finite state machines are believed to be the building blocks of life. They are used that are able to learn from prior experience. Artificial neural networks are based on such simple FSM's and are going to help computers learn from both human and machine behavior. IEEE MIPS software has been used in conjunction with a game AI. An artificial neural network was capable of learning the moves of the game from a small number of games, and was used to construct a classifier that could distinguish between human and machine play. A computer system based on neural networks could go on to train itself without special assistance. It could compare the values from its previous experience against a new experience and conclude, so says the IEEE MIPS article, that a given situation is similar to one that is familiar. It could then insert the more common response into its code. In this way the machine, short of being told, would learn its own best play.
Furthermore, the author suggests using computer vision engines may lead to the creation of smart systems that may be able to run AI based software on their own. The report used an image recognition software package to detect a set of four cards that were used to play poker. This report showed that AI engines were able to recognize that these were cards that might be used in a game of poker. The report also suggested that the AI-based software was able to recognize when it had been dealt a set of given cards from a deck and recognize the appropriate play. An early example of such a system would be a chess-playing computer. The same AI engine was used to recognize chess moves that could be made. The same system could then develop, by itself, its own optimal play for the game.
Neural network based programs are not new. For example, the MIT computer vision group used them for the players described in the paper. However, this group only applied neural networks to localization, while the paper suggested that much more was possible. A person walking in front of a pair of still image cameras could be identified, even though that person was seen from certain distances and under varying lighting conditions. What is more one person even in motion could be recognized. The AI could conclude which person was moving about in a room full of people, the table of the room may be inserted into the software. d2c66b5586