An application of artificial intelligence in real world is decision making algorithms that are inspired from the human cognitive property to make judgements based on different situations. Decision making algorithms help individuals or organisations to take decisions and make choices, usually based on ranking prioritizing or choosing from many options. To take the decisions, decisions tables are used which are a concise visual representation with a complete detail on which action to perform on what situation. The information stored in these tables are stored in form of decision trees or as a snippet of a programming language with a series of if- then-else and switch-case statements. Modelling of Agent Cognition in Extensive Games via Artificial Neural Networks This paper focuses on developing a framework that will have an artificial neural network which will try implement the decision-making algorithm. The aim is to develop an agent with self-cognition which will play the famous Go game with a moderately experienced human player of that game. An algorithm was introduced that had a series of continued conditional processes which used a training data of previous games and find the best possible move to make that had a successful result in those games. In the experiment done, a player was asked to play Go game, with an artificially developed agent. Go is a strategic board game for two players, in which the aim is to surround more territory than the opponent. Each move that was supposed to be taken out very carefully so that the opponent would not have any chance to surround the player. Now, to make the decisions the artificial agent was provided with a backtracking algorithm as he decision making algorithm in which the system must look back in the past moves and would draw the new moves as per the success rate of the previous moves. Every time the plays makes his move, the agent would backtrack the game, select a matching situation, see the result of the decision taken at that point, and would commence the same if it was successful and would take a risk to play a new move if the last move with same situation was a failure. The human player on the other hand was not capable of such processing in such a short time also to be kept in knowledge that the player was a moderately experienced player in that game. As a result, the artificial agent with artificially developed decision-making process won the game that suggests that there exist alternative ways to implement an artificial neural network that is capable to make judgements. The question that was left unanswered was that the backtracking algorithm used to make decisions didn’t had any risk analysis function in it that would have managed the risk on making a new move by checking the probability of winning of that move in other situations. If that would have been kept in check, the results could have been better.
Source: Chanjuan Liu, Enqiang Zhu, et al (2017). Modeling of Agent Cognition in Extensive Games via Artificial Neural Networks. In proceeding of the 2017 IEEE 12th International Conference on Neural Networks and Learning Systems, 2278-2285