A FACT360 BLOG SERIES – PART 3
In his third post in ‘The Science of AI’ series (read the first in the series here), FACT360’s Chief Scientific Adviser, Professor Mark Bishop outlines how our understanding of ‘thought’ has changed throughout history and how it influences the design of AI systems today…
In the classical era, as Aristotle famously explored in De Motu Animalium (350B.C.), the idea of the mind and thought was encompassed in the idea of the soul. In stark contrast to modern theory that identifies the mind with the brain, this posited ‘the heart as the seat of the senses’, the controller of both voluntary and involuntary movement and that the heart really did rule the mind.
These ideas had influence well into the renaissance period when, in 1649, Descartes laid the foundations of a new Dualist division of body (res extensa) and the immaterial soul/mind (res cogitans). In ‘Cartesian Dualism’ the immaterial mind interacted with the material organ of the brain via a mysterious interface located within the pineal gland. Subsequent support for Descartes conception of the brain – as the organ which physically controlled behaviour – came from the work of the English physician Thomas Willis who, in 1667, identified links between the physical structure of the brain and certain pathological behaviours (e.g. epilepsy and other convulsive diseases).
Knowledge from experience
Emerging from what later became known as the British Empiricist school of philosophy, John Locke was perhaps the first philosopher to identify ‘the self’ with a ‘continuity of consciousness’. In addition, in his 1690 ‘Essay concerning human understanding’, Locke famously suggested that at birth the mind was to be considered a blank slate (‘tabula rasa’). Contrary to the prevailing Cartesian view, which held the basic processes of thought and mind to be ‘innate’, Locke maintained that all knowledge comes from experience, is causally dependent on experience and is justified solely by experience.
Furthermore, Locke suggested that the mind is organised, at least in part, by Association, such that items which ‘go together’ in experience will ‘go together’ in thought. Subsequently, David Hume refined Locke’s vague notion of ‘going together’ by reducing it to three core ‘principles of association’: identity, contiguity in time/place and cause/effect.
The ‘associated ideas’ could be memories, images, thoughts etc, with complex ideas being constructed from ‘simples’ and simple ideas being derived directly from raw sensations. Thus, for Hume and Locke, as raw sensation is caused by things ‘outside the head’, thought is ultimately grounded by our interaction with ‘objects of the world’.
During the late nineteenth century, the great American psychologist and philosopher, William James also engaged with Associationism. James considered three broad forms of associative thought:
- ‘general association’, in which there is unrestricted association between concepts (e.g. The memory of a walk followed by a romantic dinner);
- ‘partial-association’, in which only some prior experiences lead to associated consequences
- ‘focussed-association’, which bridges thoughts that do not naturally ‘go together’. As an example, James considers: ‘How might thoughts of a gas flame lead to thoughts of football?’, and unfolds the sequence in the following way: think of the gas-flame which conjures up visions of the moon (via a shared ‘similarity of colour’ – pale-whiteness); this, in turn, is linked to the thoughts of football (via ‘similarity of shape’ – roundness). So, via focussed recall, thought can move from gas-flame to football, even though neither gas-flame nor football have any properties directly in common.
To determine what experiences are ‘partially associated’ together, James sketched four additional principles:
- habit (the more often something done the more likely it is to be associated)
- recency (more recent events are more likely to be recalled)
- vividness (the more intense an experience the more likely it is to be recalled)
- emotional congruity (similar emotion backgrounds are more likely to be associated together).
At any time, the strongest principle of association is that which pertains. In this way, James sketched a general ‘Associationist theory’ of thought.
Artificial Neural Networks
Computer science has been influenced by these interpretations of thought and Artificial Neural Networks (ANNs) have a strong correspondence to Associationism. For example, the input nodes of an ANN could represent ‘sensations’ (data from sensory transducers); the internal nodes encode ‘ideas and thoughts’; the inter-node weights encode strengths between ‘thoughts’ and the output nodes encode ‘behaviour’.
However, thought is not as straight forward as you may think. And while ANNs’ mechanistic architecture provides the support for Associationist psychology and philosophy, it also entails that the various well-rehearsed criticisms levelled at Associationism also apply to ANNs and pose deep, theoretical challenges to neural network approaches to Artificial Intelligence.
I will examine some of these challenges in more detail as this series progresses and show the latest thinking that attempts to solve them. Next time, I look at the importance of language. And if you thought thought was challenging for AI, that that exists exists in that that that that exists exists in.*
*That that exists exists in that that that that exists exists in.
This grammatically correct sentence using the various forms of ‘that’ highlights some of the challenges AI has to overcome to ‘understand’ the intricacies of language. It also links to classical criticism of ANN based AI which I will cover in more detail later in the series, but in the meantime you can solve the sentence yourself:
Switch the demonstrative adjective ‘that’ with ‘this’ and the relative pronoun ‘that’ with ‘which.’ And when there are two examples of the verb ‘exist’ next to each other, change the second to ‘occur’:
[The fact] that ‘that’ exists occurs in a situation which this ‘that’ exists [also] occurs in.
Professor Mark Bishop is FACT360’s Chief Scientific Adviser and to see how these leading-edge scientific techniques can be applied in your organisation download our White Paper “The Science of FACT360”or get in touch [email protected].