problem solving agents

Problem-Solving Agents In Artificial Intelligence

In artificial intelligence, a problem-solving agent refers to a type of intelligent agent designed to address and solve complex problems or tasks in its environment. These agents are a fundamental concept in AI and are used in various applications, from game-playing algorithms to robotics and decision-making systems. Here are some key characteristics and components of a problem-solving agent:

• Perception : Problem-solving agents typically have the ability to perceive or sense their environment. They can gather information about the current state of the world, often through sensors, cameras, or other data sources.
• Knowledge Base : These agents often possess some form of knowledge or representation of the problem domain. This knowledge can be encoded in various ways, such as rules, facts, or models, depending on the specific problem.
• Reasoning : Problem-solving agents employ reasoning mechanisms to make decisions and select actions based on their perception and knowledge. This involves processing information, making inferences, and selecting the best course of action.
• Planning : For many complex problems, problem-solving agents engage in planning. They consider different sequences of actions to achieve their goals and decide on the most suitable action plan.
• Actuation : After determining the best course of action, problem-solving agents take actions to interact with their environment. This can involve physical actions in the case of robotics or making decisions in more abstract problem-solving domains.
• Feedback : Problem-solving agents often receive feedback from their environment, which they use to adjust their actions and refine their problem-solving strategies. This feedback loop helps them adapt to changing conditions and improve their performance.
• Learning : Some problem-solving agents incorporate machine learning techniques to improve their performance over time. They can learn from experience, adapt their strategies, and become more efficient at solving similar problems in the future.

Problem-solving agents can vary greatly in complexity, from simple algorithms that solve straightforward puzzles to highly sophisticated AI systems that tackle complex, real-world problems. The design and implementation of problem-solving agents depend on the specific problem domain and the goals of the AI application.

Hello, I’m Hridhya Manoj. I’m passionate about technology and its ever-evolving landscape. With a deep love for writing and a curious mind, I enjoy translating complex concepts into understandable, engaging content. Let’s explore the world of tech together

Which Of The Following Is A Privilege In SQL Standard

Implicit Return Type Int In C

Leave a Comment Cancel reply

Save my name, email, and website in this browser for the next time I comment.

Reach Out to Us for Any Query

SkillVertex is an edtech organization that aims to provide upskilling and training to students as well as working professionals by delivering a diverse range of programs in accordance with their needs and future aspirations.

© 2024 Skill Vertex

Understanding Problem Solving Agents in Artificial Intelligence

Have you ever wondered how artificial intelligence systems are able to solve complex problems? Problem solving agents play a key role in AI, using algorithms and strategies to find solutions to a variety of challenges.

Problem-solving agents in artificial intelligence are a type of agent that are designed to solve complex problems in their environment. They are a core concept in AI and are used in everything from games like chess to self-driving cars.

In this blog, we will explore problem solving agents in artificial intelligence, types of problem solving agents in AI, real-world applications, and many more.

Table of Contents

What is problem solving agents in artificial intelligence, type 1: simple reflex agents, type 2: model-based agents, type 3: goal-based agents, 2. knowledge base, 3. reasoning engine, 4. actuators, gaming agents, virtual assistants, recommendation systems, scheduling and planning.

A Problem-Solving Agent is a special computer program in Artificial Intelligence. It can perceive the world around it through sensors. Sensors help it gather information.

The agent processes this information using its knowledge base. A knowledge base is like the agent’s brain. It stores facts and rules. Using its knowledge, the agent can reason about the best actions. It can then take those actions to achieve goals.

In simple words, a Problem-Solving Agent observes its environment. It understands the situation. Then it figures out how to solve problems or finish tasks.

These agents use smart algorithms. The algorithms allow them to think and act like humans. Problem-solving agents are very important in AI. They help tackle complex challenges efficiently.

Types of Problem Solving Agents in AI

There are different types of Problem Solving Agents in AI. Each type works in its own way. Below are the different types of problem solving agents in AI:

Simple Reflex Agents are the most basic kind. They simply react to the current situation they perceive. They don’t consider the past or future.

For example, a room thermostat is a Simple Reflex Agent. It turns the heat on or off based only on the current room temperature.

Model-based agents are more advanced. They create an internal model of their environment. This model helps them track how the world changes over time.

Using this model, they can plan ahead for future situations. Self-driving cars use Model-Based Agents to predict how traffic will flow.

Goal-based agents are the most sophisticated type. They can set their own goals and figure out sequences of actions to achieve those goals.

These agents constantly update their knowledge as they pursue their goals. Virtual assistants like Siri or Alexa are examples of Goal-Based Agents assisting us with various tasks.

Each type has its own strengths based on the problem they need to solve. Simple problems may just need Reflex Agents, while complex challenges require more advanced Model-Based or Goal-Based Agents.

Components of a Problem Solving Agent in AI

A Problem Solving Agent has several key components that work together. Let’s break them down:

Sensors are like the agent’s eyes and ears. They collect information from the environment around the agent. For example, a robot’s camera and motion sensors act as sensors.

The Knowledge Base stores all the facts, rules, and information the agent knows. It’s like the agent’s brain full of knowledge. This knowledge helps the agent understand its environment and make decisions.

The Reasoning Engine is the thinking part of the agent. It processes the information from sensors using the knowledge base. The reasoning engine then figures out the best action to take based on the current situation.

Finally, Actuators are like the agent’s hands and limbs. They carry out the actions decided by the reasoning engine. For a robot, wheels and robotic arms would be its actuators.

All these components work seamlessly together. Sensors gather data, the knowledge base provides context, the reasoning engine makes a plan, and actuators implement that plan in the real world.

Real-world Applications of Problem Solving Agents in AI

Problem Solving Agents are not just theoretical concepts. They are actively used in many real-world applications today. Let’s look at some examples:

Problem solving agents are widely used in gaming applications. They can analyze the current game state, consider possible future moves, and make the optimal play. This allows them to beat human players in complex games like chess or go.

Robots in factories and warehouses heavily rely on problem solving agents. These agents perceive the environment around the robot using sensors. They then plan efficient paths and control the robot’s movements and actions accordingly.

Smart home devices like Alexa or Google Home use goal-based problem solving agents. They can understand your requests, look up relevant information from their knowledge base, and provide useful responses to assist you.

Online retailers suggest products you may like based on recommendations from problem solving agents. These agents analyze your past purchases and preferences to make personalized product suggestions.

Scheduling apps help plan your day efficiently using problem solving techniques. The agents consider your appointments, priorities, and travel time to optimize your daily schedule.

Self-Driving Cars One of the most advanced applications is self-driving cars. Their problem solving agents continuously monitor surroundings, predict the movements of other vehicles and objects, and navigate roads safely without human intervention.

In conclusion, Problem solving agents are at the heart of artificial intelligence, mimicking human-like reasoning and decision-making. From gaming to robotics, virtual assistants to self-driving cars, these intelligent agents are already transforming our world. As researchers continue pushing the boundaries, problem solving agents will become even more advanced and ubiquitous in the future. Exciting times lie ahead as we unlock the full potential of this remarkable technology.

Ajay Rathod loves talking about artificial intelligence (AI). He thinks AI is super cool and wants everyone to understand it better. Ajay has been working with computers for a long time and knows a lot about AI. He wants to share his knowledge with you so you can learn too!

4 thoughts on “Understanding Problem Solving Agents in Artificial Intelligence”

• Pingback: What Is Explanation Based Learning in Artificial Intelligence?

Can you be more specific about the content of your article? After reading it, I still have some doubts. Hope you can help me.

Thanks for sharing. I read many of your blog posts, cool, your blog is very good.

Leave a comment Cancel reply

Save my name, email, and website in this browser for the next time I comment.

Problem Solving Agents in Artificial Intelligence

In this post, we will talk about Problem Solving agents in Artificial Intelligence, which are sort of goal-based agents. Because the straight mapping from states to actions of a basic reflex agent is too vast to retain for a complex environment, we utilize goal-based agents that may consider future actions and the desirability of outcomes.

You Will Learn

Problem Solving Agents

Problem Solving Agents decide what to do by finding a sequence of actions that leads to a desirable state or solution.

An agent may need to plan when the best course of action is not immediately visible. They may need to think through a series of moves that will lead them to their goal state. Such an agent is known as a problem solving agent , and the computation it does is known as a search .

The problem solving agent follows this four phase problem solving process:

• Goal Formulation: This is the first and most basic phase in problem solving. It arranges specific steps to establish a target/goal that demands some activity to reach it. AI agents are now used to formulate goals.
• Problem Formulation: It is one of the fundamental steps in problem-solving that determines what action should be taken to reach the goal.
• Search: After the Goal and Problem Formulation, the agent simulates sequences of actions and has to look for a sequence of actions that reaches the goal. This process is called search, and the sequence is called a solution . The agent might have to simulate multiple sequences that do not reach the goal, but eventually, it will find a solution, or it will find that no solution is possible. A search algorithm takes a problem as input and outputs a sequence of actions.
• Execution: After the search phase, the agent can now execute the actions that are recommended by the search algorithm, one at a time. This final stage is known as the execution phase.

Problems and Solution

Before we move into the problem formulation phase, we must first define a problem in terms of problem solving agents.

A formal definition of a problem consists of five components:

Initial State

Transition model.

It is the agent’s starting state or initial step towards its goal. For example, if a taxi agent needs to travel to a location(B), but the taxi is already at location(A), the problem’s initial state would be the location (A).

It is a description of the possible actions that the agent can take. Given a state s, Actions ( s ) returns the actions that can be executed in s. Each of these actions is said to be appropriate in s.

It describes what each action does. It is specified by a function Result ( s, a ) that returns the state that results from doing action an in state s.

The initial state, actions, and transition model together define the state space of a problem, a set of all states reachable from the initial state by any sequence of actions. The state space forms a graph in which the nodes are states, and the links between the nodes are actions.

It determines if the given state is a goal state. Sometimes there is an explicit list of potential goal states, and the test merely verifies whether the provided state is one of them. The goal is sometimes expressed via an abstract attribute rather than an explicitly enumerated set of conditions.

It assigns a numerical cost to each path that leads to the goal. The problem solving agents choose a cost function that matches its performance measure. Remember that the optimal solution has the lowest path cost of all the solutions .

Example Problems

The problem solving approach has been used in a wide range of work contexts. There are two kinds of problem approaches

• Standardized/ Toy Problem: Its purpose is to demonstrate or practice various problem solving techniques. It can be described concisely and precisely, making it appropriate as a benchmark for academics to compare the performance of algorithms.
• Real-world Problems: It is real-world problems that need solutions. It does not rely on descriptions, unlike a toy problem, yet we can have a basic description of the issue.

Some Standardized/Toy Problems

Vacuum world problem.

Let us take a vacuum cleaner agent and it can move left or right and its jump is to suck up the dirt from the floor.

The vacuum world’s problem can be stated as follows:

States: A world state specifies which objects are housed in which cells. The objects in the vacuum world are the agent and any dirt. The agent can be in either of the two cells in the simple two-cell version, and each call can include dirt or not, therefore there are 2×2×2 = 8 states. A vacuum environment with n cells has n×2 n states in general.

Initial State: Any state can be specified as the starting point.

Actions: We defined three actions in the two-cell world: sucking, moving left, and moving right. More movement activities are required in a two-dimensional multi-cell world.

Transition Model: Suck cleans the agent’s cell of any filth; Forward moves the agent one cell forward in the direction it is facing unless it meets a wall, in which case the action has no effect. Backward moves the agent in the opposite direction, whilst TurnRight and TurnLeft rotate it by 90°.

Goal States: The states in which every cell is clean.

Action Cost: Each action costs 1.

8 Puzzle Problem

In a sliding-tile puzzle , a number of tiles (sometimes called blocks or pieces) are arranged in a grid with one or more blank spaces so that some of the tiles can slide into the blank space. One variant is the Rush Hour puzzle, in which cars and trucks slide around a 6 x 6 grid in an attempt to free a car from the traffic jam. Perhaps the best-known variant is the 8- puzzle (see Figure below ), which consists of a 3 x 3 grid with eight numbered tiles and one blank space, and the 15-puzzle on a 4 x 4  grid. The object is to reach a specified goal state, such as the one shown on the right of the figure. The standard formulation of the 8 puzzles is as follows:

STATES : A state description specifies the location of each of the tiles.

INITIAL STATE : Any state can be designated as the initial state. (Note that a parity property partitions the state space—any given goal can be reached from exactly half of the possible initial states.)

ACTIONS : While in the physical world it is a tile that slides, the simplest way of describing action is to think of the blank space moving Left , Right , Up , or Down . If the blank is at an edge or corner then not all actions will be applicable.

TRANSITION MODEL : Maps a state and action to a resulting state; for example, if we apply Left to the start state in the Figure below, the resulting state has the 5 and the blank switched.

GOAL STATE :  It identifies whether we have reached the correct goal state. Although any state could be the goal, we typically specify a state with the numbers in order, as in the Figure above.

ACTION COST : Each action costs 1.

You Might Like:

• Agents in Artificial Intelligence

Types of Environments in Artificial Intelligence

• Understanding PEAS in Artificial Intelligence
• River Crossing Puzzle | Farmer, Wolf, Goat and Cabbage

Share Article:

Digital image processing: all you need to know.

• Part 2 Problem-solving »
• Chapter 3 Solving Problems by Searching
• Edit on GitHub

Chapter 3 Solving Problems by Searching 

When the correct action to take is not immediately obvious, an agent may need to plan ahead : to consider a sequence of actions that form a path to a goal state. Such an agent is called a problem-solving agent , and the computational process it undertakes is called search .

Problem-solving agents use atomic representations, that is, states of the world are considered as wholes, with no internal structure visible to the problem-solving algorithms. Agents that use factored or structured representations of states are called planning agents .

We distinguish between informed algorithms, in which the agent can estimate how far it is from the goal, and uninformed algorithms, where no such estimate is available.

3.1 Problem-Solving Agents 

If the agent has no additional information—that is, if the environment is unknown —then the agent can do no better than to execute one of the actions at random. For now, we assume that our agents always have access to information about the world. With that information, the agent can follow this four-phase problem-solving process:

GOAL FORMULATION : Goals organize behavior by limiting the objectives and hence the actions to be considered.

PROBLEM FORMULATION : The agent devises a description of the states and actions necessary to reach the goal—an abstract model of the relevant part of the world.

SEARCH : Before taking any action in the real world, the agent simulates sequences of actions in its model, searching until it finds a sequence of actions that reaches the goal. Such a sequence is called a solution .

EXECUTION : The agent can now execute the actions in the solution, one at a time.

It is an important property that in a fully observable, deterministic, known environment, the solution to any problem is a fixed sequence of actions . The open-loop system means that ignoring the percepts breaks the loop between agent and environment. If there is a chance that the model is incorrect, or the environment is nondeterministic, then the agent would be safer using a closed-loop approach that monitors the percepts.

In partially observable or nondeterministic environments, a solution would be a branching strategy that recommends different future actions depending on what percepts arrive.

3.1.1 Search problems and solutions 

A search problem can be defined formally as follows:

A set of possible states that the environment can be in. We call this the state space .

The initial state that the agent starts in.

A set of one or more goal states . We can account for all three of these possibilities by specifying an \(Is\-Goal\) method for a problem.

The actions available to the agent. Given a state \(s\) , \(Actions(s)\) returns a finite set of actions that can be executed in \(s\) . We say that each of these actions is applicable in \(s\) .

A transition model , which describes what each action does. \(Result(s,a)\) returns the state that results from doing action \(a\) in state \(s\) .

An action cost function , denote by \(Action\-Cost(s,a,s\pr)\) when we are programming or \(c(s,a,s\pr)\) when we are doing math, that gives the numeric cost of applying action \(a\) in state \(s\) to reach state \(s\pr\) .

A sequence of actions forms a path , and a solution is a path from the initial state to a goal state. We assume that action costs are additive; that is, the total cost of a path is the sum of the individual action costs. An optimal solution has the lowest path cost among all solutions.

The state space can be represented as a graph in which the vertices are states and the directed edges between them are actions.

3.1.2 Formulating problems 

The process of removing detail from a representation is called abstraction . The abstraction is valid if we can elaborate any abstract solution into a solution in the more detailed world. The abstraction is useful if carrying out each of the actions in the solution is easier than the original problem.

3.2 Example Problems 

A standardized problem is intended to illustrate or exercise various problem-solving methods. It can be given a concise, exact description and hence is suitable as a benchmark for researchers to compare the performance of algorithms. A real-world problem , such as robot navigation, is one whose solutions people actually use, and whose formulation is idiosyncratic, not standardized, because, for example, each robot has different sensors that produce different data.

3.2.1 Standardized problems 

A grid world problem is a two-dimensional rectangular array of square cells in which agents can move from cell to cell.

Vacuum world

Sokoban puzzle

Sliding-tile puzzle

3.2.2 Real-world problems 

Route-finding problem

Touring problems

Trveling salesperson problem (TSP)

VLSI layout problem

Robot navigation

Automatic assembly sequencing

3.3 Search Algorithms 

A search algorithm takes a search problem as input and returns a solution, or an indication of failure. We consider algorithms that superimpose a search tree over the state-space graph, forming various paths from the initial state, trying to find a path that reaches a goal state. Each node in the search tree corresponds to a state in the state space and the edges in the search tree correspond to actions. The root of the tree corresponds to the initial state of the problem.

The state space describes the (possibly infinite) set of states in the world, and the actions that allow transitions from one state to another. The search tree describes paths between these states, reaching towards the goal. The search tree may have multiple paths to (and thus multiple nodes for) any given state, but each node in the tree has a unique path back to the root (as in all trees).

The frontier separates two regions of the state-space graph: an interior region where every state has been expanded, and an exterior region of states that have not yet been reached.

3.3.1 Best-first search 

In best-first search we choose a node, \(n\) , with minimum value of some evaluation function , \(f(n)\) .

3.3.2 Search data structures 

A node in the tree is represented by a data structure with four components

\(node.State\) : the state to which the node corresponds;

\(node.Parent\) : the node in the tree that generated this node;

\(node.Action\) : the action that was applied to the parent’s state to generate this node;

\(node.Path\-Cost\) : the total cost of the path from the initial state to this node. In mathematical formulas, we use \(g(node)\) as a synonym for \(Path\-Cost\) .

Following the \(PARENT\) pointers back from a node allows us to recover the states and actions along the path to that node. Doing this from a goal node gives us the solution.

We need a data structure to store the frontier . The appropriate choice is a queue of some kind, because the operations on a frontier are:

\(Is\-Empty(frontier)\) returns true only if there are no nodes in the frontier.

\(Pop(frontier)\) removes the top node from the frontier and returns it.

\(Top(frontier)\) returns (but does not remove) the top node of the frontier.

\(Add(node, frontier)\) inserts node into its proper place in the queue.

Three kinds of queues are used in search algorithms:

A priority queue first pops the node with the minimum cost according to some evaluation function, \(f\) . It is used in best-first search.

A FIFO queue or first-in-first-out queue first pops the node that was added to the queue first; we shall see it is used in breadth-first search.

A LIFO queue or last-in-first-out queue (also known as a stack ) pops first the most recently added node; we shall see it is used in depth-first search.

3.3.3 Redundant paths 

A cycle is a special case of a redundant path .

As the saying goes, algorithms that cannot remember the past are doomed to repeat it . There are three approaches to this issue.

First, we can remember all previously reached states (as best-first search does), allowing us to detect all redundant paths, and keep only the best path to each state.

Second, we can not worry about repeating the past. We call a search algorithm a graph search if it checks for redundant paths and a tree-like search if it does not check.

Third, we can compromise and check for cycles, but not for redundant paths in general.

3.3.4 Measuring problem-solving performance 

COMPLETENESS : Is the algorithm guaranteed to find a solution when there is one, and to correctly report failure when there is not?

COST OPTIMALITY : Does it find a solution with the lowest path cost of all solutions?

TIME COMPLEXITY : How long does it take to find a solution?

SPACE COMPLEXITY : How much memory is needed to perform the search?

To be complete, a search algorithm must be systematic in the way it explores an infinite state space, making sure it can eventually reach any state that is connected to the initial state.

In theoretical computer science, the typical measure of time and space complexity is the size of the state-space graph, \(|V|+|E|\) , where \(|V|\) is the number of vertices (state nodes) of the graph and \(|E|\) is the number of edges (distinct state/action pairs). For an implicit state space, complexity can be measured in terms of \(d\) , the depth or number of actions in an optimal solution; \(m\) , the maximum number of actions in any path; and \(b\) , the branching factor or number of successors of a node that need to be considered.

3.4 Uninformed Search Strategies 

3.4.1 breadth-first search .

When all actions have the same cost, an appropriate strategy is breadth-first search , in which the root node is expanded first, then all the successors of the root node are expanded next, then their successors, and so on.

Breadth-first search always finds a solution with a minimal number of actions, because when it is generating nodes at depth \(d\) , it has already generated all the nodes at depth \(d-1\) , so if one of them were a solution, it would have been found.

All the nodes remain in memory, so both time and space complexity are \(O(b^d)\) . The memory requirements are a bigger problem for breadth-first search than the execution time . In general, exponential-complexity search problems cannot be solved by uninformed search for any but the smallest instances .

3.4.2 Dijkstra’s algorithm or uniform-cost search 

When actions have different costs, an obvious choice is to use best-first search where the evaluation function is the cost of the path from the root to the current node. This is called Dijkstra’s algorithm by the theoretical computer science community, and uniform-cost search by the AI community.

The complexity of uniform-cost search is characterized in terms of \(C^*\) , the cost of the optimal solution, and \(\epsilon\) , a lower bound on the cost of each action, with \(\epsilon>0\) . Then the algorithm’s worst-case time and space complexity is \(O(b^{1+\lfloor C^*/\epsilon\rfloor})\) , which can be much greater than \(b^d\) .

When all action costs are equal, \(b^{1+\lfloor C^*/\epsilon\rfloor}\) is just \(b^{d+1}\) , and uniform-cost search is similar to breadth-first search.

3.4.3 Depth-first search and the problem of memory 

Depth-first search always expands the deepest node in the frontier first. It could be implemented as a call to \(Best\-First\-Search\) where the evaluation function \(f\) is the negative of the depth.

For problems where a tree-like search is feasible, depth-first search has much smaller needs for memory. A depth-first tree-like search takes time proportional to the number of states, and has memory complexity of only \(O(bm)\) , where \(b\) is the branching factor and \(m\) is the maximum depth of the tree.

A variant of depth-first search called backtracking search uses even less memory.

3.4.4 Depth-limited and iterative deepening search 

To keep depth-first search from wandering down an infinite path, we can use depth-limited search , a version of depth-first search in which we supply a depth limit, \(l\) , and treat all nodes at depth \(l\) as if they had no successors. The time complexity is \(O(b^l)\) and the space complexity is \(O(bl)\)

Iterative deepening search solves the problem of picking a good value for \(l\) by trying all values: first 0, then 1, then 2, and so on—until either a solution is found, or the depth- limited search returns the failure value rather than the cutoff value.

Its memory requirements are modest: \(O(bd)\) when there is a solution, or \(O(bm)\) on finite state spaces with no solution. The time complexity is \(O(bd)\) when there is a solution, or \(O(bm)\) when there is none.

In general, iterative deepening is the preferred uninformed search method when the search state space is larger than can fit in memory and the depth of the solution is not known .

3.4.5 Bidirectional search 

An alternative approach called bidirectional search simultaneously searches forward from the initial state and backwards from the goal state(s), hoping that the two searches will meet.

3.4.6 Comparing uninformed search algorithms 

3.5 Informed (Heuristic) Search Strategies 

An informed search strategy uses domain–specific hints about the location of goals to find colutions more efficiently than an uninformed strategy. The hints come in the form of a heuristic function , denoted \(h(n)\) :

\(h(n)\) = estimated cost of the cheapest path from the state at node \(n\) to a goal state.

3.5.1 Greedy best-first search 

Greedy best-first search is a form of best-first search that expands first the node with the lowest \(h(n)\) value—the node that appears to be closest to the goal—on the grounds that this is likely to lead to a solution quickly. So the evaluation function \(f(n)=h(n)\) .

Latest Cloud, Data, DevOps Technologies

• Artificial Intelligence

What is the problem-solving agent in artificial intelligence?

Are you curious to know how machines can solve complex problems, just like humans? Enter the world of artificial intelligence and meet one of its most critical players- the Problem-Solving Agent. In this blog post, we’ll explore what a problem-solving agent is, how it works in AI systems and some exciting real-world applications that showcase its potential. So, buckle up for an insightful journey into the fascinating world of AI problem solvers!

Problem-solving in artificial intelligence can be quite complex, requiring the use of multiple algorithms and data structures. One critical player is the Problem-Solving Agent, which helps machines find solutions to problems. In this blog post, we’ll explore what a problem-solving agent is, how it works in AI systems and some exciting real-world applications that showcase its potential. So, buckle up for an insightful journey into the fascinating world of AI problem solvers!

Table of Contents

What is Problem Solving Agent?

Problem-solving in artificial intelligence is the process of finding a solution to a problem. There are many different types of problems that can be solved, and the methods used will depend on the specific problem. The most common type of problem is finding a solution to a maze or navigation puzzle.

Other types of problems include identifying patterns, predicting outcomes, and determining solutions to systems of equations. Each type of problem has its own set of techniques and tools that can be used to solve it.

There are three main steps in problem-solving in artificial intelligence:

1) understanding the problem: This step involves understanding the specifics of the problem and figuring out what needs to be done to solve it.

2) generating possible solutions: This step involves coming up with as many possible solutions as possible based on information about the problem and what you know about how computers work.

3) choosing a solution: This step involves deciding which solution is best based on what you know about the problem and your options for solving it.

Types of Problem-Solving Agents

Problem-solving agents are a type of artificial intelligence that helps automate problem-solving. They can be used to solve problems in natural language, algebra, calculus, statistics, and machine learning.

There are three types of problem-solving agents: propositional, predicate, and automata. Propositional problem-solving agents can understand simple statements like “draw a line between A and B” or “find the maximum value of x.” Predicate problem-solving agents can understand more complex statements like “find the shortest path between two points” or “find all pairs of snakes in a jar.” Automata is the simplest form of problem-solving agent and can only understand sequences of symbols like “draw a square.”

Classification of Problem-Solving Agents

Problem-solving agents can be classified as general problem solvers or domain-specific problem solvers. General problem solvers can solve a wide range of problems, while domain-specific problem solvers are better suited for solving specific types of problems.

General problem solvers include AI programs that are designed to solve general artificial intelligence (AI) problems such as learning how to navigate a 3D environment or playing games. Domain-specific problem solvers include programs that have been specifically tailored to solve certain types of problems, such as photo editing or medical diagnosis.

Both general and domain-specific problem-solving agents can be used in conjunction with other AI tools, including natural language processing (NLP) algorithms and machine learning models. By combining these tools, we can achieve more effective and efficient outcomes in our data analysis and machine learning processes.

Applications of Problem-Solving Agents

Problem-solving agents can be used in a number of different ways in artificial intelligence. They can be used to help find solutions to specific problems or tasks, or they can be used to generalize a problem and find potential solutions. In either case, the problem-solving agent is able to understand complex instructions and carry out specific tasks.

Problem-solving is an essential skill for any artificial intelligence developer. With AI becoming more prevalent in our lives, it’s important that we have a good understanding of how to approach and solve problems. In this article, we’ll discuss some common problem-solving techniques and provide you with tips on how to apply them when developing AI applications. By applying these techniques systematically, you can build robust AI solutions that work correctly and meet the needs of your users.

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

Save my name, email, and website in this browser for the next time I comment.

Related Articles

Intel AI openVino toolkit

NVIDIA GPU FOR THE GOOGLE CLOUD PLATFORM

Nvidia GPU family for Deep Learning

INTEL AI-BASED DEVELOPMENT

Search Algorithms Part 1: Problem Formulation and Searching for Solutions

Kredo.ai Engineering

In the previous series of blogs, we have seen the different structures of the agents based on the nature of the environment it is operating in. In the current series, we will discuss more on the goal-based agent, and the search algorithms that gives the ‘solution’ to the ‘problem’ in these agents.

As mentioned previously, these blogs are very similar to the book “Artificial Intelligence: A Modern Approach”. In fact, this series can be seen as a shorthand version of the book.

Types of Goal-based agents

We have seen that the reflex agents, whose actions are a direct mapping from the states of the environment, consumes a large space to store the mapping table and is inflexible. The goal-based agents consider the long-term actions and the desirability of the outcome, which is easier to train and is adaptable to the changing environment.

There are two kinds of goal-based agents: problem-solving agents and planning agents . Problem-solving agents consider each states of the world as indivisible , with no internal structure of the states visible to the problem-solving algorithms. Planning agents split up each state into variables and establishes relationship between them.

In this series, we will discuss more on problem-solving agents and the algorithms associated with them. We’ll keep the discussion on the planning agents for some other time.

In this post (and further too), as an example to explain the various algorithms, we consider the problem of traveling from one place to another (single-source single-destination path). Figure 1 gives the road-map of a part of Romania.

The problem is to travel from Arad to Bucharest in a day. For the agent, the goal will be to reach Bucharest the following day. Courses of action that doesn’t make agent to reach Bucharest on time can be rejected without further consideration, making the agent’s decision problem simplified.

Problem definition and formulation

Before we jump on to finding the algorithm for evaluating the problem and searching for the solution, we first need to define and formulate the problem.

Problem formulation involves deciding what actions and states to consider, given the goal. For example, if the agent were to consider the action to be at the level of “move the left foot by one inch” or “turn the steering wheel by 1 degree left”, there would be too many steps for the agent to leave the parking lot, let alone to Bucharest. In general, we need to abstract the state details from the representation.

A problem can be defined formally by 5 components:

• The initial state of the agent. In this case, the initial state can be described as In: Arad
• The possible actions available to the agent, corresponding to each of the state the agent resides in. For example, ACTIONS( In: Arad ) = { Go: Sibiu , Go: Timisoara , Go: Zerind }
• The transition model describing what each action does. Let us represent it by RESULT(s, a) where s is the state the action is currently in and a is the action performed by the agent. In this example, RESULT( In: Arad, Go: Zerind ) = In: Zerind.
• The goal test , determining whether the current state is a goal state. Here, the goal state is { In: Bucharest }
• The path cost function, which determine the cost of each path, which is reflecting in the performance measure. For the agent trying to reach Bucharest, time is essential, so we can set the cost function to be the distance between the places. (Here, we are ignoring the other factors that influence the traveling time). By convention, we define the cost function as c(s, a, s’) , where s is the current state and a is the action performed by the agent to reach state s’.

The initial state, the actions and the transition model together define the state space of the problem — the set of all states reachable by any sequence of actions. Figure 1 is the graphical representation of the state space of the traveling problem. A path in the state space is a sequence of states connected by a sequence of actions.

The solution to the given problem is defined as the sequence of actions from the initial state to the goal states. The quality of the solution is measured by the cost function of the path, and an optimal solution has the lowest path cost among all the solutions.

Searching for Solutions

We can form a search tree from the state space of the problem to aid us in finding the solution. The initial state forms the root node and the branches from each node are the possible actions from the current node (state) to the child nodes (next states).

The six nodes in Figure 2, which don’t have any children (at least until now) are leaf nodes . The set of all leaf nodes available for expansion at any given point is called the frontier . The search strategy involves the expansion of the nodes in the frontier until the solution (or the goal state) is found (or there are no more nodes to expand).

We have to notice one peculiar thing in the search tree in Figure 2. There is a path from Arad to Sibiu, and back to Arad again. We say that In(Arad) is a repeated state, generated by a loopy path . This means that the search tree for Romania is infinite , even though the search space is limited. These loopy paths makes some of the algorithms to fail, making the problem seem unsolvable. In fact, a loopy path is a special case of redundant paths , where there are more than one paths from one state to another (for example, Arad — Sibiu and Arad — Zerind — Oradea — Sibiu).

The redundant path situation occurs in almost every problem, and often makes the solution algorithm less efficient, worsening the performance of the searching agent. One way to eliminate the redundancy is to utilize the advantage given by the problem definition itself. For example, in the case of traveling from Arad to Bucharest, since the path costs are additive and step costs are non-negative, only one path among the various redundant paths has the least cost (and it is the shortest distance between the two states), and loopy paths are never better than the same path with loops removed.

Another idea to avoid exploring redundant paths is to remember which states have been visited previously. Along with the search tree, an explored set is maintained which contains all the states previously visited. Newly generates which matches the previously generated nodes can be discarded. In this way, every step moves the states in the frontier into the explored region, and some states in the unexplored region into the frontier, until the solution is found.

Performance measure of Problem-solving Algorithms

We can evaluate an algorithm’s performance with these metrics:

• Completeness : Is the algorithm guaranteed to find a solution if there exist one?
• Optimality : Does the algorithm find the optimal solution?
• Time complexity : How long does it take for the algorithm to find a solution?
• Space complexity : How much memory is consumed in finding the solution?

In graph theory, the time and space complexity is measured using |V| and |E|, where V and E are the number of vertices and the number of edges in the graph respectively. But in AI, we explore the state space (which is a graph) of the problem using its equivalent search tree. So it is meaningful if we use b and d to measure the complexity, where b is the branching factor of the tree (maximum number of successors of any node) and d is the depth of the shallowest goal node.

In this post we have discussed how to define the problem so as to assist in formulation of the problem and to effectively find a solution. We have also seen the use of search tree in finding the solution and the ways to avoid the problem of redundancy. Finally, we have listed the metrics for measuring the performance of the search algorithms.

In the next blog, we will discuss the classical search algorithms,starting with uninformed search algorithms and then moving on to heuristic, or informed search algorithms.

Stuart Russel and Peter Norvig, Artificial Intelligence: A Modern Approach, 4th edition

Written by Rithesh K

An AI Enthusiast trying to explore the world with the love for Mathematics.

Text to speech

Member-only story

A Deep Exploration of Search-based Agents

Search-based agents are great at sniffing out optimal solutions to complex problems..

Jason M. Pittman

Better Programming

So far, we’ve covered Rule-based and Utility-based agents. Both are fundamental ways to address narrow problems using AI. This is in contrast to Synthetic Intelligence and Artificial Life. The latter emulates biological systems while I’ve argued the former, in simple terms, is the logical combination of AI and Alife. Eventually, we will swing back around to Alife and Synthetic Intelligence. For now, I want to move onto the third type of AI agent. Let’s dig into Search-based agents.

Like we experienced with Rule-based and Utility-based agents, there are different types of search-based agents. We can categorize designs and implementations based on the nature of the search algorithms each employ. Each also matches more closely with various problem domains.

Here are four common types of Search-based Agents:

• Uninformed Search Agents : Uninformed search agents use basic search algorithms, such as breadth-first search (BFS) or depth-first search (DFS), to explore the problem space systematically without considering any additional information. These agents have no prior knowledge about the problem domain and rely solely on the search algorithm to explore possible…

Written by Jason M. Pittman

I am a forward-leaning innovator committed to solving tomorrow’s grand challenges by developing cutting-edge research and technology today.

Text to speech

• AI Education in India
• Speakers & Mentors
• AI services

Examples of Problem Solving Agents in Artificial Intelligence

In the field of artificial intelligence, problem-solving agents play a vital role in finding solutions to complex tasks and challenges. These agents are designed to mimic human intelligence and utilize a range of algorithms and techniques to tackle various problems. By analyzing data, making predictions, and finding optimal solutions, problem-solving agents demonstrate the power and potential of artificial intelligence.

One example of a problem-solving agent in artificial intelligence is a chess-playing program. These agents are capable of evaluating millions of possible moves and predicting the best one to make based on a wide array of factors. By utilizing advanced algorithms and machine learning techniques, these agents can analyze the current state of the game, anticipate future moves, and make strategic decisions to outplay even the most skilled human opponents.

Another example of problem-solving agents in artificial intelligence is autonomous driving systems. These agents are designed to navigate complex road networks, make split-second decisions, and ensure the safety of both passengers and pedestrians. By continuously analyzing sensor data, identifying obstacles, and calculating optimal paths, these agents can effectively solve problems related to navigation, traffic congestion, and collision avoidance.

Definition and Importance of Problem Solving Agents

A problem solving agent is a type of artificial intelligence agent that is designed to identify and solve problems. These agents are programmed to analyze information, develop potential solutions, and select the best course of action to solve a given problem.

Problem solving agents are an essential aspect of artificial intelligence, as they have the ability to tackle complex problems that humans may find difficult or time-consuming to solve. These agents can handle large amounts of data and perform calculations and analysis at a much faster rate than humans.

Problem solving agents can be found in various domains, including healthcare, finance, manufacturing, and transportation. For example, in healthcare, problem solving agents can analyze patient data and medical records to diagnose diseases and recommend treatment plans. In finance, these agents can analyze market trends and make investment decisions.

The importance of problem solving agents in artificial intelligence lies in their ability to automate and streamline processes, improve efficiency, and reduce human error. These agents can also handle repetitive tasks, freeing up human resources for more complex and strategic work.

In addition, problem solving agents can learn and adapt from past experiences, making them even more effective over time. They can continuously analyze and optimize their problem-solving strategies, resulting in better decision-making and outcomes.

In conclusion, problem solving agents are a fundamental component of artificial intelligence. Their ability to analyze information, develop solutions, and make decisions has a significant impact on various industries and fields. Through their automation and optimization capabilities, problem solving agents contribute to improving efficiency, reducing errors, and enhancing decision-making processes.

Problem Solving Agent Architecture

A problem-solving agent is a central component in the field of artificial intelligence that is designed to tackle complex problems and find solutions. The architecture of a problem-solving agent consists of several key components that work together to achieve intelligent problem-solving.

One of the main components of a problem-solving agent is the knowledge base. This is where the agent stores relevant information and data that it can use to solve problems. The knowledge base can include facts, rules, and heuristics that the agent has acquired through learning or from experts in the domain.

Another important component of a problem-solving agent is the inference engine. This is the part of the agent that is responsible for reasoning and making logical deductions. The inference engine uses the knowledge base to generate possible solutions to a problem by applying various reasoning techniques, such as deduction, induction, and abduction.

Furthermore, a problem-solving agent often includes a search algorithm or strategy. This is used to systematically explore possible solutions and search for the best one. The search algorithm can be guided by various heuristics or constraints to efficiently navigate through the solution space.

In addition to these components, a problem-solving agent may also have a learning component. This allows the agent to improve its problem-solving capabilities over time through experience. The learning component can help the agent adapt its knowledge base, refine its inference engine, or adjust its search strategy based on feedback or new information.

Overall, the architecture of a problem-solving agent is designed to enable intelligent problem-solving by combining knowledge representation, reasoning, search, and learning. By utilizing these components, problem-solving agents can tackle a wide range of problems and find effective solutions in various domains.

Uninformed Search Algorithms

In the field of artificial intelligence, problem-solving agents are often designed to navigate a large search space in order to find a solution to a given problem. Uninformed search algorithms, also known as blind search algorithms, are a class of algorithms that do not use any additional information about the problem to guide their search.

Breadth-First Search (BFS)

Breadth-First Search (BFS) is one of the most basic uninformed search algorithms. It explores all the neighbor nodes at the present depth before moving on to the nodes at the next depth level. BFS is implemented using a queue data structure, where the nodes to be explored are added to the back of the queue and the nodes to be explored next are removed from the front of the queue.

For example, BFS can be used to find the shortest path between two cities on a road map, exploring all possible paths in a breadth-first manner to find the optimal solution.

Depth-First Search (DFS)

Depth-First Search (DFS) is another uninformed search algorithm that explores the deepest path first before backtracking. It is implemented using a stack data structure, where nodes are added to the top of the stack and the nodes to be explored next are removed from the top of the stack.

DFS can be used in situations where the goal state is likely to be far from the starting state, as it explores the deepest paths first. However, it may get stuck in an infinite loop if there is a cycle in the search space.

For example, DFS can be used to solve a maze, exploring different paths until the goal state (exit of the maze) is reached.

Overall, uninformed search algorithms provide a foundational approach to problem-solving in artificial intelligence. They do not rely on any additional problem-specific knowledge, making them applicable to a wide range of problems. While they may not always find the optimal solution or have high efficiency, they provide a starting point for more sophisticated search algorithms.

Breadth-First Search

Breadth-First Search is a problem-solving algorithm commonly used in artificial intelligence. It is an uninformed search algorithm that explores all the immediate variations of a problem before moving on to the next level of variations.

Examples of problems that can be solved using Breadth-First Search include finding the shortest path between two points in a graph, solving a sliding puzzle, or searching for a word in a large text document.

How Breadth-First Search Works

The Breadth-First Search algorithm starts at the initial state of the problem and expands all the immediate successor states. It then explores the successor states of the expanded states, continuing this process until a goal state is reached.

At each step of the algorithm, the breadth-first search maintains a queue of states to explore. The algorithm removes a state from the front of the queue, explores its successor states, and adds them to the back of the queue. This ensures that states are explored in the order they were added to the queue, resulting in a breadth-first exploration of the problem space.

The algorithm also keeps track of the visited states to avoid revisiting them in the future, preventing infinite loops in cases where the problem space contains cycles.

Benefits and Limitations

Breadth-First Search guarantees that the shortest path to a goal state is found, if such a path exists. It explores all possible paths of increasing lengths until a goal state is reached, ensuring that shorter paths are explored first.

However, the main limitation of Breadth-First Search is its memory requirements. As it explores all immediate successor states, it needs to keep track of a large number of states in memory. This can become impractical for problems with a large state space. Additionally, Breadth-First Search does not take into account the cost or quality of the paths it explores, making it less suitable for problems with complex cost or objective functions.

Depth-First Search

Depth-First Search (DFS) is a common algorithm used in the field of artificial intelligence to solve various types of problems. It is a search strategy that explores as far as possible along each branch of a tree-like structure before backtracking.

In the context of problem-solving agents, DFS is often used to traverse graph-based problem spaces in search of a solution. This algorithm starts at an initial state and explores all possible actions from that state until a goal state is found or all possible paths have been exhausted.

One example of using DFS in artificial intelligence is solving mazes. The agent starts at the entrance of the maze and explores one path at a time, prioritizing depth rather than breadth. It keeps track of the visited nodes and backtracks whenever it encounters a dead end, until it reaches the goal state (the exit of the maze).

Another example is solving puzzles, such as the famous Eight Queens Problem. In this problem, the agent needs to place eight queens on a chessboard in such a way that no two queens threaten each other. DFS can be used to explore all possible combinations of queen placements, backtracking whenever a placement is found to be invalid, until a valid solution is found or all possibilities have been exhausted.

DFS has advantages and disadvantages. Its main advantage is its simplicity and low memory usage, as it only needs to store the path from the initial state to the current state. However, it can get stuck in infinite loops if not implemented properly, and it may not always find the optimal solution.

In conclusion, DFS is a useful algorithm for problem-solving agents in artificial intelligence. It can be applied to a wide range of problems and provides a straightforward approach to exploring problem spaces. By understanding its strengths and limitations, developers can effectively utilize DFS to find solutions efficiently.

Iterative Deepening Depth-First Search

Iterative Deepening Depth-First Search (IDDFS) is a popular search algorithm used in problem solving within the field of artificial intelligence. It is a combination of depth-first search and breadth-first search algorithms and is designed to overcome some of the limitations of traditional depth-first search.

IDDFS operates in a similar way to depth-first search by exploring a problem space depth-wise. However, it does not keep track of the visited nodes in the search tree as depth-first search does. Instead, it uses a depth limit, which is gradually increased with each iteration, to restrict the depth to which it explores the search tree. This allows IDDFS to gradually explore the search space, starting from a shallow depth and progressively moving to deeper depths.

The iterative deepening depth-first search algorithm works by repeatedly performing depth-limited searches, incrementing the depth limit by one with each iteration. It performs a depth-first search to a given depth limit and if the goal state is not found, it increases the depth limit and performs the search again. This iterative process continues until the goal state is found or the entire search space has been explored.

IDDFS combines the advantages of both depth-first search and breadth-first search. It has the completeness of breadth-first search, meaning it is guaranteed to find a solution if one exists in the search space. At the same time, it preserves the memory efficiency of depth-first search by only keeping track of the current path being explored. This makes it an efficient algorithm for solving problems that have large or infinite search spaces.

Advantages of Iterative Deepening Depth-First Search

1. Completeness: IDDFS is a complete algorithm, meaning it is guaranteed to find a solution if one exists.

2. Memory efficiency: IDDFS only keeps track of the current path being explored, making it memory-efficient compared to breadth-first search which needs to store the entire search tree in memory.

Disadvantages of Iterative Deepening Depth-First Search

1. Redundant work: IDDFS performs multiple depth-limited searches, which can result in redundant work as nodes may be explored multiple times at different depths.

2. Inefficient for non-uniform branching factors: If the branching factor of the search tree varies greatly across different levels, IDDFS may spend a significant amount of time exploring deep levels with high branching factors, leading to inefficiency.

In conclusion, iterative deepening depth-first search is a powerful algorithm used in problem solving within artificial intelligence. It combines the efficiency of depth-first search with the completeness of breadth-first search, making it a valuable tool for solving problems that involve large or infinite search spaces.

Informed Search Algorithms

In artificial intelligence, problem-solving agents are designed to find solutions to complex problems by applying search algorithms. One class of search algorithms is known as informed search algorithms, which make use of additional knowledge or heuristics to guide the search process.

These algorithms are particularly useful when the problem space is large and the search process needs to be optimized. By using heuristics, informed search algorithms can prioritize certain paths or nodes that are more likely to lead to a solution.

Examples of Informed Search Algorithms

• A* algorithm: This is a widely used informed search algorithm that combines the benefits of both breadth-first search and best-first search approaches. It uses a heuristic function to estimate the cost from a given node to the goal state, and selects the path with the lowest estimated cost.
• Greedy Best-First Search: This algorithm uses a heuristic function to prioritize nodes based on their estimated distance to the goal. It always chooses the path that appears to be closest to the goal, without considering the overall cost of the path.
• IDA* algorithm: Short for Iterative Deepening A*, this algorithm is an optimization of the A* algorithm. It performs a depth-first search with an increasing maximum depth limit, guided by a heuristic function. This allows it to find the optimal solution with less memory usage.

These are just a few examples of the many informed search algorithms that exist in the field of artificial intelligence. Each algorithm has its own advantages and is suitable for different types of problems. By applying these algorithms, problem-solving agents can efficiently navigate through complex problem spaces and find optimal solutions.

Uniform-Cost Search

In the field of artificial intelligence, problem-solving agents are designed to find optimal solutions to given problems. One common approach is the use of search algorithms to explore the problem space and find the best path from an initial state to a goal state. Uniform-cost search is one such algorithm that is widely used in various problem-solving scenarios.

Uniform-cost search works by maintaining a priority queue of states, with the cost of reaching each state as the priority. The algorithm starts with an initial state and repeatedly selects the state with the lowest cost from the queue for expansion. It then generates all possible successors of the selected state and adds them to the queue with their respective costs. This process continues until the goal state is reached or the queue is empty.

To illustrate the use of uniform-cost search, let’s consider an example of finding the shortest path from one city to another on a map. The map can be represented as a graph, with cities as the nodes and roads as the edges. Each road has a cost associated with it, representing the distance between the two cities it connects.

Using uniform-cost search, the algorithm would start from the initial city and explore the neighboring cities, considering the cost of each road. It would then continue expanding the cities with the lowest cumulative costs, gradually moving towards the goal city. The algorithm terminates when it reaches the goal city or exhausts all possible paths.

Uniform-cost search is particularly useful in scenarios where the goal is to find the optimal solution with the lowest cost. It guarantees the discovery of the optimal path by exploring all possible paths in a systematic way. However, it can be computationally expensive in terms of time and memory requirements, especially in large problem spaces.

In conclusion, uniform-cost search is an effective algorithm used by problem-solving agents in artificial intelligence to find optimal solutions. It systematically explores all possible paths, guaranteeing the discovery of the optimal solution. However, it can be computationally expensive and requires significant memory usage, making it less suitable for problems with large or infinite state spaces.

Greedy Best-First Search

Greedy Best-First Search (GBFS) is a problem-solving algorithm used in artificial intelligence. It is an example of an intelligent agent that aims to find the most promising solution based solely on its heuristic function.

The GBFS algorithm starts by initializing the initial state of the problem. Then, it evaluates all the neighboring states using a heuristic function, which estimates the cost or value of each state based on certain criteria. The algorithm selects the state that has the lowest heuristic value as the next state to explore.

This means that GBFS always chooses the path that seems most promising at the current moment, without considering the global picture or evaluating future consequences. It follows a greedy approach by making locally optimal decisions. This can sometimes lead to suboptimal solutions if the initial path chosen ends up being a dead-end or if there is a better path further down the line.

GBFS can be used in various problem-solving scenarios. One example is the traveling salesman problem, where the goal is to find the shortest possible route that visits a set of cities and returns to the starting point. The algorithm can evaluate the heuristic value of each potential next city based on its proximity to the current city and select the city with the shortest distance as the next destination.

Another example is the maze-solving problem, where GBFS can be used to navigate through a maze by evaluating the heuristic value of each possible move, such as the distance to the exit or the number of obstacles in the path. The algorithm then chooses the move that leads to the most promising outcome based on the heuristic evaluation.

Overall, GBFS is an example of an intelligent agent in artificial intelligence that utilizes a heuristic function to make locally optimal decisions in problem-solving scenarios. While it may not always guarantee the optimal solution, it can often provide a good approximation and is efficient in many practical applications.

A* search is a widely used algorithm in artificial intelligence for problem-solving. It is an informed search algorithm that combines the features of uniform-cost search with heuristic functions to find an optimal path from a start state to a goal state.

The A* search algorithm is especially useful when dealing with problems that have a large search space or multiple possible paths to the goal state. It uses a heuristic function to estimate the cost of reaching the goal from each state and adds this estimated cost to the actual cost of getting to that state so far. The algorithm then explores the states with the lowest total cost first, making it a best-first search algorithm.

How A* Search Works

At each step of the A* search algorithm, it selects the state with the lowest total cost from the open set of states to explore next. The total cost is calculated as the sum of the actual cost of reaching the state plus the estimated cost of reaching the goal from that state. The open set is initially populated with the start state, and the algorithm continues until the goal state is reached or the open set is empty.

To estimate the cost of reaching the goal, A* search uses a heuristic function, often denoted as h(n), which provides an optimistic estimate of the cost from a given state to the goal. This heuristic function is problem-specific and can be defined based on various factors, such as distance, time, or other relevant considerations.

One commonly used heuristic function is the Manhattan distance, which calculates the distance between two points in a grid-like environment by summing the absolute differences of their x and y coordinates. Another example is the Euclidean distance, which calculates the straight-line distance between two points in a continuous space.

Examples of A* Search

A* search has been successfully applied to various problem-solving scenarios. Some examples include:

• Pathfinding in a grid-based environment, such as finding the shortest path in a maze or a game level.
• Optimal route planning for vehicles or delivery services, considering factors like traffic conditions or fuel consumption.
• Puzzle solving, such as finding the minimum number of moves to solve a sliding puzzle or the Tower of Hanoi problem.
• Scheduling and resource allocation, where the objective is to minimize costs or maximize efficiency.

These examples demonstrate the versatility and effectiveness of A* search in solving a wide range of problems in artificial intelligence.

Constraint Satisfaction Problems

In the field of artificial intelligence, constraint satisfaction problems (CSPs) are a type of problem-solving agent that deals with a set of variables and a set of constraints that define the relationships between those variables. The aim is to find an assignment of values to the variables that satisfies all the given constraints.

One example of a CSP is the Sudoku puzzle. In this puzzle, the variables are the empty cells, and the constraints are that each row, column, and 3×3 subgrid must contain distinct numbers from 1 to 9. The problem-solving agent must find a valid assignment of numbers to the variables in order to solve the puzzle.

Another example of a CSP is the map coloring problem. In this problem, the variables are the regions on a map, and the constraints are that adjacent regions cannot have the same color. The problem-solving agent must assign a color to each region in such a way that no adjacent regions have the same color.

CSPs can be solved using various algorithms, such as backtracking, constraint propagation, and local search. These algorithms iteratively explore the search space of possible variable assignments, while taking into account the constraints, in order to find a valid solution.

Overall, constraint satisfaction problems provide a framework for modeling and solving a wide range of problems in artificial intelligence, from puzzles to planning and scheduling problems. By representing the problem as a set of variables and constraints, problem-solving agents can efficiently search for solutions that satisfy all the given constraints.

Backtracking

Backtracking is a common technique used in solving problems in artificial intelligence. It is particularly useful when exploring all possible solutions to a problem. Backtracking involves a systematic approach to finding a solution by incrementally building a potential solution, and when a dead-end is encountered, it backtracks and tries a different path.

One example of backtracking is the n-queens problem . In this problem, the goal is to place n queens on an n x n chessboard such that no two queens can attack each other. Backtracking can be used to find all possible solutions to this problem by systematically placing queens on the board and checking if the current configuration is valid. If a configuration is not valid, the algorithm backtracks and tries a different position.

Another example of backtracking is the knight’s tour problem . In this problem, the goal is to find a sequence of moves for a knight on a chessboard such that it visits every square exactly once. Backtracking can be used to explore all possible paths the knight can take, and when a dead-end is encountered, it backtracks and tries a different path.

Backtracking algorithms can be time-consuming as they may need to explore a large number of potential solutions. However, they are powerful and flexible, making them suitable for solving a wide range of problems. In artificial intelligence, backtracking is often used in problem-solving agents to find optimal solutions or to explore the space of possible solutions.

Forward Checking

Forward Checking is a technique used by problem-solving agents in artificial intelligence to improve the efficiency and effectiveness of their search algorithms. It is particularly useful when dealing with constraint satisfaction problems, where there are variables that need to be assigned values while satisfying certain constraints.

How does it work?

When a variable is assigned a value, forward checking updates the remaining domains of the variables by removing any values that are inconsistent with the assigned value, based on the constraints. This helps reduce the search space and allows the agent to explore more promising paths towards a solution.

For example, let’s consider a Sudoku puzzle, which is a classic constraint satisfaction problem. The goal is to fill a 9×9 grid with digits from 1 to 9, such that each row, each column, and each of the nine 3×3 subgrids contains all of the digits from 1 to 9 without repetition.

When forward checking is applied to solve a Sudoku puzzle, the agent starts by assigning a value to an empty cell. Then, it updates the domains of the remaining variables (empty cells) by removing any values that violate the Sudoku constraints. This reduces the number of possible values for the remaining variables and improves the efficiency of the search algorithm.

Advantages of Forward Checking

Forward checking has several advantages when used by problem-solving agents:

• It helps reduce the search space by eliminating values that are inconsistent with the constraints.
• It can lead to more efficient search algorithms by guiding the agent towards more promising paths.
• It can improve the accuracy of the search algorithm by considering the constraints during the assignment of values.

Overall, forward checking is an important technique used by problem-solving agents to efficiently solve constraint satisfaction problems, such as Sudoku puzzles, and improve the effectiveness of their search algorithms.

Arc Consistency

Arc consistency is a key concept in artificial intelligence problem-solving agents, specifically in constraint satisfaction problems (CSPs). CSPs are mathematical problems that involve finding a solution that satisfies a set of constraints.

In a CSP, variables are assigned values from a domain, and constraints define the relationships between the variables. Arc consistency is a technique used to reduce the search space by ensuring that all values in the domain are consistent with the constraints.

For example, consider a scheduling problem where we need to assign tasks to workers. We have a set of constraints that specify which tasks can be assigned to which workers. Arc consistency would involve checking each constraint to ensure that the assigned values satisfy the constraints. If a constraint is not satisfied, the agent would backtrack and try a different assignment.

The arc consistency technique uses a process called domain filtering, which iteratively eliminates values from the domain that are not consistent with the current assignments and constraints. This process continues until no more values can be removed or until a solution is found.

In this example, initially both Task 1 and Task 2 can be assigned to both Worker A and Worker B. However, by applying arc consistency, we can eliminate the assignments that violate the constraints. After applying arc consistency, we end up with the following assignments:

By applying arc consistency, we have reduced the solution space and ensured that all assignments satisfy the constraints. This allows the problem-solving agent to search for a solution more efficiently.

Game Playing Agents

Game playing agents are artificial intelligence agents that are designed to play games. These agents are capable of making decisions and taking actions in order to achieve the goal of winning the game. They use various problem solving techniques and strategies to analyze the current state of the game and make the best possible move.

There are several examples of game playing agents in artificial intelligence:

Game playing agents have been a subject of research and development in artificial intelligence for many years. They have contributed to advancements in areas such as machine learning, pattern recognition, and decision-making algorithms.

Minimax Algorithm

The Minimax Algorithm is a common solving approach used by intelligent agents in the field of artificial intelligence. It is primarily used in scenarios where an agent needs to make decisions in a competitive setting with an opponent.

The goal of the Minimax Algorithm is to determine the best possible move for an agent, assuming that the opponent is also playing optimally. It works by exploring all potential moves and their resulting outcomes, ultimately selecting the move that minimizes the maximum possible outcome for the opponent.

One example of the Minimax Algorithm in action is in the game of Chess. The agent (player) evaluates the potential moves it can make and computes the possible moves the opponent (opponent player) can make in response. The agent then simulates each possible sequence of moves, looking several moves ahead, and assigns a score to each sequence based on the predicted outcome. The agent selects the move that leads to the sequence with the lowest score, assuming the opponent will always make the move that maximizes their score.

Another example is in the game of Tic Tac Toe. The agent and the opponent each take turns making moves on a 3×3 grid. The agent uses the Minimax Algorithm to explore the possible outcomes of each move and selects the move that minimizes the maximum potential outcome for the opponent.

The Minimax Algorithm is a powerful tool for solving problems in artificial intelligence, as it allows intelligent agents to make optimal decisions in competitive settings. It can be applied to a wide range of scenarios beyond games, including decision-making processes in robotics, resource allocation, and strategic planning.

Alpha-Beta Pruning

In the field of artificial intelligence, one of the key techniques used by problem-solving agents is called alpha-beta pruning. This technique is employed in game playing algorithms, where the agent needs to make decisions that maximize its chances of winning.

The goal of alpha-beta pruning is to reduce the number of nodes that need to be evaluated in a game tree, without compromising the correctness of the agent’s decision. By pruning branches of the tree that are deemed to be less promising, the agent can save significant computational resources and make faster decisions.

How Alpha-Beta Pruning Works

Alpha-beta pruning is based on the concept of minimax algorithm, which explores the entire game tree to find the optimal move for the agent. However, unlike minimax, alpha-beta pruning stops exploring certain branches when it is determined that they will not affect the final decision.

The algorithm maintains two values called alpha and beta, which represent the best values achievable for the maximizing player and the minimizing player, respectively. As the agent explores the tree, it updates these values based on the current position and the possible moves.

If the agent finds a move that yields a value greater than or equal to the beta value, it means that the minimizing player can force a value greater than or equal to beta, so there is no need to explore that branch further. Similarly, if the agent finds a move that yields a value less than or equal to the alpha value, it means that the maximizing player can force a value less than or equal to alpha, so there is no need to explore that branch further either.

Benefits of Alpha-Beta Pruning

Alpha-beta pruning is a powerful technique that can greatly improve the efficiency of problem-solving agents in artificial intelligence. By avoiding the evaluation of unnecessary nodes in the game tree, agents can make faster decisions without sacrificing accuracy.

This technique is particularly useful in games with large branching factors, where the game tree can be extremely large. Alpha-beta pruning allows agents to focus their computational resources on the most promising branches, leading to more effective decision-making and improved gameplay.

Monte Carlo Tree Search

Monte Carlo Tree Search (MCTS) is a popular algorithm used in solving complex problems by artificial intelligence agents. It is particularly effective in problem domains with large state spaces and difficult decision-making processes.

MCTS simulates the problem-solving process by traversing a tree of possible actions and outcomes. It uses random sampling, or “Monte Carlo” simulations, to estimate the potential value or utility of each action. This allows the agent to focus its search on promising actions and avoid wasting time exploring unpromising ones.

The MCTS algorithm consists of four main steps: selection, expansion, simulation, and backpropagation. In the selection step, the algorithm chooses a node from the tree based on a selection policy, typically the Upper Confidence Bound (UCB). The expansion step adds child nodes to the selected node, representing possible actions. The simulation step performs a Monte Carlo simulation by randomly selecting actions and obtaining a simulated outcome. Finally, the backpropagation step updates the values of the nodes in the tree based on the simulation results.

By iteratively performing these steps, MCTS gradually builds up knowledge about the problem domain and improves its decision-making capabilities. It can be used in a wide range of problem-solving scenarios, such as playing board games, optimizing resource allocation, or finding optimal strategies in complex environments.

Overall, Monte Carlo Tree Search is an effective algorithm for solving problems in artificial intelligence. Its ability to balance exploration and exploitation allows agents to efficiently search large state spaces and find optimal solutions to complex problems.

Expert Systems

Expert systems are a type of problem-solving agents in the field of artificial intelligence. They are designed to mimic the behavior and knowledge of human experts in a specific domain. These systems use a combination of rules, inference engines, and knowledge bases to solve complex problems and provide expert-level solutions.

Expert systems can be found in various industries and domains, including healthcare, finance, manufacturing, and customer support. They are used to assist professionals in making complex decisions, troubleshoot problems, and provide expert advice.

One example of an expert system is IBM Watson, which gained fame for its victory on the television quiz show Jeopardy! Watson is designed to understand natural language, process large amounts of data, and provide accurate answers to questions. It utilizes machine learning techniques to improve its performance over time.

Another example is Dendral, an expert system developed in the 1960s to solve problems in organic chemistry. Dendral was able to analyze mass spectrometry data and identify the structure of organic compounds. It was one of the first successful applications of expert systems in the field of chemistry.

Expert systems can be classified as rule-based systems, where a set of rules is defined to guide the decision-making process. These rules are usually created by domain experts and encoded in the knowledge base of the system. The inference engine then uses these rules to reason and make inferences.

Overall, expert systems play a crucial role in artificial intelligence by combining human expertise and machine learning techniques to solve complex problems in various domains. They provide valuable insights and solutions, making them powerful tools for professionals in different industries.

Rule-Based Systems

Rule-based systems are a common type of problem-solving agent in artificial intelligence. These systems use a set of rules or “if-then” statements to solve problems. Each rule consists of a condition and an action. If the condition is met, then the action is performed.

Example 1: Expert Systems

One example of a rule-based system is an expert system. Expert systems are designed to mimic the decision-making abilities of human experts in a specific domain. They use a knowledge base of rules to provide advice or make decisions. For example, a medical expert system could use rules to diagnose a patient’s symptoms and recommend a course of treatment.

Example 2: Production Systems

Another example of a rule-based system is a production system. Production systems are commonly used in manufacturing and planning domains. They consist of rules that describe the steps to be taken in a production process. For example, a production system for building a car could have rules for assembling different components in a specific order.

In conclusion, rule-based systems are a powerful tool in artificial intelligence for solving problems. They use a set of rules to make decisions or perform actions based on specific conditions. Examples include expert systems and production systems.

Fuzzy Logic

Fuzzy logic is a branch of artificial intelligence that deals with reasoning that is approximate rather than precise. In contrast to traditional logic, which is based on binary true/false values, fuzzy logic allows for degrees of truth. This makes it particularly useful for problem solving agents in artificial intelligence, as it enables them to work with uncertain or ambiguous information.

One of the key advantages of fuzzy logic is its ability to handle imprecise data and make decisions based on incomplete or uncertain information. This makes it well-suited for applications such as decision-making systems, control systems, and expert systems.

One example of fuzzy logic in action is in weather forecasting. Since weather conditions can be difficult to predict with complete accuracy, fuzzy logic can be used to analyze various factors such as temperature, humidity, and wind speed, and make a determination about the likelihood of rain or sunshine.

Another example is in autonomous vehicles. Fuzzy logic can be used to interpret sensor data, such as distance, speed, and road conditions, and make decisions about how to navigate and respond to the environment. This allows the vehicle to adapt and make intelligent decisions in real-time.

Bayesian Networks

Bayesian Networks are a powerful tool in the field of Artificial Intelligence, used by problem-solving agents to model uncertain knowledge and make decisions based on probability.

Bayesian Networks are graphical models that represent a set of variables and their probabilistic relationships through a directed acyclic graph. The nodes in the graph represent the variables, while the edges represent the dependencies between the variables.

These networks are widely used in various domains, including healthcare, finance, and robotics, to name a few. They are particularly useful when dealing with uncertain and complex situations, where decisions need to be made based on incomplete or imperfect information.

Examples of Bayesian Networks:

• Medical Diagnosis: Bayesian Networks can be used to model and diagnose diseases based on symptoms, medical history, and test results. The network can update the probabilities of different diseases based on new evidence and help in making accurate diagnoses.
• Weather Prediction: Bayesian Networks can be used to model the relationships between different weather variables such as temperature, humidity, and wind speed. By updating the probabilities of these variables based on observed data, the network can predict the likelihood of different weather conditions.

In both examples, Bayesian Networks provide a systematic framework for combining prior knowledge with observed evidence to make informed decisions. They enable problem-solving agents to reason under uncertainty and update beliefs in a principled and consistent manner.

Machine Learning Agents

Machine learning agents are a subset of artificial intelligence agents that utilize machine learning algorithms to solve problems. These agents are capable of learning from experience and improving their performance over time. They are trained on large datasets and use various techniques to analyze and interpret the data, such as deep learning and reinforcement learning.

One example of a machine learning agent is a predictive model that is trained to predict future outcomes based on historical data. For example, in finance, machine learning agents can be used to predict stock prices or identify patterns in market data to make informed investment decisions.

Another example of a machine learning agent is a virtual assistant, such as Siri or Alexa, that uses natural language processing and machine learning techniques to understand and respond to user queries and commands. These virtual assistants continuously learn from user interactions and improve their accuracy in interpreting and responding to user inputs.

Machine learning agents have revolutionized many industries and have the potential to drive innovation and improve efficiency in various domains. By leveraging the power of data and advanced algorithms, these agents can solve complex problems and make intelligent decisions that were previously not possible.

Reinforcement Learning Agents

Reinforcement learning agents are a type of problem-solving agent in artificial intelligence. These agents are designed to learn and improve their behavior through trial and error, using a system of rewards and punishments.

One example of a reinforcement learning agent is an autonomous robot that learns to navigate its environment. The robot starts with no prior knowledge of the environment and must explore and interact with its surroundings to learn how to reach a specific goal. It receives positive reinforcement, such as a reward, when it successfully performs the desired action, and negative reinforcement, such as a punishment or penalty, when it makes a mistake.

Another example of a reinforcement learning agent is a computer program that learns to play a game. The program is initially unaware of the rules and strategies of the game and must learn through repeated play. It receives positive reinforcement when it makes a winning move or achieves a high score, and negative reinforcement when it makes a losing move or receives a low score. Over time, the program learns to make better decisions and improve its performance.

Reinforcement Learning Process

The reinforcement learning process consists of the following steps:

• Observation: The agent observes the current state of the environment.
• Action: The agent selects an action to perform based on its current knowledge and strategy.
• Reward: The agent receives a reward or punishment based on the outcome of its action.
• Learning: The agent adjusts its strategy and behavior based on the received reward or punishment.
• Iteration: The process is repeated, with the agent continuously learning and improving over time.

Applications of Reinforcement Learning Agents

Reinforcement learning agents have various applications in artificial intelligence, including:

• Autonomous robotics
• Game playing
• Optimization problems
• Resource allocation
• Financial trading

These examples demonstrate how reinforcement learning agents can adapt and improve their behavior in different environments and problem-solving scenarios.

Genetic Algorithms

Genetic Algorithms are a type of problem-solving technique used in artificial intelligence. They are inspired by the process of natural selection and genetic inheritance in living organisms. These algorithms use a population of possible solutions to a problem and apply genetic operators such as selection, crossover, and mutation to evolve and improve the solutions over time.

Genetic Algorithms have been successfully applied to various optimization problems, such as finding the best combination of parameters for a machine learning model or optimizing the routing of vehicles in logistics. They are particularly useful in problems where there is no deterministic algorithm to find an optimal solution.

Here are a few examples of how Genetic Algorithms can be used:

In each of these examples, Genetic Algorithms can be used to search the solution space more efficiently and find near-optimal or optimal solutions. The population-based approach of Genetic Algorithms allows for exploration of multiple potential solutions simultaneously, increasing the chances of finding a good solution.

Overall, Genetic Algorithms are a powerful and flexible problem-solving technique in the field of artificial intelligence. They can be applied to a wide range of problems and have been proven to be effective in finding optimal or near-optimal solutions.

Swarm Intelligence

Swarm intelligence is a field of artificial intelligence that involves studying the collective behavior of multi-agent systems in order to solve complex problems. In this approach, individual agents work together as a swarm to find optimal solutions without centralized control or coordination.

Central to the concept of swarm intelligence is the idea that intelligence emerges from the interactions and cooperation of simple agents. These agents, often inspired by natural systems such as ant colonies or bird flocks, follow simple rules and communicate with each other to achieve a common goal.

Applications

• Swarm intelligence has been used in various problem-solving scenarios, including optimization problems, task allocation, and decision-making.
• One notable application is in robotics, where swarms of robots can collectively explore and map unknown environments, perform search and rescue operations, or even assemble complex structures.
• Another application is in finance, where swarm intelligence algorithms are used to analyze and predict stock market trends or optimize investment portfolios.
• One of the main advantages of swarm intelligence is its robustness and adaptability. As individual agents can communicate and adjust their behavior based on the information from their neighbors, the swarm as a whole can quickly adapt to changes or disturbances in the environment.
• Swarm intelligence also offers a scalable solution, as the performance of the swarm can improve with the addition of more agents.
• Furthermore, swarm intelligence algorithms are often computationally efficient and can handle large-scale problems that would be intractable for traditional optimization techniques.

In conclusion, swarm intelligence is a promising approach in artificial intelligence that leverages the collective intelligence of simple agents to solve complex problems. Its applications span various domains, and its advantages make it an appealing technique for solving real-world challenges.

Questions and answers

What are problem solving agents in artificial intelligence.

Problem solving agents in artificial intelligence are intelligent systems that are designed to solve complex problems by searching for the best solution based on well-defined rules and goals.

How do problem solving agents work?

Problem solving agents work by analyzing a given problem, breaking it into smaller sub-problems, and then searching for a solution by applying various problem-solving techniques, such as heuristics, pattern recognition, logical reasoning, and machine learning algorithms.

Can you give an example of a problem solving agent?

One example of a problem solving agent is a chess-playing computer program. It analyzes the current state of the chessboard, generates possible moves, evaluates their outcomes using a specified evaluation function, and then selects the move with the highest expected outcome as the solution to the problem of finding the best move.

What are some other applications of problem solving agents?

Problem solving agents have a wide range of applications in various fields. They are used in robotics to plan and execute actions, in automated planning systems to optimize resource allocation, in natural language processing to interpret and respond to user queries, and in medical diagnosis to analyze symptoms and suggest possible treatments.

Are problem solving agents capable of solving all types of problems?

No, problem solving agents are not capable of solving all types of problems. Their effectiveness depends on the specific problem domain and the availability of knowledge and resources. Some problems may be too complex or ill-defined, making it difficult for problem solving agents to find optimal solutions.

Related posts:

About the author

AI for Social Good

Add comment, cancel reply.

You must be logged in to post a comment.

BlackRock and AI: Shaping the Future of Finance

Ai and handyman: the future is here, embrace ai-powered cdps: the future of customer engagement.

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Perspective
• Published: 25 January 2022

Intelligent problem-solving as integrated hierarchical reinforcement learning

• Manfred Eppe   ORCID: orcid.org/0000-0002-5473-3221 1   nAff4 ,
• Christian Gumbsch   ORCID: orcid.org/0000-0003-2741-6551 2 , 3 ,
• Matthias Kerzel 1 ,
• Phuong D. H. Nguyen 1 ,
• Martin V. Butz   ORCID: orcid.org/0000-0002-8120-8537 2 &
• Stefan Wermter 1  

Nature Machine Intelligence volume  4 ,  pages 11–20 ( 2022 ) Cite this article

5484 Accesses

33 Citations

7 Altmetric

Metrics details

• Cognitive control
• Computational models
• Computer science
• Learning algorithms
• Problem solving

According to cognitive psychology and related disciplines, the development of complex problem-solving behaviour in biological agents depends on hierarchical cognitive mechanisms. Hierarchical reinforcement learning is a promising computational approach that may eventually yield comparable problem-solving behaviour in artificial agents and robots. However, so far, the problem-solving abilities of many human and non-human animals are clearly superior to those of artificial systems. Here we propose steps to integrate biologically inspired hierarchical mechanisms to enable advanced problem-solving skills in artificial agents. We first review the literature in cognitive psychology to highlight the importance of compositional abstraction and predictive processing. Then we relate the gained insights with contemporary hierarchical reinforcement learning methods. Interestingly, our results suggest that all identified cognitive mechanisms have been implemented individually in isolated computational architectures, raising the question of why there exists no single unifying architecture that integrates them. As our final contribution, we address this question by providing an integrative perspective on the computational challenges to develop such a unifying architecture. We expect our results to guide the development of more sophisticated cognitively inspired hierarchical machine learning architectures.

This is a preview of subscription content, access via your institution

Access options

Access Nature and 54 other Nature Portfolio journals

Get Nature+, our best-value online-access subscription

24,99 € / 30 days

cancel any time

Subscribe to this journal

Receive 12 digital issues and online access to articles

111,21 € per year

only 9,27 € per issue

Buy this article

• Purchase on Springer Link
• Instant access to full article PDF

Prices may be subject to local taxes which are calculated during checkout

Similar content being viewed by others

Phy-Q as a measure for physical reasoning intelligence

Hierarchical motor control in mammals and machines

Hierarchical generative modelling for autonomous robots

Gruber, R. et al. New Caledonian crows use mental representations to solve metatool problems. Curr. Biol. 29 , 686–692 (2019).

Article   Google Scholar  

Butz, M. V. & Kutter, E. F. How the Mind Comes into Being (Oxford Univ. Press, 2017).

Perkins, D. N. & Salomon, G. in International Encyclopedia of Education (eds. Husen T. & Postelwhite T. N.) 6452–6457 (Pergamon Press, 1992).

Botvinick, M. M., Niv, Y. & Barto, A. C. Hierarchically organized behavior and its neural foundations: a reinforcement learning perspective. Cognition 113 , 262–280 (2009).

Tomov, M. S., Yagati, S., Kumar, A., Yang, W. & Gershman, S. J. Discovery of hierarchical representations for efficient planning. PLoS Comput. Biol. 16 , e1007594 (2020).

Arulkumaran, K., Deisenroth, M. P., Brundage, M. & Bharath, A. A. Deep reinforcement learning: a brief survey. IEEE Signal Process. Mag. 34 , 26–38 (2017).

Li, Y. Deep reinforcement learning: an overview. Preprint at https://arxiv.org/abs/1701.07274 (2018).

Sutton, R. S. & Barto, A. G. Reinforcement Learning : An Introduction 2nd edn (MIT Press, 2018).

Neftci, E. O. & Averbeck, B. B. Reinforcement learning in artificial and biological systems. Nat. Mach. Intell. 1 , 133–143 (2019).

Eppe, M., Nguyen, P. D. H. & Wermter, S. From semantics to execution: integrating action planning with reinforcement learning for robotic causal problem-solving. Front. Robot. AI 6 , 123 (2019).

Oh, J., Singh, S., Lee, H. & Kohli, P. Zero-shot task generalization with multi-task deep reinforcement learning. In Proc. 34th International Conference on Machine Learning ( ICML ) (eds. Precup, D. & Teh, Y. W.) 2661–2670 (PMLR, 2017).

Sohn, S., Oh, J. & Lee, H. Hierarchical reinforcement learning for zero-shot generalization with subtask dependencies. In Proc. 32nd International Conference on Neural Information Processing Systems ( NeurIPS ) (eds Bengio S. et al.) Vol. 31, 7156–7166 (ACM, 2018).

Hegarty, M. Mechanical reasoning by mental simulation. Trends Cogn. Sci. 8 , 280–285 (2004).

Klauer, K. J. Teaching for analogical transfer as a means of improving problem-solving, thinking and learning. Instruct. Sci. 18 , 179–192 (1989).

Duncker, K. & Lees, L. S. On problem-solving. Psychol. Monographs 58, No.5 (whole No. 270), 85–101 https://doi.org/10.1037/h0093599 (1945).

Dayan, P. Goal-directed control and its antipodes. Neural Netw. 22 , 213–219 (2009).

Dolan, R. J. & Dayan, P. Goals and habits in the brain. Neuron 80 , 312–325 (2013).

O’Doherty, J. P., Cockburn, J. & Pauli, W. M. Learning, reward, and decision making. Annu. Rev. Psychol. 68 , 73–100 (2017).

Tolman, E. C. & Honzik, C. H. Introduction and removal of reward, and maze performance in rats. Univ. California Publ. Psychol. 4 , 257–275 (1930).

Google Scholar  

Butz, M. V. & Hoffmann, J. Anticipations control behavior: animal behavior in an anticipatory learning classifier system. Adaptive Behav. 10 , 75–96 (2002).

Miller, G. A., Galanter, E. & Pribram, K. H. Plans and the Structure of Behavior (Holt, Rinehart & Winston, 1960).

Botvinick, M. & Weinstein, A. Model-based hierarchical reinforcement learning and human action control. Philos. Trans. R. Soc. B Biol. Sci. 369 , 20130480 (2014).

Wiener, J. M. & Mallot, H. A. ’Fine-to-coarse’ route planning and navigation in regionalized environments. Spatial Cogn. Comput. 3 , 331–358 (2003).

Stock, A. & Stock, C. A short history of ideo-motor action. Psychol. Res. 68 , 176–188 (2004).

Hommel, B., Müsseler, J., Aschersleben, G. & Prinz, W. The theory of event coding (TEC): a framework for perception and action planning. Behav. Brain Sci. 24 , 849–878 (2001).

Hoffmann, J. in Anticipatory Behavior in Adaptive Learning Systems : Foundations , Theories and Systems (eds Butz, M. V. et al.) 44–65 (Springer, 2003).

Kunde, W., Elsner, K. & Kiesel, A. No anticipation-no action: the role of anticipation in action and perception. Cogn. Process. 8 , 71–78 (2007).

Barsalou, L. W. Grounded cognition. Annu. Rev. Psychol. 59 , 617–645 (2008).

Butz, M. V. Toward a unified sub-symbolic computational theory of cognition. Front. Psychol. 7 , 925 (2016).

Pulvermüller, F. Brain embodiment of syntax and grammar: discrete combinatorial mechanisms spelt out in neuronal circuits. Brain Lang. 112 , 167–179 (2010).

Sutton, R. S., Precup, D. & Singh, S. Between MDPs and semi-MDPs: a framework for temporal abstraction in reinforcement learning. Artif. Intell. 112 , 181–211 (1999).

Article   MathSciNet   MATH   Google Scholar  

Flash, T. & Hochner, B. Motor primitives in vertebrates and invertebrates. Curr. Opin. Neurobiol. 15 , 660–666 (2005).

Schaal, S. in Adaptive Motion of Animals and Machines (eds. Kimura, H. et al.) 261–280 (Springer, 2006).

Feldman, J., Dodge, E. & Bryant, J. in The Oxford Handbook of Linguistic Analysis (eds Heine, B. & Narrog, H.) 111–138 (Oxford Univ. Press, 2009).

Fodor, J. A. Language, thought and compositionality. Mind Lang. 16 , 1–15 (2001).

Frankland, S. M. & Greene, J. D. Concepts and compositionality: in search of the brain’s language of thought. Annu. Rev. Psychol. 71 , 273–303 (2020).

Hummel, J. E. Getting symbols out of a neural architecture. Connection Sci. 23 , 109–118 (2011).

Haynes, J. D., Wisniewski, D., Gorgen, K., Momennejad, I. & Reverberi, C. FMRI decoding of intentions: compositionality, hierarchy and prospective memory. In Proc. 3rd International Winter Conference on Brain-Computer Interface ( BCI ), 1-3 (IEEE, 2015).

Gärdenfors, P. The Geometry of Meaning : Semantics Based on Conceptual Spaces (MIT Press, 2014).

Book   MATH   Google Scholar  

Lakoff, G. & Johnson, M. Philosophy in the Flesh (Basic Books, 1999).

Eppe, M. et al. A computational framework for concept blending. Artif. Intell. 256 , 105–129 (2018).

Turner, M. The Origin of Ideas (Oxford Univ. Press, 2014).

Deci, E. L. & Ryan, R. M. Self-determination theory and the facilitation of intrinsic motivation. Am. Psychol. 55 , 68–78 (2000).

Friston, K. et al. Active inference and epistemic value. Cogn. Neurosci. 6 , 187–214 (2015).

Berlyne, D. E. Curiosity and exploration. Science 153 , 25–33 (1966).

Loewenstein, G. The psychology of curiosity: a review and reinterpretation. Psychol. Bull. 116 , 75–98 (1994).

Oudeyer, P.-Y., Kaplan, F. & Hafner, V. V. Intrinsic motivation systems for autonomous mental development. In IEEE Transactions on Evolutionary Computation (eds. Coello, C. A. C. et al.) Vol. 11, 265–286 (IEEE, 2007).

Pisula, W. Play and exploration in animals—a comparative analysis. Polish Psychol. Bull. 39 , 104–107 (2008).

Jeannerod, M. Mental imagery in the motor context. Neuropsychologia 33 , 1419–1432 (1995).

Kahnemann, D. & Tversky, A. in Judgement under Uncertainty : Heuristics and Biases (eds Kahneman, D. et al.) Ch. 14, 201–208 (Cambridge Univ. Press, 1982).

Wells, G. L. & Gavanski, I. Mental simulation of causality. J. Personal. Social Psychol. 56 , 161–169 (1989).

Taylor, S. E., Pham, L. B., Rivkin, I. D. & Armor, D. A. Harnessing the imagination: mental simulation, self-regulation and coping. Am. Psychol. 53 , 429–439 (1998).

Kaplan, F. & Oudeyer, P.-Y. in Embodied Artificial Intelligence , Lecture Notes in Computer Science Vol. 3139 (eds Iida, F. et al.) 259–270 (Springer, 2004).

Schmidhuber, J. Formal theory of creativity, fun, and intrinsic motivation. IEEE Trans. Auton. Mental Dev. 2 , 230–247 (2010).

Friston, K., Mattout, J. & Kilner, J. Action understanding and active inference. Biol. Cybern. 104 , 137–160 (2011).

Oudeyer, P.-Y. Computational theories of curiosity-driven learning. In The New Science of Curiosity (ed. Goren Gordon), 43-72 (Nova Science Publishers, 2018); https://arxiv.org/abs/1802.10546

Colombo, M. & Wright, C. First principles in the life sciences: the free-energy principle, organicism and mechanism. Synthese 198 , 3463–3488 (2021).

Article   MathSciNet   Google Scholar  

Huang, Y. & Rao, R. P. Predictive coding. WIREs Cogn. Sci. 2 , 580–593 (2011).

Friston, K. The free-energy principle: a unified brain theory? Nat. Rev. Neurosci. 11 , 127–138 (2010).

Knill, D. C. & Pouget, A. The Bayesian brain: the role of uncertainty in neural coding and computation. Trends Neurosci. 27 , 712–719 (2004).

Clark, A. Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behav. Brain Sci. 36 , 181–204 (2013).

Clark, A. Surfing Uncertainty : Prediction , Action and the Embodied Mind (Oxford Univ. Press, 2016).

Zacks, J. M., Speer, N. K., Swallow, K. M., Braver, T. S. & Reyonolds, J. R. Event perception: a mind/brain perspective. Psychol. Bull. 133 , 273–293 (2007).

Eysenbach, B., Ibarz, J., Gupta, A. & Levine, S. Diversity is all you need: learning skills without a reward function. In International Conference on Learning Representations (ICLR, 2019).

Frans, K., Ho, J., Chen, X., Abbeel, P. & Schulman, J. Meta learning shared hierarchies. In Proc. International Conference on Learning Representations https://openreview.net/pdf?id=SyX0IeWAW (ICLR, 2018).

Heess, N. et al. Learning and transfer of modulated locomotor controllers. Preprint at https://arxiv.org/abs/1610.05182 (2016).

Jiang, Y., Gu, S., Murphy, K. & Finn, C. Language as an abstraction for hierarchical deep reinforcement learning. In Neural Information Processing Systems ( NeurIPS ) (eds. Wallach, H. et al.) 9414–9426 (ACM, 2019).

Li, A. C., Florensa, C., Clavera, I. & Abbeel, P. Sub-policy adaptation for hierarchical reinforcement learning. In Proc. International Conference on Learning Representations https://openreview.net/forum?id=ByeWogStDS (ICLR, 2020).

Qureshi, A. H. et al. Composing task-agnostic policies with deep reinforcement learning. In Proc. International Conference on Learning Representations https://openreview.net/forum?id=H1ezFREtwH (ICLR, 2020).

Sharma, A., Gu, S., Levine, S., Kumar, V. & Hausman, K. Dynamics-aware unsupervised discovery of skills. In Proc. International Conference on Learning Representations https://openreview.net/forum?id=HJgLZR4KvH (ICLR, 2020).

Tessler, C., Givony, S., Zahavy, T., Mankowitz, D. J. & Mannor, S. A deep hierarchical approach to lifelong learning in minecraft. In Proc. 31st AAAI Conference on Artificial Intelligence 1553–1561 (AAAI, 2017).

Vezhnevets, A. et al. Strategic attentive writer for learning macro-actions. In Neural Information Processing Systems ( NIPS ) (eds. Lee, D. et al.) 3494–3502 (NIPS, 2016).

Devin, C., Gupta, A., Darrell, T., Abbeel, P. & Levine, S. Learning modular neural network policies for multi-task and multi-robot transfer. In Proc. International Conference on Robotics and Automation ( ICRA ) (eds. Okamura, A. et al.) 2169–2176 (IEEE, 2017).

Hejna, D. J., Abbeel, P. & Pinto, L. Hierarchically decoupled morphological transfer. In Proc. International Conference on Machine Learning ( ICML ) (eds. Daumé III, H. & Singh, A.) 11409–11420 (PMLR, 2020).

Hamrick, J. B. et al. On the role of planning in model-based deep reinforcement learning. In Proc. International Conference on Learning Representations https://openreview.net/pdf?id=IrM64DGB21 (ICLR, 2021).

Sutton, R. S. Integrated architectures for learning, planning, and reacting based on approximating dynamic programming. In Proc. 7th International Conference on Machine Learning ( ICML ) (eds. Porter, B. W. & Mooney, R. J.) 216–224 (Morgan Kaufmann, 1990).

Nau, D. et al. SHOP2: an HTN planning system. J. Artif. Intell. Res. 20 , 379–404 (2003).

Article   MATH   Google Scholar  

Lyu, D., Yang, F., Liu, B. & Gustafson, S. SDRL: interpretable and data-efficient deep reinforcement learning leveraging symbolic planning. In Proc. AAAI Conference on Artificial Intelligence Vol. 33, 2970–2977 (AAAI, 2019).

Ma, A., Ouimet, M. & Cortés, J. Hierarchical reinforcement learning via dynamic subspace search for multi-agent planning. Auton. Robot. 44 , 485–503 (2020).

Bacon, P.-L., Harb, J. & Precup, D. The option-critic architecture. In Proc. 31st AAAI Conference on Artificial Intelligence 1726–1734 (AAAI, 2017).

Dietterich, T. G. State abstraction in MAXQ hierarchical reinforcement learning. In Advances in Neural Information Processing Systems ( NIPS ) (eds. Solla, S. et al.) Vol. 12, 994–1000 (NIPS, 1999).

Kulkarni, T. D., Narasimhan, K. R., Saeedi, A. & Tenenbaum, J. B. Hierarchical deep reinforcement learning: integrating temporal abstraction and intrinsic motivation. In Neural Information Processing Systems ( NIPS ) (eds. Lee, D. et al.) 3675–3683 (NIPS, 2016).

Shankar, T., Pinto, L., Tulsiani, S. & Gupta, A. Discovering motor programs by recomposing demonstrations. In Proc. International Conference on Learning Representations https://openreview.net/attachment?id=rkgHY0NYwr&name=original_pdf (ICLR, 2020).

Vezhnevets, A. S., Wu, Y. T., Eckstein, M., Leblond, R. & Leibo, J. Z. Options as responses: grounding behavioural hierarchies in multi-agent reinforcement learning. In Proc. International Conference on Machine Learning ( ICML ) (eds. Daumé III, H. & Singh, A.) 9733–9742 (PMLR, 2020).

Ghazanfari, B., Afghah, F. & Taylor, M. E. Sequential association rule mining for autonomously extracting hierarchical task structures in reinforcement learning. IEEE Access 8 , 11782–11799 (2020).

Levy, A., Konidaris, G., Platt, R. & Saenko, K. Learning multi-level hierarchies with hindsight. In Proc. International Conference on Learning Representations https://openreview.net/pdf?id=ryzECoAcY7 (ICLR, 2019).

Nachum, O., Gu, S., Lee, H. & Levine, S. Data-efficient hierarchical reinforcement learning. In Proc. 32nd International Conference on Neural Information Processing Systems (NIPS) (eds. Bengio, S. et al.) 3307–3317 (NIPS, 2018).

Rafati, J. & Noelle, D. C. Learning representations in model-free hierarchical reinforcement learning. In Proc. 33rd AAAI Conference on Artificial Intelligence 10009–10010 (AAAI, 2019).

Röder, F., Eppe, M., Nguyen, P. D. H. & Wermter, S. Curious hierarchical actor-critic reinforcement learning. In Proc. International Conference on Artificial Neural Networks ( ICANN ) (eds. Farkaš, I. et al.) 408–419 (Springer, 2020).

Zhang, T., Guo, S., Tan, T., Hu, X. & Chen, F. Generating adjacency-constrained subgoals in hierarchical reinforcement learning. In Neural Information Processing Systems ( NIPS ) (eds. Larochelle, H. et al.) 21579-21590 (NIPS, 2020).

Lample, G. & Chaplot, D. S. Playing FPS games with deep reinforcement learning. In Proc. 31st AAAI Conference on Artificial Intelligence 2140–2146 (AAAI, 2017).

Vezhnevets, A. S. et al. FeUdal networks for hierarchical reinforcement learning. In Proc. 34th International Conference on Machine Learning ( ICML ) (eds. Precup, D. & Teh, Y. W.) Vol. 70, 3540–3549 (PMLR, 2017).

Wulfmeier, M. et al. Compositional Transfer in Hierarchical Reinforcement Learning. In Robotics: Science and System XVI (RSS) (eds. Toussaint M. et al.) (Robotics: Science and Systems Foundation, 2020); https://arxiv.org/abs/1906.11228

Yang, Z., Merrick, K., Jin, L. & Abbass, H. A. Hierarchical deep reinforcement learning for continuous action control. IEEE Trans. Neural Netw. Learn. Syst. 29 , 5174–5184 (2018).

Toussaint, M., Allen, K. R., Smith, K. A. & Tenenbaum, J. B. Differentiable physics and stable modes for tool-use and manipulation planning. In Proc. Robotics : Science and Systems XIV ( RSS ) (eds. Kress-Gazit, H. et al.) https://ipvs.informatik.uni-stuttgart.de/mlr/papers/18-toussaint-RSS.pdf (Robotics: Science and Systems Foundation, 2018).

Akrour, R., Veiga, F., Peters, J. & Neumann, G. Regularizing reinforcement learning with state abstraction. In Proc. IEEE / RSJ International Conference on Intelligent Robots and Systems ( IROS ) 534–539 (IEEE, 2018).

Schaul, T. & Ring, M. Better generalization with forecasts. In Proc. 23rd International Joint Conference on Artificial Intelligence ( IJCAI ) (ed. Rossi, F.) 1656–1662 (AAAI, 2013).

Colas, C., Akakzia, A., Oudeyer, P.-Y., Chetouani, M. & Sigaud, O. Language-conditioned goal generation: a new approach to language grounding for RL. Preprint at https://arxiv.org/abs/2006.07043 (2020).

Blaes, S., Pogancic, M. V., Zhu, J. J. & Martius, G. Control what you can: intrinsically motivated task-planning agent. Neural Inf. Process. Syst. 32 , 12541–12552 (2019).

Haarnoja, T., Hartikainen, K., Abbeel, P. & Levine, S. Latent space policies for hierarchical reinforcement learning. In Proc. International Conference on Machine Learning ( ICML ) (eds. Dy, J. & Krause, A.) Vol. 4, 2965–2975 (PMLR, 2018).

Rasmussen, D., Voelker, A. & Eliasmith, C. A neural model of hierarchical reinforcement learning. PLoS ONE 12 , e0180234 (2017).

Riedmiller, M. et al. Learning by playing—solving sparse reward tasks from scratch. In Proc. International Conference on Machine Learning ( ICML ) (eds. Dy, J. & Krause, A.) Vol. 10, 6910–6919 (PMLR, 2018).

Yang, F., Lyu, D., Liu, B. & Gustafson, S. PEORL: integrating symbolic planning and hierarchical reinforcement learning for robust decision-making. In Proc. 27th International Joint Conference on Artificial Intelligence ( IJCAI ) (ed. Lang, J.) 4860–4866 (IJCAI, 2018).

Machado, M. C., Bellemare, M. G. & Bowling, M. A Laplacian framework for option discovery in reinforcement learning. In Proc. International Conference on Machine Learning (ICML) (eds. Precup, D. & Teh, Y. W.) Vol. 5, 3567–3582 (PMLR, 2017).

Pathak, D., Agrawal, P., Efros, A. A. & Darrell, T. Curiosity-driven exploration by self-supervised prediction. In Proc. 34th International Conference on Machine Learning ( ICML ) (eds. Precup, D. & Teh, Y. W.) 2778–2787 (PMLR, 2017).

Schillaci, G. et al. Intrinsic motivation and episodic memories for robot exploration of high-dimensional sensory spaces. Adaptive Behav. 29 549–566 (2020).

Colas, C., Fournier, P., Sigaud, O., Chetouani, M. & Oudeyer, P.-Y. CURIOUS: intrinsically motivated modular multi-goal reinforcement learning. In Proc. International Conference on Machine Learning ( ICML ) (eds. Chaudhuri, K. & Salakhutdinov, R.) 1331–1340 (PMLR, 2019).

Hafez, M. B., Weber, C., Kerzel, M. & Wermter, S. Improving robot dual-system motor learning with intrinsically motivated meta-control and latent-space experience imagination. Robot. Auton. Syst. 133 , 103630 (2020).

Yamamoto, K., Onishi, T. & Tsuruoka, Y. Hierarchical reinforcement learning with abductive planning. In Proc. ICML / IJCAI / AAMAS 2018 Workshop on Planning and Learning ( PAL-18 ) (2018).

Wu, B., Gupta, J. K. & Kochenderfer, M. J. Model primitive hierarchical lifelong reinforcement learning . In Proc. International Joint Conference on Autonomous Agents and Multiagent Systems ( AAMAS ) (eds. Agmon, N. et al.) Vol. 1, 34–42 (IFAAMAS, 2019).

Li, Z., Narayan, A. & Leong, T. Y. An efficient approach to model-based hierarchical reinforcement learning. In Proc. 31st AAAI Conference on Artificial Intelligence 3583–3589 (AAAI, 2017).

Hafner, D., Lillicrap, T. & Norouzi, M. Dream to control: learning behaviors by latent imagination. In Proc. International Conference on Learning Representations https://openreview.net/pdf?id=S1lOTC4tDS (ICLR, 2020).

Deisenroth, M. P., Rasmussen, C. E. & Fox, D. Learning to control a low-cost manipulator using data-efficient reinforcement learning. In Robotics : Science and Systems VII ( RSS ) (eds. Durrant-Whyte, H. et al.) 57–64 (Robotics: Science and Systems Foundation, 2011).

Ha, D. & Schmidhuber, J. Recurrent world models facilitate policy evolution. In Proc. 32nd International Conference on Neural Information Processing Systems (NeurIPS) (eds. Bengio, S. et al.) 2455–2467 (NIPS, 2018).

Battaglia, P. W. et al. Relational inductive biases, deep learning and graph networks. Preprint at https://arxiv.org/abs/1806.01261 (2018).

Andrychowicz, M. et al. Hindsight experience replay. In Proc. Neural Information Processing Systems ( NIPS ) (eds. Guyon I. et al.) 5048–5058 (NIPS, 2017); https://papers.nips.cc/paper/7090-hindsight-experience-replay.pdf

Schwartenbeck, P. et al. Computational mechanisms of curiosity and goal-directed exploration. eLife 8 , e41703 (2019).

Haarnoja, T., Zhou, A., Abbeel, P. & Levine, S. Soft actor-critic: off-policy maximum entropy deep reinforcement learning with a stochastic actor. In Proc. International Conference on Machine Learning ( ICML ) (eds. Dy, J. & Krause, A.) 1861–1870 (PMLR, 2018).

Yu, A. J. & Dayan, P. Uncertainty, neuromodulation and attention. Neuron 46 , 681–692 (2005).

Baldwin, D. A. & Kosie, J. E. How does the mind render streaming experience as events? Top. Cogn. Sci. 13 , 79–105 (2021).

Download references

Acknowledgements

We acknowledge funding from the DFG (projects IDEAS, LeCAREbot, TRR169, SPP 2134, RTG 1808 and EXC 2064/1), the Humboldt Foundation and Max Planck Research School IMPRS-IS.

Author information

Manfred Eppe

Present address: Hamburg University of Technology, Hamburg, Germany

Authors and Affiliations

Universität Hamburg, Hamburg, Germany

Manfred Eppe, Matthias Kerzel, Phuong D. H. Nguyen & Stefan Wermter

University of Tübingen, Tübingen, Germany

Christian Gumbsch & Martin V. Butz

Max Planck Institute for Intelligent Systems, Tübingen, Germany

Christian Gumbsch

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Manfred Eppe .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary information.

Supplementary Boxes 1–6 and Table 1.

Rights and permissions

Reprints and permissions

About this article

Cite this article.

Eppe, M., Gumbsch, C., Kerzel, M. et al. Intelligent problem-solving as integrated hierarchical reinforcement learning. Nat Mach Intell 4 , 11–20 (2022). https://doi.org/10.1038/s42256-021-00433-9

Download citation

Received : 18 December 2020

Accepted : 07 December 2021

Published : 25 January 2022

Issue Date : January 2022

DOI : https://doi.org/10.1038/s42256-021-00433-9

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

This article is cited by

Efficient stacking and grasping in unstructured environments.

• Jinbiao Zhu

Journal of Intelligent & Robotic Systems (2024)

Four attributes of intelligence, a thousand questions

• Matthieu Bardal
• Eric Chalmers

Biological Cybernetics (2023)

An Alternative to Cognitivism: Computational Phenomenology for Deep Learning

• Pierre Beckmann
• Guillaume Köstner
• Inês Hipólito

Minds and Machines (2023)

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing: AI and Robotics newsletter — what matters in AI and robotics research, free to your inbox weekly.

Table of Contents

What is an agent in ai, the functions of an artificial intelligence agent, the number and types of agents in artificial intelligence, the structure of agents in artificial intelligence, what are agents in artificial intelligence composed of, how to improve the performance of intelligent agents, all about problem-solving agents in artificial intelligence, choose the right program, can you picture a career in artificial intelligence, exploring intelligent agents in artificial intelligence.

Artificial Intelligence, typically abbreviated to AI, is a fascinating field of Information Technology that finds its way into many aspects of modern life. Although it may seem complex, and yes, it is, we can gain a greater familiarity and comfort with AI by exploring its components separately. When we learn how the pieces fit together, we can better understand and implement them.

That’s why today we’re tackling the intelligent Agent in AI. This article defines intelligent agents in Artificial Intelligence , AI agent functions and structure, and the number and types of agents in AI.

Let’s define what we mean by an intelligent agent in AI.

Okay, did anyone, upon hearing the term “intelligent agent,” immediately picture a well-educated spy with a high IQ? No? Anyway, in the context of the AI field, an “agent” is an independent program or entity that interacts with its environment by perceiving its surroundings via sensors, then acting through actuators or effectors.

Agents use their actuators to run through a cycle of perception, thought, and action. Examples of agents in general terms include:

• Software: This Agent has file contents, keystrokes, and received network packages that function as sensory input, then act on those inputs, displaying the output on a screen.
• Human: Yes, we’re all agents. Humans have eyes, ears, and other organs that act as sensors, and hands, legs, mouths, and other body parts act as actuators.
• Robotic: Robotic agents have cameras and infrared range finders that act as sensors, and various servos and motors perform as actuators.

Intelligent agents in AI are autonomous entities that act upon an environment using sensors and actuators to achieve their goals. In addition, intelligent agents may learn from the environment to achieve those goals. Driverless cars and the Siri virtual assistant are examples of intelligent agents in AI.

These are the main four rules all AI agents must adhere to:

• Rule 1: An AI agent must be able to perceive the environment.
• Rule 2: The environmental observations must be used to make decisions.
• Rule 3: The decisions should result in action.
• Rule 4: The action taken by the AI agent must be a rational. Rational actions are actions that maximize performance and yield the best positive outcome.

Artificial Intelligence agents perform these functions continuously:

• Perceiving dynamic conditions in the environment
• Acting to affect conditions in the environment
• Using reasoning to interpret perceptions
• Problem-solving
• Drawing inferences
• Determining actions and their outcomes

There are five different types of intelligent agents used in AI. They are defined by their range of capabilities and intelligence level:

• Reflex Agents: These agents work here and now and ignore the past. They respond using the event-condition-action rule. The ECA rule applies when a user initiates an event, and the Agent turns to a list of pre-set conditions and rules, resulting in pre-programmed outcomes.
• Model-based Agents: These agents choose their actions like reflex agents do, but they have a better comprehensive view of the environment. An environmental model is programmed into the internal system, incorporating into the Agent's history.
• Goal-based agents: These agents build on the information that a model-based agent stores by augmenting it with goal information or data regarding desirable outcomes and situations.
• Utility-based agents: These are comparable to the goal-based agents, except they offer an extra utility measurement. This measurement rates each possible scenario based on the desired result and selects the action that maximizes the outcome. Rating criteria examples include variables such as success probability or the number of resources required.
• Learning agents: These agents employ an additional learning element to gradually improve and become more knowledgeable over time about an environment. The learning element uses feedback to decide how the performance elements should be gradually changed to show improvement.

Agents in Artificial Intelligence follow this simple structural formula:

Architecture + Agent Program = Agent

These are the terms most associated with agent structure:

• Architecture: This is the machinery or platform that executes the agent.
• Agent Function: The agent function maps a precept to the Action, represented by the following formula: f:P* - A
• Agent Program: The agent program is an implementation of the agent function. The agent program produces function f by executing on the physical architecture.

Many AI Agents use the PEAS model in their structure. PEAS is an acronym for Performance Measure, Environment, Actuators, and Sensors. For instance, take a vacuum cleaner.

• Performance: Cleanliness and efficiency
• Environment: Rug, hardwood floor, living room
• Actuator: Brushes, wheels, vacuum bag
• Sensors: Dirt detection sensor, bump sensor

Here’s a diagram that illustrates the structure of a utility-based agent, courtesy of Researchgate.net.

Become a AI & Machine Learning Professional

• $267 billion Expected Global AI Market Value By 2027
• 37.3% Projected CAGR Of The Global AI Market From 2023-2030
• $15.7 trillion Expected Total Contribution Of AI To The Global Economy By 2030

Artificial Intelligence Engineer

• Industry-recognized AI Engineer Master’s certificate from Simplilearn
• Dedicated live sessions by faculty of industry experts

Professional Certificate Program in Generative AI and Machine Learning

Here's what learners are saying regarding our programs:.

Indrakala Nigam Beniwal

Technical consultant , land transport authority (lta) singapore.

I completed a Master's Program in Artificial Intelligence Engineer with flying colors from Simplilearn. Thanks to the course teachers and others associated with designing such a wonderful learning experience.

Aman Kukreti

Business process analyst , ds group.

The generative ai machine learning module was an eye opener for me. The mentor's teaching style in the weekend sessions was spot on. While there's a lot of Gen AI content online, this module stood out for its structured approach and the mentor's interactive guidance. It made learning about AI easy, even for beginners like me.

Agents in Artificial Intelligence contain the following properties:

• Enrironment

Flexibility

• Proactiveness

Using Response Rules

Now, let's discuss these in detail.

Environment

The agent is situated in a given environment.

The agent can operate without direct human intervention or other software methods. It controls its activities and internal environment. The agent independently which steps it will take in its current condition to achieve the best improvements. The agent achieves autonomy if its performance is measured by its experiences in the context of learning and adapting.

• Reactive: Agents must recognize their surroundings and react to the changes within them.
• Proactive: Agents shouldn’t only act in response to their surroundings but also be able to take the initiative when appropriate and effect an opportunistic, goal-directed performance.
• Social: Agents should work with humans or other non-human agents.
• Reactive systems maintain ongoing interactions with their environment, responding to its changes.
• The program’s environment may be guaranteed, not concerned about its success or failure.
• Most environments are dynamic, meaning that things are constantly in a state of change, and information is incomplete.
• Programs must make provisions for the possibility of failure.

Pro-Activeness

Taking the initiative to create goals and try to meet them.

The goal for the agent is directed behavior, having it do things for the user.

• Mobility: The agent must have the ability to actuate around a system.
• Veracity: If an agent’s information is false, it will not communicate.
• Benevolence: Agents don’t have contradictory or conflicting goals. Therefore, every Agent will always try to do what it is asked.
• Rationality: The agent will perform to accomplish its goals and not work in a way that opposes or blocks them.
• Learning: An agent must be able to learn.

When tackling the issue of how to improve intelligent Agent performances, all we need to do is ask ourselves, “How do we improve our performance in a task?” The answer, of course, is simple. We perform the task, remember the results, then adjust based on our recollection of previous attempts.

Artificial Intelligence Agents improve in the same way. The Agent gets better by saving its previous attempts and states, learning how to respond better next time. This place is where Machine Learning and Artificial Intelligence meet.

Problem-solving Agents in Artificial Intelligence employ several algorithm s and analyses to develop solutions. They are:

• Search Algorithms: Search techniques are considered universal problem-solving methods. Problem-solving or rational agents employ these algorithms and strategies to solve problems and generate the best results.

Uninformed Search Algorithms: Also called a Blind search, uninformed searches have no domain knowledge, working instead in a brute-force manner.

Informed Search Algorithms: Also known as a Heuristic search, informed searches use domain knowledge to find the search strategies needed to solve the problem.

• Hill Climbing Algorithms: Hill climbing algorithms are local search algorithms that continuously move upwards, increasing their value or elevation until they find the best solution to the problem or the mountain's peak.

Hill climbing algorithms are excellent for optimizing mathematical problem-solving. This algorithm is also known as a "greedy local search" because it only checks out its good immediate neighbor.

• Means-Ends Analysis: The means-end analysis is a problem-solving technique used to limit searches in Artificial Intelligence programs , combining Backward and Forward search techniques.

The means-end analysis evaluates the differences between the Initial State and the Final State, then picks the best operators that can be used for each difference. The analysis then applies the operators to each matching difference, reducing the current and goal state difference.

Supercharge your career in AI and ML with Simplilearn's comprehensive courses. Gain the skills and knowledge to transform industries and unleash your true potential. Enroll now and unlock limitless possibilities!

Program Name AI Engineer Post Graduate Program In Artificial Intelligence Post Graduate Program In Artificial Intelligence Geo All Geos All Geos IN/ROW University Simplilearn Purdue Caltech Course Duration 11 Months 11 Months 11 Months Coding Experience Required Basic Basic No Skills You Will Learn 10+ skills including data structure, data manipulation, NumPy, Scikit-Learn, Tableau and more. 16+ skills including chatbots, NLP, Python, Keras and more. 8+ skills including Supervised & Unsupervised Learning Deep Learning Data Visualization, and more. Additional Benefits Get access to exclusive Hackathons, Masterclasses and Ask-Me-Anything sessions by IBM Applied learning via 3 Capstone and 12 Industry-relevant Projects Purdue Alumni Association Membership Free IIMJobs Pro-Membership of 6 months Resume Building Assistance Upto 14 CEU Credits Caltech CTME Circle Membership Cost $$ $$$$ $$$$ Explore Program Explore Program Explore Program

As you can infer from what we’ve covered, the field of Artificial Intelligence is complicated and involved. However, AI is the way of the future and is making its way into every area of our lives. If you want to join the AI revolution and pursue a career in the field, Simplilearn has everything you need.

The Caltech Post Graduate Program in AI & ML program, built and delivered in partnership with Caltech CTME and IBM, will help you master vital Artificial Intelligence concepts such as Data Science with Python , Machine Learning , Deep Learning, and Natural Language Programming (NLP). In addition, the course offers exclusive hackathons and “Ask me anything” sessions held by IBM. Before you know it, the live sessions, practical labs, and hands-on projects give you job-ready AI certification.

Glassdoor says that Artificial Intelligence Engineers in the United States can earn an average of $119,316 per year. However, a similar position in India makes a yearly average of ₹986,682.

Check out Simplilearn today, and get started on that exciting new career in Artificial Intelligence!

1. What are Intelligent Agents in Artificial Intelligence?

Intelligent Agents in AI are autonomous entities that perceive their environment and make decisions to achieve specific goals.

2. How do Intelligent Agents contribute to AI?

Intelligent Agents enhance AI by autonomously processing information and performing actions to meet set objectives.

3. What are examples of Intelligent Agents in AI?

Examples include recommendation systems, self-driving cars, and voice assistants like Siri or Alexa.

4. How do Intelligent Agents perceive their environment?

Intelligent Agents use sensors to perceive their environment, gathering data for decision-making.

5. What role do Intelligent Agents play in Machine Learning?

In Machine Learning, Intelligent Agents can learn and improve their performance without explicit programming.

6. Are Intelligent Agents the same as AI robots?

Not all Intelligent Agents are robots, but all AI robots can be considered Intelligent Agents.

7. What's the future of Intelligent Agents in AI?

The future of Intelligent Agents is promising, with potential advancements in automation, decision-making, and problem-solving.

8. How do Intelligent Agents impact everyday life?

Intelligent Agents impact our lives by providing personalized recommendations, automating tasks, and enhancing user experiences.

9. How do Intelligent Agents make decisions in AI?

Intelligent Agents make decisions based on their perception of the environment and pre-defined goals.

10. Can anyone use Intelligent Agents in AI?

Yes, anyone with the right tools and understanding can utilize Intelligent Agents in AI.

Our AI & Machine Learning Courses Duration And Fees

AI & Machine Learning Courses typically range from a few weeks to several months, with fees varying based on program and institution.

Get Free Certifications with free video courses

AI & Machine Learning

Machine Learning using Python

Artificial Intelligence Beginners Guide: What is AI?

Learn from Industry Experts with free Masterclasses

Kickstart Your Gen AI & ML Career on a High-Growth Path in 2024 with IIT Kanpur

Ethics in Generative AI: Why It Matters and What Benefits It Brings

Navigating the GenAI Frontier

Recommended Reads

Artificial Intelligence Career Guide: A Comprehensive Playbook to Becoming an AI Expert

What is Artificial Intelligence and Why Gain AI Certification

What is Artificial Intelligence: Types, History, and Future

Introduction to Artificial Intelligence: A Beginner's Guide

Top Artificial Intelligence Interview Questions

How Does AI Work

Get Affiliated Certifications with Live Class programs

• PMP, PMI, PMBOK, CAPM, PgMP, PfMP, ACP, PBA, RMP, SP, and OPM3 are registered marks of the Project Management Institute, Inc.

The Design and application of RAG-based conversational agents for collaborative problem solving

New citation alert added.

This alert has been successfully added and will be sent to:

You will be notified whenever a record that you have chosen has been cited.

To manage your alert preferences, click on the button below.

New Citation Alert!

Please log in to your account

Information & Contributors

Bibliometrics & citations, recommendations, exploring effects of conversational fillers on user perception of conversational agents.

Through technological advancements in various areas of our lives, Conversational Agents progressed in their human-likeness. In the field of HCI, however, the use of conversational fillers (e.g., "um," "uh," etc.) by Conversational Agents have not been ...

Acoustic-Prosodic Entrainment and Rapport in Collaborative Learning Dialogues

In spoken dialogue analysis, the speech signal is a rich source of information. We explore in this paper how low level features of the speech signal, such as pitch, loudness, and speaking rate, can inform a model of student interaction in collaborative ...

Gaze awareness in conversational agents: Estimating a user's conversational engagement from eye gaze

In face-to-face conversations, speakers are continuously checking whether the listener is engaged in the conversation, and they change their conversational strategy if the listener is not fully engaged. With the goal of building a conversational agent ...

Information

Published in.

Association for Computing Machinery

New York, NY, United States

Publication History

Permissions, check for updates, author tags.

• Collaborative problem solving
• Conversational agent
• Research-article
• Refereed limited

Funding Sources

• the National Natural Science Foundation in China

Contributors

Other metrics, bibliometrics, article metrics.

• 0 Total Citations
• 0 Total Downloads
• Downloads (Last 12 months) 0
• Downloads (Last 6 weeks) 0

View Options

Login options.

Check if you have access through your login credentials or your institution to get full access on this article.

Full Access

View options.

View or Download as a PDF file.

View online with eReader .

HTML Format

View this article in HTML Format.

Share this Publication link

Copying failed.

Share on social media

Affiliations, export citations.

• Please download or close your previous search result export first before starting a new bulk export. Preview is not available. By clicking download, a status dialog will open to start the export process. The process may take a few minutes but once it finishes a file will be downloadable from your browser. You may continue to browse the DL while the export process is in progress. Download
• Download citation
• Copy citation

We are preparing your search results for download ...

We will inform you here when the file is ready.

Your file of search results citations is now ready.

Your search export query has expired. Please try again.

Published:  06 August 2024 Contributor : Anna Gutowska

A multiagent system (MAS) consists of multiple artificial intelligence (AI) agents working collectively to perform tasks on behalf of a user or another system.

Each agent within a MAS has individual properties but all agents behave collaboratively to lead to desired global properties. 1 Multiagent systems are valuable in completing large-scale, complex tasks that can encompass hundreds, if not thousands, of agents. 2

Central to this idea are artificial intelligence (AI)  agents. An AI agent refers to a system or program capable of autonomously performing tasks on behalf of a user or another system by designing its workflow and using available tools. At the core of AI agents are  large language models (LLMs) . These intelligent agents leverage the advanced natural language processing techniques of LLMs to comprehend and respond to user inputs. Agents work through problems step-by-step and determine when to call on external tools. What differentiates AI agents from traditional LLMs is the use of tools and the ability to design a plan of action. The tools available to an agent can include external datasets, web searches and application programming interfaces (APIs). Similarly to human decision-making, AI agents can also update their memory as they acquire new information. The information-sharing, tool usage and adaptive learning allow AI agents to be more general purpose than traditional LLMs.

For more information about single agent systems, see our detailed AI agent content . 

Use this model selection framework to choose the most appropriate model while balancing your performance requirements with cost, risks and deployment needs.

Single agent intelligent systems engage with their environment to autonomously plan, call tools and produce responses. The tools made available to an agent provide information that is otherwise unavailable to the agent. As previously described, this information can be a database acquired through an API or another agent. There is a distinction here between single and multiagent systems. When calling another agent as a tool, that secondary agent is part of the original agent’s environmental stimuli. That information is acquired and no further cooperation takes place. Whereas multiagent systems differ by involving all agents within the environment to model each other’s goals, memory and plan of action. 4 Communication between agents can be direct or indirect through altering the shared environment.

Each entity within a multiagent system is an autonomous agent to some extent. This autonomy is typically seen by the agent’s planning, tool calling and general reasoning. In a multiagent system, agents remain autonomous but also cooperate and coordinate in agent structures. 3 To solve complex problems, agent communication and distributed problem-solving are key. This type of agent interaction can be described as multiagent  reinforcement learning . The information shared through this form of learning can include instantaneous information acquired through sensors or actions. Additionally, an agent’s experiences in the form of episodic information can be shared. These episodes can be sequences of sensations, actions and learned policies. Finally, agents can share their experiences in real-time to prevent other agents from repetitively learning the same policies. 5

Individual agents are powerful on their own. They can create subtasks, use tools and learn through their interactions. The collective behavior of multiagent systems increases the potential for accuracy, adaptability and scalability. Multiagent systems tend to outperform single-agent systems due to the larger pool of shared resources, optimization and automation. Instead of multiple agents learning the same policies, one can share learned experiences to optimize time complexity and efficiency. 5

Centralized networks

Multiagent systems can operate under various architectures. In centralized networks, a central unit contains the global knowledge base, connects the agents and oversees their information. A strength of this structure is the ease of communication between agents and uniform knowledge. A weakness of the centrality is the dependence on the central unit; if it fails, the entire system of agents fails. 6

Decentralized networks

Agents in decentralized networks share information with their neighboring agents instead of a global knowledge base. Some benefits of decentralized networks are robustness and modularity. The failure of one agent does not cause the overall system to fail since there is no central unit. One challenge of decentralized agents is coordinating their behavior to benefit other cooperating agents. 7

There are also many ways of organizing agents within a multiagent system including:

Hierarchical structure

A hierarchical structure is tree-like and contains agents with varying levels of autonomy. Within a simple hierarchical structure, one agent can have the decision-making authority. In a uniform hierarchical structure, the responsibility can be distributed among multiple agents. 8

Holonic structure

Within this architecture type, agents are grouped into holarchies. A holon is an entity that cannot operate without its components. For instance, the human body is a holon because it cannot function without working organs. 9 Similarly, in holonic multiagent systems, the leading agent can have multiple subagents while appearing to be a singular entity. 8 These subagents can also play roles in other holons. These hierarchical structures are self-organized and created to achieve a goal through the collaboration of the subagents.

Coalition structure

Coalitions are helpful in cases of underperforming singular agents in a group. In these situations, agents temporarily unite to boost utility or performance. Once the desired performance is reached, the coalitions are disbursed. It can become difficult to maintain these coalitions in dynamic environments. Regrouping is often necessary to enhance performance. 9

Teams are similar in structure to coalitions. In teams, agents cooperate to improve the performance of the group. Agents in teams do not work independently, unlike in coalitions. Agents in teams are much more dependent on one another and their structure is more hierarchical than coalitions. 8

The behaviors of agents within a multiagent system often reflect behaviors occurring in nature. The following agent behaviors can apply to both multisoftware and multirobot agents.

The collective behavior seen in multiagent systems can resemble that of birds, fish and humans. In these systems, agents share an objective and require some organization to coordinate their behavior. Flocking pertains to directional synchronization and the structure of these flocks can be described by these heuristics:  10

• Separation: attempt to avoid collision with nearby agents.
• Alignment: attempt to match the velocity of nearby agents.
• Cohesion: attempt to remain close to other agents.

In the context of software agents, this coordination is crucial for multiagent systems managing transportation networks such as railroad systems.

The spatial positioning of agents in a multiagent system can be compared to the swarming that occurs in nature. For instance, birds fly in sync by adjusting to neighboring birds. From a technical perspective, swarming is the emergent self-organization and aggregation among software agents with decentralized control. 11 A benefit of swarming is that one operator can be trained to manage a swarm of agents. This method is less computationally expensive and more reliable than training an operator for each agent. 12

Multiagent systems can solve many complex, real-world tasks. Some examples of applicable domains include:

Multiagent systems can be used to manage transportation systems. The qualities of multiagent systems that allow for the coordination of complex transportation systems are communication, collaboration, planning and real-time information access. Examples of distributed systems that might benefit from MAS are railroad systems, truck assignments and marine vessels visiting the same ports. 13

Multiagent systems can be used for various specific tasks in the healthcare field. These agent-based systems can aid in disease prediction and prevention through genetic analysis. Medical research about cancer might be one application. 14 In addition, multiagent systems can serve as tools for preventing and simulating epidemic spread. This forecasting is made possible by using epidemiologically informed neural networks and machine learning (ML) techniques to manage large datasets. These findings can affect public health and public policy. 15

Numerous factors affect a supply chain. These factors range from the creation of goods to the consumer purchase. Multiagent systems can use their vast informational resources, versatility and scalability to connect the components of supply chain management . To best navigate this intelligent automation , virtual agents should negotiate with one another. This negotiation is important for agents collaborating with other agents that have conflicting goals. 16

Multiagent systems can aid in strengthening defense systems. Potential threats can include both physical national security issues and cyberattacks. Multiagent systems can use their tools to simulate potential attacks. One example is a maritime attack simulation. This scenario would involve agents working in teams to capture the interactions between encroaching terrorist boats and defense vessels. 17  Also, by working in cooperative teams, agents can monitor different areas of the network to detect incoming threats such as distributed denial of service (DDoS)  flooding attacks. 18

There are several characteristics of multiagent systems that provide advantages including:

Flexibility

Multiagent systems can adjust to varying environments by adding, removing or adapting agents.

Scalability

The cooperation of several agents allows for a greater pool of shared information. This collaboration allows multiagent systems to solve more complex problems and tasks than single-agent systems.

Domain specialization

Single agent systems require one agent to perform tasks in various domains, whereas each agent in a multiagent system can hold specific domain expertise.

Greater performance

Multiagent frameworks tend to outperform singular agents. 19 This is because the more action plans are available to an agent, the more learning and reflection occur. An AI agent incorporating knowledge and feedback from other AI agents with specialties in related areas can be useful for information synthesis. This backend collaboration of AI agents and the ability to fill information gaps are unique to agentic frameworks, making them a powerful tool and a meaningful advancement in artificial intelligence.

There are several challenges in designing and implementing multiagent systems including:

Agent malfunctions

Multiagent systems built on the same  foundation models  can experience shared pitfalls. Such weaknesses might cause a system-wide failure of all involved agents or expose vulnerability to adverse attacks. 20  This highlights the importance of data governance in building foundation models and the need for thorough training and testing processes.

Coordination complexity

One of the greatest challenges with building multiagent systems is developing agents that can coordinate and negotiate with one another. This cooperation is essential for a functioning multiagent system.

Unpredictable behavior

The agents performing autonomously and independently in decentralized networks can experience conflicting or unpredictable behavior. Detecting and managing issues within the larger system might be difficult under these conditions.

Leverage IBM watsonx Assistant, an enterprise-grade AI chatbot, which is equipped with generative abilities to produce accurate, conversational answers. 

Increase developer productivity through the use of IBM watsonx Code Assistant powered by IBM Granite™ foundation models. Features include code generation, matching and modernization.

Lighten workload and improve productivity using generative AI assistants with IBM watsonx Orchestrate, powered by LLMs. 

Climate change is posing challenges for operating and designing critical infrastructure. Increasingly, AI has been used to enhance these decision making process.

An artificial intelligence (AI) agent refers to a system or program that is capable of autonomously performing tasks on behalf of a user or another system by designing its workflow and utilizing available tools.

Conversational artificial intelligence (AI) refers to technologies, such as chatbots or virtual agents, that users can talk to. They combine natural language processing (NLP) with machine learning to help imitate human interactions.

Train, validate, tune and deploy generative AI, foundation models and machine learning capabilities with IBM watsonx.ai™, a next-generation enterprise studio for AI builders. Build AI applications in a fraction of the time with a fraction of the data.

1  Edmund H. Durfee and Jeffrey S. Rosenschein, "Distributed problem solving and multi-agent systems: Comparisons and examples." In Proceedings of the Thirteenth International Distributed Artificial Intelligence Workshop , 1994,  https://aaai.org/papers/000-ws94-02-004/ (link resides outside ibm.com)

² David Kinny and Michael Georgeff, "Modelling and design of multi-agent systems," International Workshop on Agent Theories, Architectures, and Languages , 1996, https://link.springer.com/chapter/10.1007/BFb0013569  (link resides outside ibm.com)

³ Michael Wooldridge, An introduction to multiagent systems . John Wiley & Sons, 2009, https://dl.acm.org/doi/10.5555/1695886 (link resides outside ibm.com)

⁴ Peter Stone and Manuela Veloso, “Multiagent Systems: A Survey from a Machine Learning Perspective,” Autonomous Robotics, 2000, https://link.springer.com/article/10.1023/A:1008942012299  (link resides outside ibm.com)

⁵ Ming Tan, “Multi-Agent Reinforcement Learning: Independent versus Cooperative Agent,” Proceedings of the tenth international conference on machine learning, 1993, https://web.media.mit.edu/~cynthiab/Readings/tan-MAS-reinfLearn.pdf (link resides outside ibm.com)

⁶ Jianan Wang, Chunyan Wang, Ming Xin, Zhengtao Ding and Jiayuan Shan, Cooperative Control of Multi-Agent Systems: An Optimal and Robust Perspective , Academic Press, 2020, https://www.sciencedirect.com/book/9780128201183/cooperative-control-of-multi-agent-systems?via=ihub=  (link resides outside ibm.com)

⁷ Lucian Busoniu, Bart De Schutter and Robert Babuska, “Decentralized reinforcement learning control of a robotic manipulator,” Proceedings of the 9th International Conference on Control, Automation, Robotics and Vision , 2006,  https://ieeexplore.ieee.org/document/4150192  (link resides outside ibm.com)

⁸ Parasumanna Gokulan Balaji and Dipti Srinivasan, "An Introduction to Multi-Agent Systems,” Innovations in Multi-Agent Systems and Applications - 1 , 2010, https://link.springer.com/chapter/10.1007/978-3-642-14435-6_1  (link resides outside ibm.com)

⁹ Vincent Hilaire, Abder Koukam and Sebastian Rodriguez, "An adaptative agent architecture for holonic multi-agent systems," ACM Transactions on Autonomous and Adaptive Systems (TAAS) , 2008,  https://dl.acm.org/doi/10.1145/1342171.1342173  (link resides outside ibm.com)

¹⁰ Reza Olfati-Saber, “Flocking for Multi-Agent Dynamic Systems: Algorithms and Theory,” EEE Transactions on automatic control 51, no. 3, 2006, https://ieeexplore.ieee.org/document/1605401  (link resides outside ibm.com)

¹¹ H. Van Dyke Parunak and Sven A. Brueckner, "Engineering swarming systems," Methodologies and software engineering for agent systems , 2004, https://link.springer.com/chapter/10.1007/1-4020-8058-1_21  (link resides outside ibm.com)

¹² Ross Arnold, Kevin Carey, Benjamin Abruzzo and Christopher Korpela, "What is a robot swarm: a definition for swarming robotics," IEEE 10th annual ubiquitous computing, electronics & mobile communication conference (uemcon) , 2019, https://ieeexplore.ieee.org/document/8993024  (link resides outside ibm.com)

¹³ Hans Moonen, Multi-agent systems for transportation planning and coordination, 2009.

¹⁴ Elhadi Shakshuki and Malcolm Reid, “Multi-Agent System Applications in Healthcare: Current Technology and Future Roadmap,” Procedia Comput Sci, 2015, https://www.sciencedirect.com/science/article/pii/S1877050915008716?via%3Dihub  (link resides outside ibm.com)

¹⁵ Alexander Rodríguez, "AI & Multi-agent Systems for Data-centric Epidemic Forecasting," AAMAS , 2023, https://dl.acm.org/doi/10.5555/3545946.3599132  (link resides outside ibm.com)

¹⁶ Ksenija Mandic and Boris Delibašić, “Application Of Multi-Agent Systems In Supply Chain Management,” Management Journal of Sustainable Business and Management Solutions in Emerging Economies , 2012,  https://scindeks.ceon.rs/article.aspx?artid=0354-86351263075M (link resides outside ibm.com)

¹⁷ Thomas W. Lucas, Susan M. Sanchez, Lisa R. Sickinger, Felix Martinez and Jonathan W. Roginski, 2007 Winter Simulation Conference , 2007, https://ieeexplore.ieee.org/document/4419596  (link resides outside ibm.com)

¹⁸ Igor Kotenko, Multi-agent Modelling and Simulation of Cyber-Attacks and Cyber-Defense for Homeland Security, IEEE International Workshop on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications, 2007, https://ieeexplore.ieee.org/document/4488494 (link resides outside ibm.com)

¹⁹ Junyou Li, Qin Zhang, Yangbin Yu, Qiang Fu and Deheng Ye. "More agents is all you need."  arXiv preprint, 2024,  https://arxiv.org/abs/2402.05120   (link resides outside ibm.com)

²⁰ Alan Chan, Carson Ezell, Max Kaufmann, Kevin Wei, Lewis Hammond, Herbie Bradley, Emma Bluemke, Nitarshan Rajkumar, David Krueger, Noam Kolt, Lennart Heim and Markus Anderljung, “Visibility into AI Agents,” The 2024 ACM Conference on Fairness, Accountability, and Transparency, 2024,  https://arxiv.org/abs/2401.13138  (link resides outside ibm.com)

Grab your spot at the free arXiv Accessibility Forum

Help | Advanced Search

Computer Science > Artificial Intelligence

Title: creative problem solving in artificially intelligent agents: a survey and framework.

Abstract: Creative Problem Solving (CPS) is a sub-area within Artificial Intelligence (AI) that focuses on methods for solving off-nominal, or anomalous problems in autonomous systems. Despite many advancements in planning and learning, resolving novel problems or adapting existing knowledge to a new context, especially in cases where the environment may change in unpredictable ways post deployment, remains a limiting factor in the safe and useful integration of intelligent systems. The emergence of increasingly autonomous systems dictates the necessity for AI agents to deal with environmental uncertainty through creativity. To stimulate further research in CPS, we present a definition and a framework of CPS, which we adopt to categorize existing AI methods in this field. Our framework consists of four main components of a CPS problem, namely, 1) problem formulation, 2) knowledge representation, 3) method of knowledge manipulation, and 4) method of evaluation. We conclude our survey with open research questions, and suggested directions for the future.

Submission history

Access paper:.

• Other Formats

References & Citations

• Google Scholar
• Semantic Scholar

BibTeX formatted citation

Bibliographic and Citation Tools

Code, data and media associated with this article, recommenders and search tools.

• Institution

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .

• Data Science
• Data Analysis
• Data Visualization
• Machine Learning
• Deep Learning
• Computer Vision
• Artificial Intelligence
• AI ML DS Interview Series
• AI ML DS Projects series
• Data Engineering
• Web Scrapping

How does an agent formulate a problem?

In artificial intelligence (AI) and machine learning, an agent is an entity that perceives its environment, processes information and acts upon that environment to achieve specific goals. The process by which an agent formulates a problem is critical, as it lays the foundation for the agent’s decision-making and problem-solving capabilities.

This article explores the steps and considerations involved in problem formulation by an intelligent agent.

Table of Content

Understanding Problem Formulation

Example: problem formulation for a package delivery by an autonomous drone, step 1: define the initial state, step 2: define actions and transition model, step 3: define the goal state and objective function, importance of problem formulation, challenges in problem formulation.

Problem formulation is the process by which an agent defines the task it needs to solve. This involves specifying the initial state, goal state, actions, constraints, and the criteria for evaluating solutions. Effective problem formulation is crucial for the success of the agent in finding optimal or satisfactory solutions.

Steps in Problem Formulation

• Example: In a navigation problem, the initial state could be the agent’s starting location on a map.
• Example: For the navigation problem, the goal state is the destination location.
• Example: In a robot navigation scenario, actions could include moving forward, turning left, or turning right.
• Example: In a game, the transition model would include the rules that specify how the game state changes based on the player’s moves.
• Example: For a delivery drone, constraints might include battery life, weight capacity, and no-fly zones.
• Example: In route planning, the cost function could represent the distance traveled, time taken, or energy consumed.
• Example: For a puzzle-solving agent, success criteria could be the completion of the puzzle within the shortest time or the fewest moves.

We will demonstrate how to formulate the problem of package delivery by an autonomous drone, implementing the concepts in Python code. The drone needs to navigate from an initial location to a customer’s location while avoiding no-fly zones and managing its battery life.

The initial state includes the drone’s starting location and its battery level.

We create a Drone class with an initializer ( __init__ method) that sets the initial location, battery level, no-fly zones, and goal location.

The drone can take various actions such as taking off, landing, and moving in different directions. The transition model updates the drone’s state based on the action taken.

The takeoff , land , and move methods define how the drone’s state changes with each action. The transition_model method uses these actions to update the drone’s state.

The goal state is the customer’s location. The objective function evaluates the drone’s performance based on whether it reaches the goal and the remaining battery life.

The objective_function method returns a high score if the drone reaches the goal and otherwise returns the remaining battery level.

Complete Implementation

Now let’s put the problem formulation for a package delivery by an autonomous drone into practice:

We instantiate a Drone , execute a sequence of actions, and print the final location, battery level, and objective function score.

Effective problem formulation is essential because:

• Clarity : It provides a clear understanding of the problem, making it easier to devise a solution.
• Efficiency : Proper formulation can significantly reduce the computational resources required to solve the problem.
• Optimal Solutions : It helps in finding the most optimal or satisfactory solution by accurately defining the goals and constraints.
• Incomplete Information : The agent may not have access to all the necessary information about the environment.
• Dynamic Environments : The environment may change unpredictably, requiring the agent to adapt its problem formulation.
• Complex Constraints : Managing and incorporating complex constraints can be challenging.

A key step in artificial intelligence is problem formulation, which has a big influence on how well an agent completes its duties. An agent may efficiently traverse its environment and accomplish desired results by providing precise definitions for the starting state, actions, target state, restrictions, transition model, and objective function. By using a structured approach, the agent is guaranteed to be able to tackle complicated issues methodically and make well-informed judgments that result in effective and efficient solutions. The examples given show how issue formulation is used in a variety of contexts, underscoring its adaptability and significance in the area of artificial intelligence. Problem formulation techniques will continue to be essential to creating intelligent agents that can solve an ever-expanding array of problems as AI develops.

Please Login to comment...

Similar reads.

• AI-ML-DS With Python
• Data Science Blogathon 2024

Improve your Coding Skills with Practice

What kind of Experience do you want to share?

Artificial Intelligence

Control System

• Interview Q

Intelligent Agent

Problem-solving, adversarial search, knowledge represent, uncertain knowledge r., subsets of ai, artificial intelligence mcq, related tutorials.

• Send your Feedback to [email protected]

Help Others, Please Share

Learn Latest Tutorials

Transact-SQL

Reinforcement Learning

R Programming

React Native

Python Design Patterns

Python Pillow

Python Turtle

Preparation

Verbal Ability

Interview Questions

Company Questions

Trending Technologies

Cloud Computing

Data Science

Machine Learning

B.Tech / MCA

Data Structures

Operating System

Computer Network

Compiler Design

Computer Organization

Discrete Mathematics

Ethical Hacking

Computer Graphics

Software Engineering

Web Technology

Cyber Security

C Programming

Data Mining

Data Warehouse