next up previous contents
Next: Techniques in Qualitative Reasoning Up: The Origins of Qualitative Previous: Early Qualitative Reasoning Systems

Naive Physics

``Naive Physics'' is a term coined by Hayes [Hay78] in 1978, to describe his approach to developing a ``large-scale formalism'' of commonsense knowledge about the world. This concern with real world knowledge can be related to a general awareness amongst AI workers that future progress in AI depends on intensive knowledge being made available to reasoning systems.3.3

The aim of naive physics as stated in [Hay83] is to formally describe the world in the way that most people think about it, rather than describing it in the way that physicists think about it. This description should attempt reasonable completeness - that is, it should describe a significant portion of the way we understand the world, rather than just small pieces.

The use of the word ``naive'' indicates that this description must include commonsense knowledge that is normally taken for granted in formal physics, and it therefore may include elements outside what we consider to be the field of physics. It may also choose to describe phenomena in a way that is familiar to ``the man in the street'', but would not be considered appropriate to a physicist. Two examples are the ``force'' of sucking, and ``impetus'' theories of motion, both of which adequately describe everyday phenomena, and are prevalent theories amongst intelligent people [McC83], even though they are considered to be inappropriate for physicists.

The ``Naive Physics Manifesto'', and the its revision in the ``Second Naive Physics Manifesto'', proposed the construction of a formalisation of commonsense knowledge which covered a broad range of knowledge using a common framework, and included dense factual detail. It specifically did not recommend the construction of programs or new formal description methods. It did dwell on the importance of spatial representation, although it excluded the sort of spatial reasoning that is necessary to plan physical motion3.4.

Nearly all qualitative physics literature has cited the ``Naive Physics Manifesto'' (including Forbus [For81], Faltings [Fal87], Kuipers [Kui82], Stanfill [Sta83a], etc.), and it can therefore be considered to be a foundational work for qualitative physics.

The wide influence of the manifesto can be easily accounted for - it is compelling reading, pointing out clearly the deficiencies of much work in A.I., while proposing a clear, exciting, and apparently practical vision for future progress. However, most qualitative physics work does not follow Hayes' recommendations. In particular, an emphasis on restricted domains, and on construction of reasoning programs rather than on knowledge representation, separates qualitative physics from naive physics as proposed by Hayes.

In addition, the importance of spatial reasoning in the commonsense world as pointed out by Hayes has not resulted in an emphasis on spatial problem solving in qualitative reasoning. On the contrary, most work concentrates on problem domains which can be abstractly represented without spatial information. This thesis addresses some issues in spatial representation which are obviously related to commonsense spatial reasoning ability, and they may assist progress toward some of the goals that Hayes has set.

next up previous contents
Next: Techniques in Qualitative Reasoning Up: The Origins of Qualitative Previous: Early Qualitative Reasoning Systems
Alan Blackwell