Next: Naive Physics Up: The Origins of Qualitative Previous: The Origins of Qualitative

## Early Qualitative Reasoning Systems

The first (1975) published paper which is referred to in qualitative physics literature was de Kleer [dK79]3.1. The system described in this paper was intended to model human problem solving in Newtonian mechanics. Several people had previously worked on systems which could solve problems at the level of ``high school physics'', but de Kleer was the first to use qualitative methods in an attempt to model the thought process of a student more closely - the others simply extracted numbers from a problem description, and used linguistic cues to select the mathematical operations which might be appropriate.

De Kleer's system, called NEWTON, described motion of a block on a roller coaster in terms of the speed of the block and the slope of the roller coaster. The speed could have one of three qualitative values - positive, negative, or zero. The slope of the roller coaster was likewise described by one of three qualitative slopes - up, down, or level. NEWTON performed an initial qualitative analysis to find places at which the behaviour of the block was likely to be interesting, and later applied numerical techniques if operations using the qualitative values were not sufficient to reach a conclusion about behaviour at particular locations. The use of multiple representations to support both qualitative and numeric reasoning was the major emphasis of the research.

In 1979 de Kleer extended qualitative techniques to the domain of electronic circuits3.2. The techniques used for reasoning about circuits retained NEWTON's use of three qualitative values. An emphasis of the work was establishing causal relationships between values at different nodes of the circuit. The use of the circuit domain removed any need for spatial representation in this research, since a circuit can be completely represented by its topology, without any spatial information.

In 1980, de Kleer and Brown described a system which modelled human understanding of mechanical devices [dKB80], [dKB83]. The input to the system was a structural description of the device. From this information the system would create an ``envisionment'' of its likely behaviour. The example used was an electro-mechanical buzzer. The analysis of the buzzer proceeded by analysing how some components could cause changes in the state of others, thus changing the state of the whole system. The oscillation of the buzzer could be deduced from the closed loop in the state transition diagram.

In 1981, Forbus published a description of a system which reasoned about the motion of a bouncing ball through a two dimensional scene consisting of free space bounded by vertical and horizontal surfaces [For81]. ``FROB'' could, given a description of the initial trajectory of the bouncing ball, predict where it might come to rest. The main tools used by the system were the ``metric diagram'' and the ``place vocabulary''. The ``metric diagram'' was a representation of the scene that was intended to model the human capability to reason using an internal image of a situation [For83]. The ``place vocabulary'' was a subdivision of free space into regions with different qualitative features. Motion was then described as a transition between regions. This part was purely qualitative, and included no numerical analysis components.

Forbus developed a new technique for qualitative reasoning in 1981, which he called ``qualitative process theory'' [For84a]. Qualitative process theory describes physical quantities such as temperature and pressure in terms of a ``quantity space'', which is an ordering of values that have qualitative significance for the process being described (for example - boiling point is a qualitatively significant value for temperature when a water heating process is being described). The process is described as a set of ``influences'' on quantities associated with objects in the situation described. For example, heating might be described an influence exerted by a heater on the temperature quantity associated with a fluid.

A large body of research has developed from Forbus' qualitative process theory, and from de Kleer's work on causal reasoning about circuits. A comprehensive collection of this work is the book ``Qualitative Reasoning about Physical Systems'', published in 1984 [Bob84a]. Since this time, the field has continued to grow, and I will mention later work only if it has a bearing on spatial or mechanical reasoning.

Next: Naive Physics Up: The Origins of Qualitative Previous: The Origins of Qualitative
Alan Blackwell
2000-11-17