How understanding valence could help make future AIs safer-continued
A reply to Mike’s reply
Hi Mike,
Having been through several conscious loops of over-riding and reconfiguring “valence” at many levels (in training for deep diving, in training for survival in very cold conditions, in surviving a terminal cancer diagnosis, and others) it seems clear to me that while valence may start with relatively simple biochemical roots, it can actually become a very complex function in and of itself. It seems to me, from about 50 years of active investigation on myself (over 40 of those within the context of having a very high probability that indefinite life extension was a real possibility that might be available in my lifetime, and from my interests in biochemistry, AI, logic, computing, complexity, evolution and life more generally), that valence can itself become a recursively complex system with many modifier functions.
I am now very confident that the “physics” of consciousness is essentially a software on software thing – a software entity existing in a subconsciously created software model of reality, and of course all software is influenced by the “hardware” it is running on, and we have many levels of highly evolved hardware systems within us. So it seems clear to me that it is a very messy, very complex, highly recursive set of systems. And having been writing computer systems for over 40 years, I know how easy it is to create systems whose behaviour is not predictable by any method faster than simply letting them do what they do – maximal computational complexity has that messy property. I like both Jo Ito’s and David Snowden’s approaches to such things.
Ted Howard NZ's Blog 
Hi Ted,
Thanks for the remarks. Here are some off-the-cuff thoughts back.
You note that “while valence may start with relatively simple biochemical roots, it can actually become a very complex function in and of itself”. I would actually go the other way– that it necessarily has a complex biochemical profile, because our brains are irreducibly complex. Valence, however, could be a fairly ‘simple’ property of conscious systems. (Or not- we just don’t know yet.)
Agreed that the human-brain-as-software is hugely recursive. (Sidenote: if IIT is right, and qualia can be modeled as a set of vectors in some very-high-dimensional space, what does the messy recursion & self/world models look like in terms of this geometric representation?)
But I think a key fork in the road, so to speak, is whether to understand consciousness, we should pay attention to what the ‘bits’ are doing, or the ‘quarks’. I suspect quarks, since the physicist’s ontology maps better to reality than the software engineer’s. But both seem possible at this point in time.
LikeLiked by 1 person
Hi Mike,
I am kind of dreading replying to this, because of the levels of complexity I “see” and the near impossibility of communicating them.
I started to get seriously interested in biology about 50 years ago, and got seriously interested in computation, computers, neurophysiology, cybernetics etc a little over 40 years ago, and have been experimenting on myself for most of that time, to some degree or other, as well as looking at what others have done and published or communicated to me personally.
I am very confident that valence in conscious systems is at least as complex as in the biochemical systems, and seems to be infinitely extensible (through recursion to earlier systems). And I say this being very conscious that consciousness is but the tiny tip of the vast computational “iceberg” that is us. I don’t consciously create these words, they come from my subconscious, I simply consciously set the context, then consciously review what comes out – doing occasional edits.
I consciously reprogrammed my tastes for food, and that process isn’t complete, the old desires are still present, but the power of their pull is somewhat reduced. It has amazed me how many people would rather die than make that conscious effort.
I guess I have been consciously aware since 1974 that indefinite life extension was possible, and have been considering possible risk mitigation strategies in all other domains of risk (particularly our social, political and technical institutions), so that everyone might have a reasonable chance of living a very long time. So for me the idea of a future of exponentially expanding rewards outweighs any present costs – the thing is to survive. Survival trumps everything else.
That thought is so deeply woven in every level of consciously influenced valence within me, and has allowed me to do things like going over 2,000 days without missing a twice daily dose of vitamin C, and not consciously eating anything with added refined sugar (a rather restricted diet in today’s world).
And long before cancer, I spent several years of consistent daily practice that resulted in quadrupling my lung capacity, and allowing me to hold my breath for over 7 minutes. Though achieving that required learning how to consciously override the CO2 breathing reflex, and to alter my metabolism in other more subtle ways, and to sustain a very low level of consciousness, even after senses like vision and hearing stopped working – the sense of temperature actually seems to be one of the last useful ones to go – at least in my case.
When you ask is it about the bits or the quarks that is very interesting.
Most people when they think of computation think binary computers – two state systems – 0 or 1, true or false, source or sink.
Some of us are old enough to have experienced analog computation.
Some of us have worked with non-binary logics, leading to probability systems.
One can create analog systems that simulate binary computers.
One can create binary systems that simulate analog computers.
It seems that any computational modality can be expressed as some function of any other system – but biology tends to add the constraint of having to do so in a useful time frame.
So while Searle’s Chinese Room Argument might actually be able to generate a consciousness of a sort, it would not be one with any particularly significant survival probability in the reality human beings face, and evolution, at every level, is ultimately about survival – what works.
It seems clear to me that all of Kant’s “a priori” categories are at base simply heuristic “hacks” that have worked in practice (in the evolutionary contexts of their specific developmental pathways).
So it isn’t for me so much a matter of bits or quarks, but more about what one thinks that the notion of quark might contain.
If one conceives of quarks as complex systems, stochastic functions operating within probability constraints (a mix of the lawful and the random, displaying properties that support a close approximation to causality in some contexts {like Feynman’s sum over life history approximation} and allowing for freedom in other contexts {like human free will, in as much as it can be said to exist}) then it provides a model of reality that can be reasonably usefully approximated in computational systems.
Programmers have to approximate stochastic systems frequently, particularly when using MCMC models like in population studies.
When even the simplest system imaginable (a one dimensional array of a two state system, that has only 256 possible rule states) can display maximal computational complexity in rule 30, how much more complex is our reality?
When one looks closely at it, the notion of hard causality, of strict Boolean logic, is little more than a childish myth that has bedeviled philosophers since time immemorial.
Reality doesn’t seem to be constrained by such simple notions, though it does approximate them at some levels, and our consciousness needs to have simplifying assumptions that work with sufficient reliability to be useful in evolutionary contexts.
I don’t see much difference between physics and software engineering – all of our conceptions are ultimately software, that much seems clear beyond any reasonable doubt.
The more deeply one delves into notions like quantum mechanics, the standard model of cosmology, string theory, the more reliant one becomes on software both to gather observations, and to generate models that have any meaning to other humans.
One simply cannot draw a clear distinction – the systems are far too complex, far too tightly integrated, far too much flexibility and permeability in the boundaries.
There is real power present, and also real dangers.
Our minds have to make simplifying assumptions to be able to deal with such things. It becomes very difficult to retain conscious track of the implications of those assumptions. The levels of error possible expand infinitely. Uncertainty reigns.
And there is a place for beauty, for art, for creativity, in the heart of science; where else can one get a hypothesis to test?
One fascinating fact to come out of database theory, is that for a fully loaded processor, the most efficient search possible is the fully random search.
My processor has been fairly fully loaded for a very long time 😉
LikeLike