“Artificial Intelligence – a personal view” by D. Marr. Contribution: 1) Marr introduces “Method” and “Result”. Method is a term used to address implementation useful to solve information-processing problems. A method has to be abstract and it is equivalent to a theorem in Mathematic. Result consists of solution to particular information processing problem and statement of method on resolving it. Method is usually implemented through some algorithms and it is important to note that once a method is implemented it should never be re-done again. 2) Marr stressed on understanding information-processing problems before relating an algorithm to resolve the problem. The overall structure of the problem is the main concerned and not the way of implementing the method. 3) Marr categorized information-processing problems into two categories. Category one is called Type-1 theory and category two is called Type-2 theory. Type-1 theory is when someone already knew the detail about the problem and he or she could apply some algorithm to resolve it. In this case a method that uses algorithm for implementation. Type-2 theory is more like prediction to solve the problem. Simultaneous actions are executed by several processes to achieve the final result. Questions: 1) I understand Marr is critical of AI research. Why? I personally don’t think that Marr is critical of AI research. He indeed has some brilliant contributions but I don’t think that his ideas in this paper would have made any significant improvement in AI. However I like his idea of categorizing problems into Type-1 and Type-2 theories. Type-1 theory is always the best if you understand the problem and have a good algorithm that you could use. Type-2 theory to me is like try and error until you achieve the satisfying result. Marr mentioned that many problems were implemented using Type-2 theory, which they could actually be implemented using Type-1 theory if one puts more effort in understanding the problem. He brought out a very good point when come to create an Artificial Intelligent. One should always starts resolving simple problem before getting into complex problem like creating human brain functions. 2) What do you think Minsky would think of Marr’s paper? First of all, Minsky and Marr contributed differently in Artificial Intelligent. Minsky concentrated more on improving heuristic programming. He also pointed out areas that needed for improvement. Improvements are in areas like Search, Pattern-Recognition, Learning, Planning, and Induction. Marr is interested in categorizing problems into two types. I think if Minsky would read Marr’s paper, I think he will agreed with Marr in attacking simple problem before trying to attack complex problem. 3) What do you think Marr would think of Minsky paper? Steps… Since Marr interested more in understanding problem rather than method implementation, I think Marr would agreed with Minsky in identifying areas where improvement could be done. Minsky introduced improvement in Search, Pattern-Recognition, Learning, Planning, and Induction after thorough research and understanding on why those areas should be improved for heuristic programming. 4) What do you think Marr would think of Brook’s work? I think Marr would agree on Brook’s idea of steps development. Brook’s robot is built based on the idea of information processing at different stages. Each does not need to know how the other stage implement or process information. 5) What do you think Marr would think of connectionism? I think Marr would treat connectionism as Type-2 theorem. Connectionism is about processes connected in some point and creating some actions or purposes. Similar to Type-2 theorem where simultaneous actions are executed by several processes to achieve the final result. 6) What do you think Marr would think deep blue? I am assuming it is asking about the chess “Deep Blue” program. I personally think that Marr might not be very interested in “Deep Blue” program. As you know, Marr is big in solving simple problems so complex problem like “Deep blue: might not be that important to him at that time. He is also not a big fan on knowing how some algorithm would be implemented to resolve problem. Thus he will not be interested in the logic of the Deep Blue program too. 7) What do you think Chomsky would think of Marr’s paper? I think Chomsky would agree on separating Method from detail of implementing the method. Anticipated questions in class: 1) How much do you agree with Marr about categorizing theory to Type-1 and Type-2? Would there be a type 3 theory? 2) If you were to choose one of the most important contributions of Marr in AI, what would it be? Why?