So, a long time ago I made a casual promise on twitter to write a blog about connectionist implications of approaches to theories of mind and consciousness. Foolish. However, that idea has bubbled away and recently re-emerged in discussing some aspects on the inter-relationship between language, mind, and truth. I’ve edited this and reproduced it here. Make of it what you will.
Connectionist approaches to mind, particularly Smolensky’s sub-symbolic approach, would say in my interpretation of it that words are something like socioculturally negotiated categorical descriptors and are connected associatively to world-events by dint of learning-training experiences. Our experience of language/words/thoughts/beliefs in a computational-philosophical framework is of a virtual machine underpinned by Connectionist mechanisms that are highly attenuated informationally-primed sensitive pattern matching machines – amongst other things. Indeed the best connectionism networks we have to date ‘deep belief networks’ which are forms of boltzmann machines are able to make categorical distinctions of written numers for example – you can show them lots of example ‘1’s’ and it can recognise that it is a ‘1’ even when highly ambiguous or the mechanism is damaged. The learning process that does this is never specifically tutored on what is a ‘1’ or told the desired output it learns to discriminate without such labels. What I take from this is that language is arbitrary and negotiated and built on top of lower-level sub-symbolic cognitive consciousness supporting mechanisms that are primarily concerned with finding statistical regularities – reducing surprisal and anticipating and reducing information-error. Language itself is a virtual machine on top that enables us to express and converse but is subject to sociocultural influences and is in of itself not a necessary prerequisite of conscious experience. On this basis whilst I consider words to be necessarily useful and meaning rich vehicles for experience they can only ever be at best to ourselves partial approximate descriptors of ‘raw experience’ (or qualia) and are always thus potentially subject to revision, oppression and the imposition of others mediated ‘truth’.
This is a talk by Geoff Hinton on deep-belief nets recognising numbers, without instruction. The video starts at 21 minutes – his demonstration of brain states and mind states is cool. https://www.youtube.com/watch?v=AyzOUbkUf3M
With autoethnography I’m with Ellis and am not so concerned with ‘truth’ more with stories that impact emotionally upon the reader and stimulate empathy compassion. ‘Evocative’ autoethnography if you will to my mind has as much value – if not potentially more than ‘Analytical’ strands. I get different things from both. We are tasked in the nursing-education world, courtesy of the Francis report and our CNO, via the 6C’s, with imbuing nursing with requisite care and compassion – (amongst other empirically opaque ‘C’s’) and are forced to leave aside the political injustice of the government’s reduction of the myriad of recommendations to holding our profession accountable for these failings as we do so. I would argue that these, along with the Courage C, are sound reasons to promote the autoethnographic movement within mental health nursing in particular, and nursing in general.
In applying a connectionist approach I’m seeking to undermine the ‘psychological truth’ of psychological models by exposing what I consider to be the shaky foundations they rest on.
Geoff Hinton’s talk I linked you to is a classic example – he seriously talks about models of brain state and models of mental state. They are linked but not the same. As I’ve expressed before elsewhere in various talks about more philosophical-existential aspects of mind, based on my understanding and previous work on connectionist theory and models I see the mind as a ‘virtual machine’ or ‘extra-dimensional property’ – they are inherently hypothetical. Minds themselves and the associated language labels – beliefs, thoughts etc employed in cognitive and clinical psychology are a construct. I may behave and have experiences that can be described as ‘having a belief’ and that is a convenient way of talking about that experience and associated behaviour, but I’m quite convinced that what ‘really’ underlies this is nothing like either the folk-psychological account of mind and mental behaviour nor the clinical psychological version – they are both kinds of approximation. This is evident to me from how classical-AI systems can be built out of sub-symbolic AI systems. The former has rules, categorical labels and so forth much like a lingua franca of the mind. The latter is the machinery underneath that scaffolds all this and has no clue or interest in the above. Thus I would argue that connectionist approaches are potentially quite liberating for this reason, and that it allows us to hold these folk-psychological, cognitive-psychological and clinical psychological models lightly as mere approximations, not by necessity truth.