Next: Emotional factors Previous: The thesis Up: How To Do Research In the MIT AI Lab
This section is weak. Please contribute!
A research methodology defines what the activity of research is, how to proceed, how to measure progress, and what constitutes success. AI methodology is a jumbled mess. Different methodologies define distinct schools which wage religious wars against each other.
Methods are tools. Use them; don't let them use you. Don't fall for slogans that raise one above the others: ``AI research needs to be put on firm foundations;'' ``Philosophers just talk. AI is about hacking;'' ``You have to know what's computed before you ask how.'' To succeed at AI, you have to be good at technical methods and you have to be suspicious of them. For instance, you should be able to prove theorems and you should harbor doubts about whether theorems prove anything.
Most good pieces of AI delicately balance several methodologies. For example, you must walk a fine line between too much theory, possibly irrelevant to any real problem, and voluminous implementation, which can represent an incoherent munging of ad-hoc solutions. You are constantly faced with research decisions that divide along a boundary between ``neat'' and ``scruffy.'' Should you take the time to formalize this problem to some extent (so that, for example, you can prove its intractability), or should you deal with it in its raw form, which ill-defined but closer to reality? Taking the former approach leads (when successful) to a clear, certain result that will usually be either boring or at least will not Address the Issues; the latter approach runs the risk of turning into a bunch of hacks. Any one piece of work, and any one person, should aim for a judicious balance, formalizing subproblems that seem to cry for it while keeping honest to the Big Picture.
Some work is like science. You look at how people learn arithmetic, how the brain works, how kangaroos hop, and try to figure it out and make a testable theory. Some work is like engineering: you try to build a better problem solver or shape-from algorithm. Some work is like mathematics: you play with formalisms, try to understand their properties, hone them, prove things about them. Some work is example-driven, trying to explain specific phenomena. The best work combines all these and more.
Methodologies are social. Read how other people attacked similar problems, and talk to people about how they proceeded in specific cases.
A whole lot of people at MIT