Consciousness in the Age of Inherited, Substratal Mimicry

Presuppositions about the Inevitable Achievements of A.I.

In 2022, ChatGPT (the latest iteration of large language models) was unleashed upon the internet, and debate about the nature of artificial minds took center stage. One of the popular opinions offered by the zealots of this technology was that it heralded the inevitability of emergent consciousness in the realm of artificial intelligence (A.I.) and even the promise of synthetic minds indistinguishable (in their behavior) from those of humans. I considered these ideas entirely unconvincing. However, having never sat down and processed exactly why, I found myself plagued by a frustration I could not justify. Of course, the vagueness of my intuitions did not diminish the intensity of my frustration. It probably increased it. The claim of emergent consciousness in the context of A.I.  felt so wrong to me; but of course, such intuitions are useless unless examined.

Upon reflection, I found that there are two troublesome presuppositions smuggled into most discussions of A.I.’s “inevitable” consciousness. The first is the presupposed possibility that the vehicle of synthetic materials can ever arrive at the destination of consciousness at all. Consciousness is, at bottom, a binary phenomenon, though once it has emerged certainly does exist in degrees of intensity. Because of this, “progress” towards its emergence cannot be plotted in the way progress is plotted for those qualities which exist entirely on a continuum. And further, the prerequisite qualities for emergent consciousness remain unknown. In fact, complexity is the only vague quality to which these zealots appeal: an appeal which doesn’t even attempt to quantify requisite levels of complexity.

Ultimately, this presupposition relies on nothing quantifiable at all, just an unsupported acceptance that given enough time, every imaginable technological advancement is possible. The fact that humans achieved things which earlier generations didn’t think were possible is used as “proof” that the innovations many people now do not think are possible must be achievable. This issue was the one that frustrated me from the edge of my awareness. Some part of my subconscious understood that the zealots knew they were tossing about claims which could never be truly examined or debunked, because the deadline for the “inevitable” self-awareness could be infinitely extended info the future. It’s a debt to one’s sense of reason that can never be collected.  

The second troublesome presupposition (and the primary focus of this essay) is that we can evaluate the intelligence and potential for consciousness in A.I. the same way we evaluate these things in biological organisms. This topic contains a form of an ancient philosophical question: how do we know anything else (or anyone else) has subjective experience? One cannot jump into the mind of another to verify. Traditionally, the way we deal with the uncertainty of experience in other entities is through rational extrapolation.

 The extrapolation proceeds as follows. First, we assume that other humans have experiences much like our own. This is, of course, the only rational way to run a decent society. The outcomes of not doing so have historically been very harrowing. The second step of the rational extrapolation proceeds by supposing that the degree to which a non-human organism’s “likeness” resembles our own is correlated to that organism’s intelligence (the higher the perceived intelligence, the more likely it is to have a sense of self).

            There are exceptions to this extrapolation of course, but the trend is strong enough to ignore those for a moment. In practice, this process is the most rational process for evaluating the potential for subjective awareness in biological organisms. We know we are conscious, and so we must use ourselves, our “likeness,” as the standard. In modern times, this “likeness” has largely shifted from ascetic likeness to behavioral likeness (particularly as it relates to social behavior). As an example, we know that one of the primary reasons people today think of dolphins as highly intelligent creatures is due to their capacity for complex communication. [1] Though they do not look like us, they resemble us in observable behaviors.

I argue that the standard by which we evaluate the level of intelligence, and by extension the likelihood of consciousness, of an entity is based upon likeness to human behavior; and where applicable, specifically their ability for complex communication. This is a rational standard for biological organisms, but the extrapolation cannot rationally be applied to A.I.

Mimicry

            Unfortunately, evaluating “likeness” becomes a nearly impossible task in the digital age where any behavior (or ascetic) can be mimicked with novel mechanisms. Mimicry itself is not novel to the natural world, but the kinds we observe in biological organisms are fundamentally different from that kind employed by A.I. I’ll introduce two fairly broad meta-categories for the kind of mimicry we can observe in the natural world. In the first, mimicry is intentional, individual, and confers some self-determined advantage (usually a social one).

A silly example I’ve observed in my own life is the purposeful infliction of cauliflower ear through contrived means. In any of the grappling disciplines, competitors will often develop cauliflower ear (a permanent mangling of the ear’s cartilage). This scaring is from an injury, not a fashion statement. However, as the popularity of mixed martial arts increased, I observed that some young men (many of whom did not train mixed martial arts at all) would purposefully damage their ears’ cartilage to get cauliflower ear, in an attempt to look like fighters. This form of mimicry is not inherited. It is generated on the individual level and completely conscious.

            Mimicry in service of social validation.

            The second meta-category of mimicry for biological organisms is unconscious (or operates the same whether unconscious or not) and inherited. We find many forms of this in the animal kingdom which are the result of various evolutionary pressures. [2] One example, somewhat famously exhibited by butterflies, is called Batesian mimicry, in which non-toxic ones mimic the aesthetic features of toxic ones.[3] This mimicry affords an advantage of safety from predators who conclude that mimics are dangerous and not to be eaten. Because these forms of mimicry are genetic phenomena, they should be seen as both unconscious and inherited.

            Mimicry in service of survival.

            Both of these forms of mimicry can be aesthetic or behavioral. The fundamental purpose does not change in either case. The important thing to notice is that the mimicry is always an attempt to extract value for an underlying nature.

            A.I. produces mimicry as well. However, it is of another kind entirely: inherited (though not genetically) while also serving no inherent underlying nature. Because the kinds of mimicry we observe in biological organisms function as emergent tools in furtherance of some other goal which benefits the organism, we know the mimicry reflects something about the inherent underlying nature of the organism, even if the specifics are meant to be concealed. In the case of the want-to-be fighter, a desire for admiration. In the case of the butterfly, a highly vulnerable physical form. The mimicry of living organisms is an overlay for an inherent underlying nature. It is auxiliary. The mimicry of A.I. is the underlying nature. It is substratal.

Substratal Mimicry

What is the difference between an outfit and a disguise? The answer to this question actually becomes more complicated the longer it’s considered. But, perhaps surprisingly, I think the thought processes needed to answer it can also inform our understanding of the kinds of the systems we’re discussing. The first answer I’ll propose is that the disguise is meant to obscure an inherent underlying nature, while the outfit is meant to emphasize an inherent underlying nature. In both cases, an inherent underlying nature is being somehow reflected. The way in which this happens is completely contextual. It’s so contextual in fact that identical components can be used in service of either the outfit or the disguise. A bomber jacket with a popped collar, a cap, and a thick pair of sunglasses could be either an outfit or a disguise. Is this man an aviator or a convict on the run? So, what is the difference between an outfit and a disguise? My final answer is the nature of the wearer.  

            The language and speech of humans is the linguistic equivalent of an outfit or a disguise. We primarily use words to express or conceal our individual underlying nature. This fact is so well understood that when we observe language, we immediately use it to evaluate the nature of the speaker (or writer). We do not even pause to think whether the language reflects any inherent nature at all. For most of time, such consideration would have only been wasteful. What then can be said about the production of language by an entity which (I claim) has no inherent underlying nature?  

            I once walked into my room, in which my hat and jacket were hanging such that it looked like a man was standing in the room. The illusion was strong enough for me to start at the sight. Now, would we call the hat and jacket an outfit? A disguise? Indeed, I was fooled, so one may be tempted to call it a disguise; but remember that outfits and disguises are dependent in their status upon the wearer. There was no wearer. Those clothes, which could be considered an outfit or disguise when fitted to a person, were empty. They reflected no underlying nature, and the man which briefly appeared in my mind as a result of the image was an illusion.

The language of A.I. is an illusion, an empty suit designed with phenomenal capabilities to fool any user engaging with it. It has no inherent underlying nature, and those natures which briefly appear are transient and unmoored. And if the consequence of this technology is to become ever-more “life-like”, it will always produce results meant to fool, no matter how convincing the outputs.

            A.I. produces something different from all other forms of mimicry, because it is the expression of a base-level mimicry. There’s nothing beneath it. No organism mimics as its primary function, but A.I. does. What this means is that the intuitions and processes humans have developed over the course of millennia to detect sentience are fundamentally unreliable when evaluating A.I.

Inherited Substratal Mimicry

Currently, most people (with some notable exceptions) still consider A.I. models to be nonsentient, largely because the outputs are still reliably explained, modified, and predicted; and there still remain some tell-tale linguistic elements which signal differences between human and computer outputs. However, this situation could very well be a transitory one. In the near future, such systems will likely be able to iterate autonomously and with surprising speed, and the linguistic trends separating human and computer outputs could disappear entirely. The temptation will be to interpret the lack of need for direct human intervention and the ever-increasing complexity (and inscrutability) of the mimicry as proof of subjective awareness for such models. This would be a mistake.

I believe that anyone who claims A.I. “becoming more advanced” invariably leads to sentience is simply masking the claim that complexity equals sentience in linguistic vagueness. And further, the reason for this masking is because the claim is axiomatic. It can’t be supported (for now at any rate). If one is not initially convinced by the claim, it’s impossible to prove. This is the sneaky little problem hidden within the promise of sentience for an ever-iterating system.

The other problem with this promise is that the iterating system merely perpetuates more of the same. There’s a glaring lack of reasonable proposed mechanisms for the emergence of consciousness within such a system. Each extension of the technology just inherits and improves upon the previous one’s ability to pretend. The result is a system of inherited, substratal mimicry: an ever more convincing illusion.  

[1] Sickler, J., Lemcke, K., Gruber, S., Fraser, J., Webler, T., Reiss, D., ... & Boyle, P. (2006). Social narratives surrounding dolphins: Q method study. Society & Animals, 14(4), 351-382.

[2] Anderson, B., & de Jager, M. L. (2020). Natural selection in mimicry. Biological Reviews, 95(2), 291-304.

[3] Nishikawa, H., Iijima, T., Kajitani, R., Yamaguchi, J., Ando, T., Suzuki, Y., ... & Fujiwara, H. (2015). A genetic mechanism for female-limited Batesian mimicry in Papilio butterfly. Nature genetics, 47(4), 405-409.