Aaron Sloman

Despite AI’s enormous practical successes, some researchers focus on its potential as science and philosophy: providing answers to ancient questions about what minds are, how they work, how multiple varieties of minds can be produced by biological evolution, including minds at different stages of evolution, and different stages of development in individual organisms. AI cannot yet replicate or faithfully model most of these, including ancient, but still widely used, mathematical discoveries described by Kant as non-empirical, non-logical and non-contingent. Automated geometric theorem provers start from externally provided logical axioms, whereas for ancient mathematicians the axioms in Euclid’s Elements were major discoveries, not arbitrary starting points. Human toddlers and other animals spontaneously make similar but simpler topological and geometrical discoveries, and use them in forming intentions and planning or controlling actions. The ancient mathematical discoveries were not results of statistical/probabilistic learning, because, as noted by Kant, they provide non-empirical knowledge of possibilities, impossibilities and necessary connections. Can gaps between natural and artificial reasoning in topology and geometry be bridged if future AI systems use previously unknown forms of information processing machinery – perhaps “Super-Turing Multi-Membrane” machinery?

This is a 5 page summary with three diagrams of the main objectives and some work in progress at the University of Birmingham Cognition and Affect project. involving: Professor Glyn Humphreys (School of Psychology), and Luc Beaudoin, Chris Paterson, Tim Read, Edmund Shing, Ian Wright, Ahmed El-Shafei, and (from October 1994) Chris Complin (research students). The project is concerned with "global" design requirements for coping simultaneously with coexisting but possibly unrelated goals, desires, preferences, intentions, and other kinds of motivators, all at different stages of processing. Our work builds on and extends seminal ideas of H.A.Simon (1967). We are exploring "broad and shallow" architectures combining varied capabilities most of which are not implemented in great depth. The poster summarises some ideas about management and meta-management processes, attention filtering, and the relevance to emotional states involved "perturbances", where there is partial loss of control of attention.

Replication or even modelling of consciousness in machines requires some clarifications and refinements of our concept of consciousness. Design of, construction of, and interaction with artificial systems can itself assist in this conceptual development. We start with the tentative hypothesis that although the word “consciousness” has no well-defined meaning, it is used to refer to aspects of human and animal informationprocessing. We then argue that we can enhance our understanding of what these aspects might be by designing and building virtual-machine architectures capturing various features of consciousness. This activity may in turn nurture the development of our concepts of consciousness, showing how an analysis based on information-processing virtual machines answers old philosophical puzzles as well enriching empirical theories. This process of developing and testing ideas by developing and testing designs leads to gradual refinement of many of our pre-theoretical concepts of mind, showing how they can be construed as implicitly “architecture-based” concepts. Understanding how humanlike robots with appropriate architectures are likely to feel puzzled about qualia may help us resolve those puzzles. The concept of “qualia” turns out to be an “architecture-based” concept, while individual qualia concepts are “architecture-driven”

What we have learnt in the last six or seven decades about virtual machinery, as a result of a great deal of science and technology, enables us to offer Darwin a new defence against critics who argued that only physical form, not mental capabilities and consciousness could be products of evolution by natural selection. The defence compares the mental phenomena mentioned by Darwin’s opponents with contents of virtual machinery in computing systems. Objects, states, events, and processes in virtual machinery which we have only recently learnt how to design and build, and could not even have been thought about in Darwin’s time, can interact with the physical machinery in which they are implemented, without being identical with their physical implementation, nor mere aggregates of physical structures and processes. The existence of various kinds of virtual machinery (including both “platform” virtual machines that can host other virtual machines, e.g. operating systems, and “application” virtual machines, e.g. spelling checkers, and computer games) depends on complex webs of causal connections involving hardware and software structures, events and processes, where the specification of such causal webs requires concepts that cannot be defined in terms of concepts of the physical sciences. That indefinability, plus the possibility of various kinds of self-monitoring within virtual machinery, seems to explain some of the allegedly mysterious and irreducible features of consciousness that motivated Darwin’s critics and also more recent philosophers criticising AI. There are consequences for philosophy, psychology, neuroscience and robotics.

A.1. Some philosophers, including Tarski and Russell, have concluded from a study of various versions of the Liar Paradox ‘that there must be a hierarchy of languages, and that the words “true” and “false”, as applied to statements in any given language, are themselves words belonging to a language of higher order’. In his famous essay on truth Tarski claimed that ‘colloquial’ language is inconsistent as a result of its property of ‘universality’: that is, whatever can be said at all can in principle be said in it, with an extended vocabularly if necessary. Thus, in English we can talk about English expressions, what they denote, what they say, whether what they say is true or false, and so on: English contains its own metalanguage. This universality enables us to construct sentences which say of themselves that they are false, and by applying the law of excluded middle to them we easily derive a contradiction. Tarski concludes that ‘these antinomies seem to provide a proof that every language which is universal in the above sense, and for which the normal laws of logic hold, must be inconsistent’ . He then proposes to avoid such contradictions by the use of a hierarchy of languages such that statements about any one language can be made only in a different language at a higher level

The aim of the thesis is to show that there are some synthetic necessary truths, or that synthetic apriori knowledge is possible. This is really a pretext for an investigation into the general connection between meaning and truth, or between understanding and knowing, which, as pointed out in the preface, is really the first stage in a more general enquiry concerning meaning. (Not all kinds of meaning are concerned with truth.) After the preliminaries (chapter one), in which the problem is stated and some methodological remarks made, the investigation proceeds in two stages. First there is a detailed inquiry into the manner in which the meanings or functions of words occurring in a statement help to determine the conditions in which that statement would be true (or false). This prepares the way for the second stage, which is an inquiry concerning the connection between meaning and necessary truth (between understanding and knowing apriori). The first stage occupies Part Two of the thesis, the second stage Part Three. In all this, only a restricted class of statements is discussed, namely those which contain nothing but logical words and descriptive words, such as "Not all round tables are scarlet" and "Every three-sided figure is three-angled". (The reasons for not discussing proper names and other singular definite referring expressions are given in Appendix I.)

This paper is an attempt to summarise and justify critical comments I have been making over several decades about research on consciousness by philosophers, scientists and engineers. This includes (a) explaining why the concept of "phenomenal consciousness" (P-C), in the sense defined by Ned Block, is semantically flawed and unsuitable as a target for scientific research or machine modelling, whereas something like the concept of "access consciousness" (A-C) with which it is often contrasted refers to phenomena that can be described and explained within a future scientific theory, and (b) explaining why the "hard problem" is a bogus problem, because of its dependence on the P-C concept. It is compared with another bogus problem, "the 'hard' problem of spatial identity" introduced as part of a tutorial on semantically flawed concepts. Different types of semantic flaw and conceptual confusion not normally studied outside analytical philosophy are distinguished. The semantic flaws of the "zombie" argument, closely allied with the P-C concept are also explained. These topics are related both to the evolution of human and animal minds and brains and to requirements for human-like robots. The diversity of the phenomena related to the concept "consciousness" as ordinarily used makes it a polymorphic concept, partly analogous to concepts like "efficient", "sensitive", and "impediment" all of which need extra information to be provided before they can be applied to anything, and then the criteria of applicability differ. As a result there cannot be one explanation of consciousness, one set of neural associates of consciousness, one explanation for the evolution of consciousness, nor one machine model of consciousness. We need many of each. I present a way of making progress based on what McCarthy called "the designer stance", using facts about running virtual machines, without which current computers obviously could not work. I suggest the same is true of biological minds, because biological evolution long ago "discovered" a need for something like virtual machinery for self-monitoring and self-extending information-processing systems, and produced far more sophisticated versions than human engineers have so far achieved.

This paper outlines a design-based methodology for the study of mind as a part of the broad discipline of Artificial Intelligence. Within that framework some architectural requirements for human-like minds are discussed, and some preliminary suggestions made regarding mechanisms underlying motivation, emotions, and personality. A brief description is given of the `Nursemaid' or `Minder' scenario being used at the University of Birmingham as a framework for research on these problems. It may be possible later to combine some of these ideas with work on synthetic agents inhabiting virtual reality environments.

When scientists discuss experimental observations, they often, unfortunately, use language that evolved for informal discourse among people engaged in every day social interaction, like this: What does the infant/child/adult/chimp/crow (etc) perceive/understand/learn/intend (etc)? What is he/she/it conscious of? What does he/she/it experience/enjoy/desire? What is he/she/it attending to? Similar comments can be made about the terminology used in many philosophical discussions about minds, cognition, language, and the relationships between evolution and learning.

What is the relation between intelligence and computation? Although the difficulty of defining `intelligence' is widely recognized, many are unaware that it is hard to give a satisfactory definition of `computational' if computation is supposed to provide a non-circular explanation for intelligent abilities. The only well-defined notion of `computation' is what can be generated by a Turing machine or a formally equivalent mechanism. This is not adequate for the key role in explaining the nature of mental processes, because it is too general, as many computations involve nothing mental, nor even processes: they are simply abstract structures. We need to combine the notion of `computation' with that of `machine'. This may still be too restrictive, if some non-computational mechanisms prove to be useful for intelligence. We need a theory-based taxonomy of {\em architectures} and {\em mechanisms} and corresponding process types. Computational machines my turn out to be a sub-class of the machines available for implementing intelligent agents. The more general analysis starts with the notion of a system with independently variable, causally interacting sub-states that have different causal roles, including both `belief-like' and `desire-like' sub-states, and many others. There are many significantly different such architectures. For certain architectures (including simple computers), some sub-states have a semantic interpretation for the system. The relevant concept of semantics is defined partly in terms of a kind of Tarski-like structural correspondence (not to be confused with isomorphism). This always leaves some semantic indeterminacy, which can be reduced by causal loops involving the environment. But the causal links are complex, can share causal pathways, and always leave mental states to some extent semantically indeterminate.