AMBR is a dynamic emergent cognitive model built on the basis of the cognitive architecture DUAL. AMBR is an acronym for 'Associative Memory-Based Reasoning' ( Kokinov 1988, 1990, 1994, 1998) and has been conceived as a model with very broad scope. Much of the work on it is still in progress. The current version of the model is numbered AMBR3. Previous versions were AMBR1 ( Kokinov, 1994) and AMBR2 (Petrov, 1997). We fully recognize the fact that the model as it currently stands and is reported here is incomplete. Here and now AMBR3 is an integrated model of analogical access and mapping. We view this version only as an intermediate stage of a bigger project.
The author of AMBR is Boicho Kokinov and it is being developed at the Central and Eastern European Center for Cognitive Science at New Bulgarian University.
Since its initial conception ( Kokinov 1988) the AMBR model has advocated a set of ideas about human reasoning in general and analogy-making in particular. They have been distilled by Kokinov (1997) into the following three principles:
Integration. The reasoning process cannot be partitioned into a sequence of independent stages performed by specialized module-like components. Rather, there are subprocesses that run together and each of them is potentially influenced by the rest. Each computational mechanism is responsible not only to produce its immediate result but also to create appropriate guiding pressures for other mechanisms. That is why AMBR is designed as an integrated model based on a parallel emergent architecture.
Unification. Analogy is not a specific mode of reasoning. Rather, deduction, induction (generalization), and analogy are slightly different versions of the same uniform reasoning process. The same computational mechanisms are used in all cases--there is some sort of perceptual processing that builds internal representation of the problem being solved, there is some (sub)process that accesses relevant information from long-term memory, there is some (sub)process that tries to map the new problem to previous knowledge, etc. Deduction, induction, and analogy all fit into the same framework, the differences being in the outcome of the processing but not in the processing itself. Thus the term deduction applies to cases when the new problem happens to match with a general old schema, induction goes the other way around, and analogy applies when the two situations are at approximately equal level of abstraction. Conceptualized in this way deduction and induction are just two extremal (and hence very important) points on the analogy continuum. Therefore AMBR is designed as a general model of reasoning with emphasis on analogy-making.
Context-sensitivity. Human reasoning is context-sensitive. Its outcome depends not only on the task and long-term memory knowledge but also on the environmental setting, recent activities of the reasoner, etc. AMBR is designed with the explicit aim to reflect this context-sensitivity of human thinking.
This document focuses on the first point from this list--integration. Deduction, induction, context and priming effects are treated elsewhere (e.g. Kokinov, 1990, 1992, 1994; Kokinov & Yoveva, 1996). Hence it is warranted to explicate the principle of integration of subprocesses in more detail.
Theories of analogy-making frequently partition the process into a sequence of stages (e.g. Gentner, 1989). The computational models that stem from these theories typically involve separate 'engines' each of which works on its own and dovetails with the next. The output from the retrieval module is fed into the mapping module, whose output in turn is fed to the transfer module, etc. This 'pipeline paradigm' is illustrated in Figure 1. Each module influences the next only via the data structures it passes to it. Occasionally a module could detect a failure and loop back to some earlier module.
Figure 1. Schematic description of the 'pipeline paradigm' of analogy-making. The whole process is broken into a sequence of independent stages. They can interact only through the data structures (not shown in the figure) that each of them feeds to the next. Compare with Figure 2.
A problem with this approach is that it depends on the tacit assumption that all these stages are separable. While this assumption definitely merits careful consideration, it is questioned by a number of researchers (Chalmers, French & Hofstadter, 1992; Kokinov, 1994; Hummel & Holyoak, 1997). AMBR follows a different track. It adopts an interactionist approach and treats analog access, mapping, transfer, etc. as parallel subprocesses rather than serial stages. These subprocesses are still ordered in time as suggested by the pipeline approach--for instance the mapping begins after the retrieval has begun. However, there is no requirement that a stage must end before the next one could begin. On the contrary, subprocesses overlap considerably and can interact. This leads to the cascade illustrated in Figure 2.
Figure 2. Schematic description of the interactionist paradigm of analogy-making. There are subprocesses that overlap in time and can influence each other. Compare with Figure 1.
The interactionist approach seems problematic at first sight because each stage (or subprocess for that matter) depends on the result of the previous one. Indeed, how could the target problem be mapped to the source when it has not yet been even retrieved from memory?! It seems a logical necessity that the mapping comes after the retrieval. Similarly, the perceptual stage should come first, the transfer should follow the mapping, and so on.
AMBR overcomes this difficulty by representing information in smaller chunks. The model does not represent episodes as big units that are either manipulated wholesale or not at all. Instead, it represents them as coalitions of small elements susceptible of piecemeal manipulation. This allows each subprocess to begin as soon as the previous one has produced some partial results.
More concretely, as soon as the perceptual mechanisms have built internal representations of a few elements of the target problem, the access subprocess starts looking in the long-term memory for information that relates to these new elements. The concepts, propositions, episodes, etc. that are accessed in this way can now influence the perception of the target. In addition, they trigger the mapping subprocess which starts constructing the first tentative correspondences. If a promising candidate correspondence emerges, it could influence both perception and access. Gradually, all subprocesses are at work and more and more is perceived, accessed, mapped, transferred, and so forth.
This is the upward motion of the 'wave' of the reasoning process. Sooner or later the wave goes down. A stable representation of the target problem has been built and the perceptual mechanisms go off stage. A source episode wins the competition with alternative episodes from memory and the access subprocess diminish. One by one, all subprocesses terminate roughly in the order they have started. In this way there is something that could be characterized roughly as a sequence of stages. However, the boundaries between the AMBR 'stages' are fuzzy and each one could in principle interact with everyone else.
| Next page | > > > | AMBR Protagonists |
| [ DUAL Main Page ] | [ Back to top ] |
| [ Petrov's Home Page ][ Petrov's Publications ] | |
| [ Kokinov's Home Page ][ Kokinov's Publications ] |
Page maintained by
Alex Petrov