Artificial Intelligence – a personal view Contributions Marr has proposed the following methodology for AI research o It is important to isolate simpler information processing problem of which tangible, efficient and correct solutions can be sought o It is important to come up with ‘abstract’ method which gives a direction to solve a particular information processing problem. A particular algorithm can be devised from this method later and it is important to separate the two. o Solving AI problems are of two types. Type 1 denotes those types of problems where a simple method can be devised to solve a information processing problem. Type 2 denotes those types of problems where a problem is solved by simultaneous interaction of more than one process whose interactions form a simplest description Questions I understand Marr is critical of AI. Why? o What do you think Minsky would think of Marr’s Paper? o Marr is not critical of AI, but has some reservations towards the approach taken towards solving the problems in AI. Based on his classification of Type I and Type II methodologies, he feels that many researchers have come across solutions that are solved by convoluted Type II methods, but for which alternate Type I solution could have existed. Also, he feels that many are guilty of not separating the abstract methods to solve a given problem and the algorithms with the gory details of implementation. Furthermore, he feels that current research on AI should focus on solving simpler problems for which we can provide clean, correct and efficient solutions rather than for complex problems like cryptarithmetic for which we don’t know how exactly the human brain decomposes this problem to solve it. He also touch upon the political issue of external pressure for showing early results impacts a good judgment and results in programs that just ‘mimic’ the solution but aren’t the actual solution by any means. Misnky will agree with Marr about his methodology of isolating problems, identifying methods and then implement algorithms. Minsky himself proposes to decompose complex problems into simpler solutions and then solving simpler problems and aggregating their results to get the final ‘goal’. However, he won’t agree to Marr that we need to attack only simpler problems. Minsky would feel that we can attack complex problems as far as we can decompose them into simpler manageable problems What do you think Marr would think of Minsky’s Paper, steps..? o I think Marr would appreciate the idea of Minsky; that or breaking a complex problem of achieving a goal into simpler ones. These simpler problems are in the essence, the Type I problems that Marr is referring to. Minsky is thinking in same lines with Marr here that we need to isolate simpler solvable problems which need not be of higher complexity levels. What do you think Marr would think of Brook’s work? o What do you think Marr would think about connectionism? o Marr has expressed that the perception of an event or of an object must include the simultaneous computation or several descriptions of it that capture diverse aspects of the use, purpose or circumstances of the event or object. The connectionist model is based on simultaneous computations that capture the perception of an object or event. The perception-motor cycle mentioned in the connectionist model fits well here. Again, I feel that Marr would be supportive of this idea. What do you think Marr would think Deep blue? o Brook’s work closely fits the Type II model where a complex functionality is achieved by simultaneous interactions of many processes (layers in case of brook). Again, the idea is to decompose complex behavior into simpler manageable sub-problems. Marr never seemed to have advocated for centralized managed systems with formal representation of general concepts or against them. Brook’s work also seems to go by thinking of a method and then an algorithm (FSA for this work) and have a subtle distinction between them. My feeling is that Marr will appreciate Brook’s work in that perspective I presume the question reflects about the Deep-Blue program. Marr has the idea of first selecting simpler problems and devising solutions for them. For complex problems like playing chess, we don’t know how human brain decomposes such problem. again, he feels that just providing solutions to such complex problems isn’t enough, as many times they just ‘mimic’ the actual solutions. I don’t think Marr will be too appreciative about Deep Blue. What do you think Chomsky would think about Marr’s Paper? o Chomsky will agree to Marr’s ideas of separating Methods and Gory detailed algorithms. Marr’s idea seems to be an extension to Chomsky’s work. Anticipated questions What is Marr’s Idea of Artificial Intelligence? What level of Intelligence he expects Machines to demonstrate? What would be Marr’s vision of achieving in the field of AI in next 5 years?