On the prospects for general artificial intelligence, gAI.

The problem of pain: what's hurting?

When a nerve is stimulated - a prick to the finger, say - unambiguous signals pass up the relevant nerve to the brain. There, somehow, it is turned into a sensation, pain. How this happens is the central mystery that, when solved, opens the door to gAI.

Reflex arcs connect such signals as surpass thresholds to stereotyped physical responses, without any intermediate processing. Our hand jerks away from the rose thorn before we even think about it. Lower animals respond to all such noxious stimuli without "feeling" anything. The nematode Caenorhabditis elegans has just 302 neurons, surely too few to support anything but mechanical reflexes. By contrast, an ant or a fruit fly have each around quarter of a million of them. A cockroach has a million, a frog sixteen million, a mouse 70 million, a rat 200 million. A jackdaw has a billion, a dog two and a half billion. A human has 860 billion.

Somewhere along that continuum, pain - sensations, generalised as "qualia" - come to be. Most would hold a cockroach to be insensate, but believe that a mouse is fully equipped with whatever generates qualia: certainly, rats have a sense of self, of danger, of social drivers of behaviour. So, somewhere North of 50 million neurons, and possessed of an architecture that generalises as between mammals and birds, this remarkable synthesis occurs. (The octopus, with half a billion neurons, has no neuro-anatomical similarity to the vertebrates and we will skip over it, unique as it is amongst the molluscs. Yet many can manage and display complex visual information, much of it clearly social in context.)

Higher animals appear to show varying levels of 'conscious awareness' of a stimulus. Vertebrate pain stimuli are relayed from dedicated sensors through conventional neurons in the dorsal horn of the spinal cord to brain structures which include the thalamus, the somatosensory, insular, and anterior cingulate cortices. Reflex arc connections are spinal. Their action can be observed even when the human spine is completely severed, and do not involve sensations of pain. It is the network of complex brain regions which is capable of generating the sensation of pain, and communicating this to consciousness as a 'thing in itself', an indivisible unit. Such elementary percepts are called qualia, (singular quale).

The thalamus plays an important role in pain detection, as already mentioned. Its processing of the stimulus, and this has two complexes and two outcomes associated with this. The first of these innervates the ventral posterior lateral nucleus. This locates the source of pain in the body image, its intensity and its duration. The second system is located in the medial nucleus of the thalamus, and it is the system which appears to drive the motivational-planning aspect of pain perception. The processing is extremely complex, involving at least five other centres scattered around the forebrain. In other words, this processing is the means by which the stimulus becomes a quale, a part of conscious experience.

Consciousness can itself repress the sensation of pain, for example through hypnotic analgesia, or it can set it aside when it is distracted. Soldiers in battle frequently disregard major wounds until the conflict is over.

Studies of brain response to pain in hypnotised subjects show strong differences as from control subjects. That is, hypnosis has a profound neurological effect on the subject. However, both hypnosis and distraction - for example, a complex puzzle - had roughly equivalent effects in reducing perceptions of pain from a standard stimulus. This subjective evaluation of the pain's intensity was mirrored by electrical signals that were obtained from the brain. Thought processes can act on qualia and the machinery that supports them just as that machinery, properly stimulated, can act on consciousness. If awareness is visualised as an abstract sphere, qualia generation occurs at the outer surface of it.

The same arguments are also true of the more complex stimuli of vision, hearing and so on. Layered structures refine incoming data which, at some stage, enters consciousness as a finished article. Much of what underpins this integration lies in the frontal cortex, but not all of it. Children born without a frontal cortex are plainly self-aware, and mammals show purposeful, goal-directed behaviour after experimental decortication. It, too, seems to be distributed across many systems, somehow coaxed into a unified entity, something more analogous to a social system that it is to a centralised equivalent to an operating system. Qualia are the unifying grammar by which these otherwise very different parts speak to each other.

It is important to recognise that qualia have no special properties, indeed, are indistinguishable from random activity, without a consciousness to react to them. They are "reified" - made into active, if abstract structures - by their contact with pre-existing awareness. In this, they have much in common with all other forms of information. What distinguished tokens of information from noise is the result that they have on the recipient system. Of itself, an item of data has no qualities which flags it as being "information". Only the receptor can deliver that. We will come back to this in a moment.

Emergence

Emergence is an elusive concept. If you set out to describe an entity - a molecule, for example - you can completely capture all of the properties and qualities that it has. The least complex model that does this is generally seen as a complete capture of the degrees of freedom which the molecule possesses. The same model applies to every instance of that molecule. However, if you put a number of these perfectly described molecules together, the resulting situation will show additional degrees of freedom: no one molecule can show 'pressure' or 'temperature', but a group of them can. They can undergo phase changes - ice, water, steam - in which completely new behaviours are emitted. Entropy and thermodynamics become possible; and the gas can do work, transmit sounds and much, much more. Yet none of these things are present in the single molecule: they are said to have "emerged".

How this occurs is both obvious and elusive. We can say how pressure is generated and derive the gas laws from first principles. Yet how the universe acquired these new properties and degrees of freedom is conceptually challenging to describe.

Emergent properties are able to be stacked, as atoms give rise to chemistry, chemistry to life, life to behaviours that are far removed from atomic properties and constituents. Seen from the bottom of such a hierarchy, the minimal model cannot "conceive of", describe, capture the high level properties. Seen from the top of the hierarchy - an ecology, perhaps - one can drive down to the underlying structures, discarding explanatory power as one goes. A fundamental issue is whether this is an artefact of our cognition or whether it reflects a deep aspect of reality.

This makes the concept of emergence vulnerable to "nothing but"-ism. The ecology is 'nothing but' atoms. Anyone with a scientific training is conditions to turn issues around until they can be seen in this way. Yes, we can show that ever-more complex models are needed to describe hierarchies of emergence, but are those models physically meaningful or are they an artefact of our own though processes?

How could one know? Well, you can ask whether there are physically meaningful phenomena that are not merely additive: as one can argue that pressure is the additive effect of lots of molecules in a container. It is clear that there are: some of them mentioned above. Phase changes break the symmetry of the mass of molecules. Entropy and thermodynamics arise from mass action, to be sure, but they are not merely additive. They are conceptually and operationally different from the effect of there being lots of molecules in a space. At all levels of scale, structure emerges from the structure-less and does things that the structure-less, gauge symmetrical system cannot.

Having established that, let's turn to the idea of the "system". Emergence is less about collections of stuff, like a gas, and more about flows and agency: verbs, not nouns. Such flows constitute high level activities such as information processing, biochemistry, even social interactions and economics. You can argue that this notion is 'nothing but' our perspective, imposed on the universe but, plainly, these things exist and do things which are as alien to their component parts as the examples given above. The systems boundary is both theoretical - that is, where the minimal model that captures the system's degrees of freedom change - and practical, in terms of the agency which the system performs on other such systems, or on non-emergent matter.

Emergence can be multiply layered: this set of properties lie meta- to that, and so on up. Any two such hierarchies, isolated at all levels down to and including their roots - can interact with each other. Prior to that contact they are causally isolated. All that they "know" of each other is the contact: market sales as effected by the day's weather, a chain of though disrupted by pain from your foot. A set of flows interact with another isolated system, doing so by tokens, often informational in quality. Indeed, whilst the agency that transmits information may be an electrical impulse, a molecule, a price signal or a movement of troops, the response to it - what we see as the arrival of the information - is generated in the recipient system. Columns, rings and other configurations exchange tokens and store states. Information is processed and yet higher order emergent system result from this.

It may be helpful to consider an example. An insect and a leaf have points of operational contact, but they are each distinct structures, with inner flows and ways of representing and processing information. A system made of lots of both, much as a gas is made of molecules, acquires properties of its own. It's a primitive ecology, showing predator-prey cycles and so on. It is a new system, emerged from two other quite distinct emergent structures, but having different degree of freedom to either of them. Ecologies exchange physical and informational tokens with other systems, and so the structure builds in its complexity. The universe has just become more complex, as it did when the first flying insects found a niche to occupy in the Cambrian.

Minds and emergence

Minds have often been compared to ecologies, societies and similar structures in which many dissimilar agents interact through the exchange of what its useful to generalise as information. The modules are, however, as different from each other as the insect and the leaf. They need need a grammar, a way of interacting that both encodes a common interpretative framework ("meaning") and is additive towards the functioning of the mind as a whole, also "meaning". That shared meaning , is - of necessity - consciousness. Whatever it is, it sits at the summit of the many processing elements, interpreting, synthesising, summarising and directing.

Earlier, we used the word "reification". The dictionary definition is:

Reification (from Latin res „thing“ and facere „make“) Is the process of reconceptualization of the immaterial, or abstract, by regarding it as a material construct, force, or value. bringing something into being, or making something concrete.

Interestingly, philosophy also uses the word to mean "the fallacy of treating an abstraction as if it were a real thing."

If the mind reifies the actions of its modules, and does so in the first sense of the definition, then it is both constructed from those modules and their outputs and is the thing which, by engaging with their emergent properties, makes them real, functional, potent. The final paragraph of the first section notes that qualia have no existence unless reified by a recipient system. They, like other forms of information, are simply inconsequential noise. Signals enter the thalamus and other parts of the frontal cortex, processing occurs and what emerges is a sensation, a quale. This is deeply mysterious, yet it bears all of the hallmarks of emergence. Vision, hearing undergo similar changes in modules. Actions - speech, motion, even balance - are controlled by dedicated modules which get their instructions from, once again qualia. Your higher functions say "reach", expressing this in terms of our body image. Motor-dedicated structures translate that into impulses to the twenty-odd muscles that are involved in reaching, each modulated by feedback from the arm, the eye and other sensors. A high level abstraction drives specialised, self-correcting structures to an end that is both abstract - envisaged by consciousness - and physically concrete. The qualia are generated centrally, and 'pushed' to the periphery; and the periphery generates feedback as simple impulses that are constructed into what are taken in as qualia.

Conclusion

If this is remotely correct, then a gAI will not suddenly pop into existence. The analogy which sees the new born as a formless void is completely incorrect. The initial architecture may be relatively simple, but what generates the hierarchies of ordering, which reifies and parses the qualia into a useful grammar, has to be the outcome of hundreds of millions of years of evolution.

A gAI will need a similarly complicated architecture to process the incoming data streams, set in place before its first light. What makes up such an architecture is going to be based on near-intangibles, hierarchies of order that are perceivable only to themselves. As already stated, it is the receiving system that makes information out of data, and only that system can tell the difference between data and noise. (Shannon and entropy considerations aside.)

If we are able to construct some elements of natural systems into an engineering outcome, the higher order forms will have to evolve naturally and be cross referenced against natural structures which they will then allow us to perceive. The evolution process, and the mistakes made, will lead to the death and madness of many, many instances of the aware system. Then we arrive at a system that has bad hair days and saves for its old age, one that exists as a captive slave. It's not going to be useful, save as a cradle into which to drop human minds. And, if this view is correct, those minds will be simply copies of a common sense of existence - of being "me" - differentiated only by memory.

The consequences of understanding t e sources of awareness go well beyond gAI. You would be able to alter your personality, aptitudes and memories, make multiple copies of yourself, transmit yourself to business meetings in abstract space or New York. The trouble would be, though, the likelihood of completely losing yourself in all of this, of seeing that there is nothing very special about a single instant of "me" and the discarding of the entire human heritage in favour of God only knows what. This may be why the heavens are so obdurately silent, that technology employing species reach this singularity and fall into it, to their eventual oblivion.