Skip to main content
Digital Transformation

Typical Artificial Intelligence

By November 30, 2021No Comments

This chapter presents basic methodological assumptions of methods belonging to symbolic artificial intelligence. These methods have three core beliefs in common: a model representing an intelligent system can be defined explicitly.

  • knowledge in such a model is represented symbolically,
  • Mental/cognitive operations can be described as formal operations over symbolic expressions and structures which belong to a knowledge model.

 In symbolic AI, two subgroups of methods can be distinguished. Researchers define general (generic) models of knowledge representation and intelligent operations within the first subgroup. This subgroup includes cognitive stimulation and logic-based reasoning. In contrast, the definition of models related to specific application areas is the goal of research within the second group. Hence such models are based on representations of domain knowledge.

This subgroup includes rule-based knowledge representa[1]tion, structural knowledge representation, and an approach based on mathematical linguistics. These groups of methods are presented briefly in the following sections.

 

Cognitive Simulation

Cognitive stimulation is one of the earliest approaches to artificial intelligence. The approach was introduced by Newell and Simon, who used it to design their famous Logic Theorist and General Problem Solver (GPS) systems. The basic idea of cognitive simulation is to define heuristic algorithms3 to simulate human cognitive abilities, e.g., 

Thinking, problem-solving, object recognition, and learning.

During such a simulation, a sequence of elementary steps that are analogous to those made by a human being is performed by a computer. So, to design such algorithms, we try to discover elementary concepts and rules human beings use to solve generic problems. We introduce four fundamental concepts of cognitive stimulation: state space, problem-solving as searching state space, Means-Ends Analysis, and problem reduction.

 Let us start with the concept of state space. Its initial state represents the situation in which we start problem-solving. Let us consider the example of chess. The initial state represents the initial position of the pieces at the beginning of the game. Goal states (or the goal state, if the problem only has one solution) represent a problem when finding a solution. Thus, in chess, goal states represent all the situations in which we checkmate the opponent. The remaining (intermediate) states represent all possible situations on the way to solving the problem.

 Thus, in chess, they represent all situations that are allowed taking into account the game’s rules. A state space is a graph in which nodes correspond to states (initial, goal, and intermediate) and edges represent all allowable transitions from one state to another. Thus, for chess, a state-space can be defined in the following way. First, from the node of the initial state, we define transitions (edges) to nodes that correspond to situations after the first “white” move. Then, for each such intermediate state, we represent transitions (edges) to nodes that correspond to conditions after the first “black” move in response to a situation caused by the first “white” move, etc.

Problem-solving as searching a state space is the second concept of cognitive stimulation. The idea is straightforward: if we do not know how to formulate algo[1]rithmic rules of problem-solving, we can try to solve this problem with the method of “trial and error” (“generate and test,” “guess and check”). Of course, using this method in its “pure form” is not a good idea for, e.g., playing chess. However, sometimes this is the only method we can use in our everyday life. For example, we have forgotten the combination lock code of our suitcase, or we have promised our fiancée to make a delicious omelet in the evening, and we have lost the recipe. (How[1]ever, we have a lot of eggs in the refrigerator, so we can make some experiments.) Let us notice that such generation of potential solutions can be “blind” (we do not use codes that represent important dates) or can be limited to a certain subarea of the state space (we use important dates as codes).

What is more, we can have a certain measure, called here a heuristic function, which tells us how close we are to a satisfactory solution (e.g., the “tasting measure,” which tells us how close we are to a delicious omelet). Then, if our experimental omelet is almost delicious, we should modify the recent result only a little bit. A correct procedure to generate possible solutions should have three properties:

  1. It should be complete, i.e., it should be able to generate every potential solution.
  2. It should not generate a potential solution more than once.
  3. It should make use of information that restricts the state space to its subarea and measures/functions of the quality of the potential solutions generated.

 Suppose we have a function that assesses the quality of potential solutions. In that case, we can use a technique called Means-Ends Analysis, MEA, to control generating possible solutions. For example, if we are at a specific state, then we use such a function to determine the difference/distance between this state and a goal state. Then, we use a means (i.e., the means is usually an operator or a procedure), reducing this difference.

For example, suppose our experimental omelet is nearly as tasty as the ideal one, and the only difference is that it is not sweet enough. In that case, we should use a “sweetening operator” to reduce this difference. We use the MEA technique iteratively, starting from the initial state until we reach a goal state. Problem reduction is the last concept. The idea involves replacing a complex problem, which is difficult to solve at once, with a sequence of simpler subproblems.

For example, if getting a driving license is our problem, we should divide it into two subproblems: passing a theory test and passing a road (practical) test. Let us notice that for many situations, their reduction to a sequence of simpler sub-problems is necessary because different state spaces must be defined for these sub-problems. As a result, other quality functions and different operators must be determined.

Resource:

Introduction to Artificial Intelligence – Mariusz Flasiński

samir faraj

Experienced, motivated, and results-oriented Managing Business Consultant with more than +16 years of proven achievements in information technology and Business Management.Certified ITIL 2011 EXPERT, ITIL 4 ® MANAGING PROFESSIONAL, ITIL EXPERT. Certified TOGAF 9.1, COBIT 5 FOUNDATION, CLOUD COMPUTING, ISO9001, ISO27001, EXECUTIVE MANAGEMENT

Leave a Reply