MIT Press 1995
Date finished: 2007-01-03
A knowledgable and opinionated tour of the landscape of artificial intelligence research circa 1995. I love books such as this, that survey a scientific field from a particular worker's point of view and discuss the author's positions on significant issues instead of remaining boringly neutral. Franklin discusses three open questions in artificial intelligence:
1. Is artificial intelligence even possible? 1. Can AI be produced by systems that manipulate symbols algorithmically, or by systems of many relatively simple and highly interconnected components? By both types of system? Neither type? 1. Do the internal structures of an AI have to represent aspects of the exterior world, or is a central representation, a world model, not necessary?
Simply understanding these questions is difficult, and how to go about answering them is far from obvious. To some degree the questions can only be answered positively by actually building an AI, and it may be impossible to prove the converse -- that it's impossible to build a mind, or to build a mind algorithmically, or to build one without a representation.
In pursuit of these questions, Franklin describes various papers and working systems that embody different positions in the debate: different models of mind such as Minsky's agent model, Selfridge's pandaemonium model, Maes's competence models, Brooks's subsumption architecture, and many more. Some of these systems are very neat: Copycat solves "what's next in this sequence" problems by analogy; Pollack's auto-associative memory translates sentences from active to passive form, yet is purely connectionist and has no internal concept of verbs, tenses, or nouns; Kanerva's sparse distributed RAM is a elegant mathematical approach to storing information.
Not knowing much about AI, I have no idea how well this book's ideas have stood up to the last 11 years, but I greatly enjoyed Franklin's opinionated tour.
%T Artificial Minds %A Franklin, Stan %K artificial intelligence, computing %@ 2007-01-03 %I MIT Press %D 1995 %G ISBN 0-262-06178-3 %P 436pp