page 1  (15 pages)
2to next section

, ,

Abstract

Stuart Watt

S.N.K.Watt@open.ac.uk

The Naive Psychology Manifesto

1 Introduction: from naive physics to naive psychology

Human Cognition Research Laboratory
The Open University Milton Keynes MK7 6AA
Email:

This paper argues that artificial intelligence has failed to address the whole problem of common sense, and that this is the cause of a recent stagnation in the field. The big gap is in common sense?or naive?psychology, our natural human ability to see one another as minds rather than as bodies. This is especially important to artificial intelligence because AI must eventually enable us humans to see computers not as grey boxes, but as minds.

The paper proposes that AI study exactly this?what is going on in people's heads that makes them see others as having minds. To illustrate this, it describes three models for common sense psychology; each of which illustrates some aspect of the human ability to see people as minds. The paper concludes by drawing some conclusions about where and how AI can adapt to deal with these issues and to provide a new and stronger foundation for future research.

Ten years ago, Hayes published the ?Second naive physics manifesto? [24]. Hayes proposed that we ?put away childish things by building large-scale formalisations,? beginning with ?our knowledge of the everyday physical world.? He, and others, have since put a lot of effort into developing models of our common sense understanding of the physical world.

But life in the common sense community isn't a bed of roses. Common sense has generally been a big problem for artificial intelligence, and despite the attempts of many brave souls (e.g. Hayes, McCarthy, McDermott, and Lenat) it hasn't really yielded at all: ?the common sense knowledge problem has blocked all progress in theoretical artificial intelligence? [17]. For twenty years there has been only slow progress in the field, and although many important technologies have been found while looking for ways to deal with common sense (e.g. nonmonotonic reasoning,) the core of the problem is still almost untouched.

The problems for common sense have generally come in two classes, technical and methodological. On the technical side, for example, there is the `frame problem' [32], which Dennett [15] describes as the ?smoking pistol? behind a lot of the