Abstract for AISB 2000: How to Design a Functioning Mind

Abstract for the
Symposium on How to Design a Functioning Mind
17-18th April 2000
At the AISB'00 Convention

AUTHOR: Marcin Chady School of Computer Science, The University of Birmingham

TITLE: Can we model the mind with its own products?


The human brain has evolved to model the world, so that we can avoid dangers and achieve our goals. It is not possible to simulate the universe exactly, so some (arguably significant) degree of approximation is required. In the complex world, in order to draw conclusions and predict future events, one must generalise. Consequently, any brain-endowed organism is a natural classifier, and we humans are a prime example of this: we categorise, generalise and organise things into hierarchies, so that we can make better sense of the world, and turn its rules into our advantage. We have developed languages, algebras and calculi. Our sophisticated system of symbols can describe abstract as well as physical things. However, as we face the enigma of our own minds, we must ask ourselves the question: Can we model the mind with its own products?

When we talk about thinking, we tend to use inherently ambiguous terms: emotions, intentions, exploration, control, or even memory. It isn't until we actually try to realise them in a physical system, that it transpires how vague they are. Many of them have been discredited from scientific discourse, and confined to the realms of folk psychology, but much of the confusion remains (cf. Smolensky 1988}. People like to view the mind as a number of parallel processes: transforming information, exchanging signals, making decisions, each specialised in a different kind of operations. It is easy to conceptualise such systems using the "divide and conquer" strategy, and, perhaps most importantly, it is easy to depict them using boxes and arrows. Even much of the non-symbolic AI, such as neural networks, inadvertently follow this path by giving parts of their systems explicitly defined functionality and combining them into interacting ensembles of modules.

It is hard not to do it this way, given that this is what our brains have been designed to do through millenia of evolution. However, one must not forget that everything our cognitive processes come up with is an artefact. It's a product of a process whose purpose is by no means general. On the contrary, its purpose is to filter out details which are irrelevant to survival, and produce a convenient model of the environment. There is no reason why our cognitive apparatus should be able to cope with its own workings, just like no one is predisposed to imagine 7-dimensional objects, which some physicists suggest our universe really consists of.

Can we escape the kaleidoscope of our cognition? Most probably not. Even the machines we create will be biased. However, we can reduce this bias by limiting our involvement in their design. Evolutionary strategies spring to mind. But we can also study complex unconventional systems and look for new insights there. Of course, every insight will be subjective too, but there is nothing we can do about it. At least we may discover a new way of thinking which will take us a further step back from the narrow vision of a "survival machine".

An example of such approach can be found in the work of Jim Hanson and Jim Crutchfield (1922). There, a complex behaviour of CA is analysed using classic paradigms from computational theory, such as Finite State Automata. The work demonstrates how complicated operation can result from simple interactions between simple units. Conversely, the use of classical computational models to describe the global behaviour shows that no explicit FSA machinery is necessary to produce FSA behaviour.

Perhaps then, there are no parts responsible for motivation, emotions, control, etc. They could be just facets of the same process, presenting themselves differently from different angles. Much like the facets of a hypercube intersected with a 3-dimensional hyperplane, or like ephemeral vortices of turbulent flow.



* BSc (Hons) in Software Engineering from Sheffield Hallam University, 1994

* MSc in Advanced Computer Science from Birmingham University, 1997

* MSc (5 year course equivalent) in Software Engineering from Technical University of Wroclaw, Poland, 1997

* Currently a research student at the School of Computer Science of the University of Birmingham. Thesis topic: Modelling higher cognitive functions with Hebbian cell assemblies

* 1 year developing software at British Telecom Research Laboratories.

* Translating English part time for a software magazine

* Part time English teaching

* Currently a teaching assistant at the SoCS, teaching Java to MSc students

* Evolution of Cellular-automaton-based Associative Memories. Co-authored with Riccardo Poli. 1997. Second On-line World Conference on Soft Computing (WSC2)

* Local Connections in a Neural Network Improve Pattern Completion. Co-authored with Rafal Bogacz. 1999. Third International Conference on Cognitive and Neural Systems

* Modelling Higher Cognitive Functions with Hebbian Cell Assemblies. 1999. Thesis summary presented at AAAI-99 Doctoral Consortium and at EmerNet: International Workshop on Emergent Neural Computational Architectures Based on Neuroscience