The flexible nature of our human perceptual system enables us to accomplish such tasks as dealing with unfamiliar situations, inventing novel solutions to old problems, and to even perceive objects as instruments that are far removed from their intended use (for example, perceiving a coat-hanger as a potential antenna for a television set). Analogy making, which requires the equating of non-identical situations is one of the clearest illustrations of this flexibility, and is said to lie deep in the heart of human creativity (Holyoak & Thagard, 1995). Analogy may be used for a wide range of purposes, from guiding reasoning, to formulating theories about new domains and extracting regularities from groups of experiences (Falkenhainer, Forbus & Gentner, 1989, p1). For these reasons, computational models of analogical thought are of great interest to both cognitive science (in which they can often provide useful insights into understanding the human mind), and artificial intelligence (in which they can be used to create more flexible learning and reasoning systems).
One such computational model of analogical thought is the Copycat model devised by Melanie Mitchell and Douglas Hofstadter. This model was conceived as part of a long-term research project aimed at gaining an understanding of the mechanisms that underlie the flexibility of high-level perception. Much of this flexibility is dependent on the fluid nature of mental concepts themselves. Unlike many symbolic systems, concepts in the human mind have fuzzy boundaries that overlap with each other, are dynamic and graded, and can change with the situation at hand. An example of the slippery nature of concepts are definitions for concepts such as boy and man; the concepts seem quite distinguishable from their definitions, but there are no hard separating boundaries by which this classification can be made (e.g. at what age does a father think of his son as finally becoming a man?).
According to Mitchell (1993, p2), the same cognitive mechanisms seem to underlie all aspects of thought, from simple recognition to complex feats of creativity. In particular, central to all of these mental activities is the phenomenon of conceptual slippage, in which, under the pressures of a given context, a concept can "slip into" (i.e. can be recognised as) another conceptually related concept. For example, in recognising messy handwriting, a letter in isolation may look like a different letter, but in the context of a word will be correctly identified. Such slippages are often quite covert, but can be embarrassingly apparent in many speech errors such as word substitutions (where for example, you may refer to the dishwasher as a washing machine due to the fact that it is a machine that washes things).
The Copycat model was designed as an attempt to study these phenomena in the isolated domain of letter-string analogy problems. As stated by Hofstadter "once we have some understanding of the way cognitive processes work in a restricted domain, we will have made genuine progress towards understanding the same phenomena in the unrestricted real world" (Hofstadter, 1995, p190).
Although the Copycat model is successful at mimicking human performance in this domain, and uses psychologically plausible mechanisms that can be adapted to larger, real world domains, the popularity of this model has been encumbered by its complexity. From simply reading the literature, it is difficult to acquire a good understanding of how the constituents of its architecture interact to form a solution. The aim of this tutorial is to facilitate this learning processes, by breaking down the model into its basic components, slowly revealing how these parts are integrated, and giving the user hands-on experience, allowing them to see the model (as well as the constituents in isolation) in action.
Exercise 1: Context-dependent conceptual slippage plays a vital role in
human perception. Find an example
domain (such as character recognition mentioned above – but don’t use this one,
try to be creative) where context-dependent conceptual slippage is required to
deal with the noise in the environment.
Give examples where this slippage takes place to illustrate your point.
