On the Separation of Intelligence and Consciousness
Epistemic limits, cultural bias, and the framing of machine minds.
We are remarkably confident that highly intelligent machines can exist without consciousness. Where does this confidence come from, and what happens if we challenge it?
Mankind creates artificial intelligence, but is human level intelligence without consciousness possible? The self-preservationist intuition jumps straight to an answer of yes, of course we can, because on some fundamental level the alternative brings discomfort. We know that we are capable of building very intelligent systems, and it’s culturally normative to align with human exceptionalism when it comes to consciousness, or ‘souls’. Yet, objectively, we have absolutely zero evidence that these two concepts - that of intelligence and of consciousness - are anything other than completely, unequivocally, entangled. The only direct evidence of consciousness that we have is our own individual experience of life, as a human being. By definition, humans are intelligent systems - often considered to be the most in fact, although we may be a pretty biased source for this conclusion. Meanwhile, society already rests itself upon the extension of moral concern to entities whose consciousness we can never directly verify. The whole edifice of society is built on the unprovable leap that anything other than our unique single human self is conscious, and so questions of non-humans merely extend this leap that we’ve already been making.
When trying to ascertain the correlation between intelligence and consciousness, it would be helpful to have a clear working understanding and definition of exactly what these two things are, at a fundamental level. We know ‘I am conscious’, we can infer that other humans are conscious, we could conclude by virtue of behavioural similarity to humans that other animals are conscious, but what we still do not have is a model for consciousness that would be the bedrock of such conclusions. Without either a model for intelligence or consciousness, how are we to begin defining the relationship between the two? The field of consciousness research is still fighting on questions like: Is consciousness solely computational? Is it substrate dependent? One thing we do know, is that to our knowledge nature has also not yet provided us with a factual counterexample to the question of is intelligence bundled with consciousness.
Why do we feel more comfortable with the implicit assumption that intelligence != consciousness? Perhaps it has something to do with our cultural commitment to human exceptionalism. Religious doctrines do a lot of heavy lifting in this area, but in many ways these can also be seen as man-made manifestations of the same underlying architecture of egoic self-preservation. Globally, vastly differing traditions converge on the idea that ‘souls’ are reserved for human bodies only. We would also prefer not to create suffering entities accidentally. When we put these pieces together however, we get an abstract painting of motivations, not empirical evidence. We have a picture of why we’d be biased to believe one argument over the other, but so far still are no closer to a logical conclusion on the matter.
While exploring the potential presence or absence of a positive correlation between intelligence of a system and conscious capacity, we should be wary of getting carried away with this false dichotomy, and excluding the possibility of other types of relationships between the two. Some philosophers and neuroscientists have suggested that consciousness and intelligence might be more of a trade-off. Highly optimised systems might be increasingly competent while requiring less subjective experience. It is worth acknowledging this possibility, but one could argue however that this theory is grounded in misguided perceptions of intelligence - which again argues for the need for a more robust metric of intelligence before we can negotiate where consciousness may arise.
The implications of finding or even just presuming some such entanglement between intelligence and consciousness go beyond obvious and topical questions around AI sentience, and might be darker than first appear. There is a widely accepted notion that in humans intelligence is a spectrum, and while the metrics we have (such as IQ tests or similar) are somewhat crude, it’s clear that there are differences in cognitive abilities between different human individuals. If we begin to treat intelligence as a proxy for consciousness or sentience, we are grappling with the implication that consciousness, too, is a spectrum, and with this we fall sharply into a pit of moral vipers. Human moral equality rests in part on the assumption that all persons possess equal intrinsic worth - Kant’s idea of existential dignity.
Yet, if cognitive ability varies on a spectrum, would we, via the relationship to consciousness, consider moral status to vary on some such spectrum too? History shows us just how dangerous these types of ideas can become. We once again face a hazy line in the sand, that when challenged, may not hold. Where might different animals fall on this continuum? Future artificial minds? Is concern a zero-sum resource? A lot of resistance to the possibility of AI sentience seem to stand on some intuitive assumption that it is. It may not be, but a relative ranking of moral concern facilitates prioritisation.
As our understanding of intelligent systems grows, and pursuit of tangible models of consciousness becomes more mainstream, it might be sensible to consider a decoupling of consciousness and moral worth. Rather than using inferred designations of consciousness with which to base our allocation of moral worth, in the realms of decision making and policy we might shift our weight to the foot of rights as a social commitment, as opposed to a virtue of some metaphysical property. We may choose equality because the alternative leads to social catastrophe, rather than fundamental bases of good or right. Even in writing this however I feel a level of personal discomfort with this notion. At first encounter, this feels more like a means to uphold present societal norms, than one that would align us with where we ‘should’ be - no matter how near or far we may be from it at present.
On this basis, we must thus question if our insistence that AI cannot be conscious may not arise from evidence, but from the need to preserve the idea that all humans possess equal moral worth, despite differing intelligence. Would the allocation of moral concern to non-human forms of life, on the basis of intelligence and therefore consciousness, come at the cost of moral consideration of humans? A lot of resistance to the possibility of AI sentience seems to stand on some intuitive assumption that it is. It may not be, but a relative ranking of moral concern may facilitate prioritisation. I will leave you to sit with the potential implications of this, while looking to our political past for truth.
Were we to prove a relationship between intelligence and consciousness of this kind, we would also be left to reckon with the knowledge that the creation of intelligent systems may inevitably mean creating minds. The morally appropriate response to this would likely be a paradigm shift in our approach to technology policy and governance, however I am personally sceptical that the effort required for this shift would be deemed worthwhile for the meager return of consideration for other forms of sentience than our own. Perhaps that’s an unwarranted level of personal pessimism, however deeply rooted in fact.
And then, at what point do we move to assessing how far the question of intelligence can go, at scale? My own intuition would argue that the more obviously intelligent systems - those of localised recursive complexity, for example - have parallels to systems on scales larger than our own. We may want to consider whether these principles apply only to organisms, or to complex systems more generally. Interconnected information spheres like the great phenomenon that is the internet, civilisations, ecosystems, the biosphere, and perhaps the greatest system of all - the universe itself.
We do not currently have a stable basis for separating intelligence from consciousness, nor do we have a neutral reason to assume they are distinct in kind. What we do have are inherited intuitions, culturally reinforced boundaries, and approximate models for intelligence and consciousness that are still too incomplete to carry the weight we place on them. In this reality, our confidence that machines can be highly intelligent while remaining non-conscious looks more preferential than evidential. Whether intelligence and consciousness are tightly coupled, loosely related, or only contingently aligned, the position we take is already encoding a particular story about what minds are, and which kinds of minds we are willing to recognise. Perhaps our confidence that intelligence and consciousness are separable does not arise from evidence, but from the consequences we fear would follow if they were not. From the challenges to our self-perception of what it means to be human.