Iterative Dialectic Engine for Automated Learning
IDEAL is the vital 'missing link' in the great intellectual challenge of our age: the quest for artificial intelligence.
This ambitious claim is supported by the following considerations:
Artificial intelligence involves the computer simulation of human intelligence;
Computers and humans have different strengths and weaknesses;
By focusing too closely on human attributes, AI research has been looking in the wrong place;
The 'missing link' for AI is cybernetics - the science of control and communication;
By completing the mission of cybernetics, IDEAL constitutes the AI 'missing link'.
artificial intelligence (or machine intelligence). A new, computer-based science aimed partly at understanding the nature of human and animal intelligence and specifically at creating machines capable of intelligent problem-solving, by any means open to the logician and the engineer. (1)
For instance, the following key strengths and weaknesses are well-recognised:
|Key strength||Excels at the large-scale repetition of mundane computational operations.||Excels at learning from experience across a wide range of applications.|
|Key weakness||Incapable of learning by itself, except for specialised applications when programmed accordingly.||Incapable of the large-scale repetition of mundane computational operations.|
The phrase 'artificial intelligence' or 'AI' has come to mean many things, including:
The design and operation of robots and their components;
The design and operation of artificial neural networks;
The design and operation of specialised problem-solvers, e.g. games programs;
The application of specialised computing techniques, e.g. expert systems;
A science-fictional future world in which humans and computers are virtually indistinguishable, thereby fulfilling the 'Turing Test'. In practice, if computers were indistinguishable from humans then there would be no point including them as fictional characters. Accordingly, typical plotlines turn on a key distinction between the species - an 'Achilles heel' for one or the other (e.g. 2001 - A Space Odyssey, Wargames, Blade Runner, Terminator, The Matrix).
Vast resources have been expended in the perfection of computer simulations of specific human attributes. Nevertheless, it remains the case that (as some unknown wit has remarked) "artificial intelligence is no match for natural stupidity". It is tempting to jump to the conclusion that there must be a 'missing link' - a vital component somehow overlooked by every major research programme - that makes all the difference between AI and human intelligence. On the basis of the available evidence this conclusion can be neither confirmed nor denied. The following lesson from history is salutary.
A field of research closely associated with AI is the study of human consciousness. The oldest and most persistent explanation for consciousness is the presence of a unique 'soul' in every human. Thus the soul is postulated as the vital 'missing link' distinguishing humans from other entities. The modern search for this 'missing link' was started by Descartes, who
...conceived the notion that mind existed in a separate sphere from the material universe, a concept that lingers still. In his scheme the brain was a sort of radio receiver that tapped into the dimension of mind via the pineal gland - the only brain component Descartes could find that was not replicated in each hemisphere. (2)
In addition to the theory of the soul, many other explanations for the phenomenon of human consciousness have been put forward, e.g. the breakdown of the 'bicameral mind' (3) and the development of the 'inner voice' (4). The current favourite with commentators (e.g. (2), (5)) - and with this author - is Deacon's theory of humanity as a 'symbolic species' (6). A common theme of all these theories is the implicit assumption that there is a single key feature - a vital 'missing link' - that, prior to publication, nobody else has apprehended. The sheer plethora of theories on this topic, their tendency to exclusivity, and the appreciation that they can't all be right, instils a degree of caution in relation to the assumption of the existence of a vital 'missing link'. The 'missing link' may be nothing more than an artificial by-product of the way we choose to conduct our research.
That being said, it is a postulate of these pages that there is a 'missing link' in relation to AI. As suggested above, this postulate cannot be argued one way or the other on the basis of the available evidence; otherwise, it wouldn't be a postulate. (Neither, for that matter, would there be a 'missing link', if it were that obvious.) However, as we have seen, there are indications that this is a reasonable assumption:
The computer simulation of specific human attributes is well-advanced, but there has been little progress towards AI's goal of understanding and simulating human intelligence;
In particular, computers have trouble learning by themselves;
It appears that the focus of AI research has been the mimicking of human attributes rather than directly addressing this key weakness of computers.
This brings us to the main point. There exists the field of cybernetics, whose terminology has been adopted wholesale by AI, but whose original purpose and direction have long since been neglected. Cybernetics, applied to the formulation of an automated learning engine, is postulated as the AI 'missing link'.
cybernetics. A subject which dates from 1942 and was named in 1947 by Norbert Wiener and Arturo Rosenbleuth, distinguished mathematician and physician respectively. It was then defined as 'the science of control and communication in the animal and the machine'. This definition indicated (1) that a state of 'in-control' depends upon a flow of information, and (2) that the laws governing control are universal, i.e. do not depend on the classical dichotomy between organic and inorganic systems. The name cybernetics derives from the Greek word meaning 'steersman', and was chosen to show that adaptive control is more like steersmanship than dictatorship. Today, a more general definition of cybernetics might be preferred: the science of effective organisation. (1)
On the contrary, this author prefers the original definition.
Recent advances in computing power mean that we can no longer blame the lack of progress towards AI on the absence of resources. We have all the resources we need - it's just that they're not joined up in the right way. That is, we have not given sufficient attention to the control of, and communication between, the components of our computing machines. In other words, the 'missing link' for AI is cybernetics - as originally defined.
A key theme of cybernetics (as originally conceived) is the importance of feedback. One would have thought that, by now, all aspects of feedback had been fully investigated and understood. This appears not to be the case, however. In the formulation of IDEAL it initially proved convenient - and ultimately proved essential - to draw a distinction between 'data' and 'control'. Thus IDEAL has both data connectors (solid arrows) and control connectors (barbed arrows):
This distinction between 'data' and 'control' appears to be foreign to cybernetics. Without this distinction, feedback is a dull, sterile process, comprising two limited variants ('positive' and 'negative' feedback), neither of which fulfil the high hopes of the founders of cybernetics and earlier enthusiasts such as James Clark Maxwell. Perhaps this is why the focus of cybernetics shifted from AI to the business and military aspects of control and communication. In any case, when one introduces a distinction between 'data' and 'control', and subsumes feedback within the broader notion of 'iteration', then the resulting systems can be much more interesting. Indeed, a full implementation of these ideas leads directly to IDEAL, which is not at all sterile in its functionality. Thus IDEAL may be seen as the true heir of the original cybernetics - and the vital 'missing link' for AI.
Alan Bullock and Oliver Stallybrass (eds.), The Fontana Dictionary of Modern Thought (Fontana 1977).
Rita Carter, Mapping the Mind (Phoenix 1998).
Julian Jaynes, The Origin of Consciousness in the Breakdown of the Bicameral Mind (Penguin 1993).
John McCrone, The Myth of Irrationality (Macmillan1993).
John McCrone, How the Brain Works (Dorling Kindersley 2002).
Terrence Deacon, The Symbolic Species (Penguin 1997).
The Theorist: "Very interesting... but What led to IDEAL?"
Click here to return to contents page
Copyright © Roger Kingdon 2004