AI as Aurochs Intelligence

Or, why ChatGPT makes you stupid

I once saw a talk by Dan Dennett in which he discussed the aurochs, the extinct ancestor of the modern cow. (I found this old TED talk where he does the same bit.) The aurochs was a large Holocene bovine that ranged across Eurasia. The last wild individual may have died in 1627 in Poland, driven to extinction by anthropogenic habitat loss and overhunting, but a domesticated variant had split off ten thousand years before. You should care about the aurochs if you care about AI. (Warning: This is going to get a little weird.)

(Aurochs painted in Laxcaux cave, France: Wikimedia Commons)

Dennett made the point that the wild aurochs had a larger brain capacity than the common cow. Aurochs had to think about foraging and predators and mating and whatever else Pleistocene megafauna have to worry about. Cows, however, do not have to think as hard about these things because their human herdsmen do the thinking for them. Cowboys drive them to pasture, look out for wolves, and play matchmaker for bulls and heifers.

Brains are expensive, metabolically speaking. So if you do not need a big brain, then you can use your energy for other things, such as getting fat and being more attractive to herdsmen (who watch over the herd! Where is your mind?). In effect, cows outsource their information processing to humans. An individual aurochs can probably outcompete an individual cow in the same situation. But a human-cow team is much more competitive than a lone aurochs. As a result, cows may be stupider than aurochs, but cows are everywhere and aurochs are extinct.

The enstupification of cows occurred through intelligent design, or human intervention into natural selection. Holocene herdsmen presumably began selectively breeding aurochs herds for meat, dairy, hides, and docility, but not for intelligence. Stupider cows were easier to manage, and maybe even tastier than wild aurochs. Humans thus constructed pastoral niches (grazing pastures, feed crops, markets, etc.) that proved much more conducive to the surival of domesticated cattle than wild niches were to the surival of aurochs (even worse for the wise aurochs, pastoralism encroached on wild habitat). This was good for both humans and cows at the population levels (although maybe not so good for individual cows, which could be arbitrarily turned into hamburger, and not so good for the environment in the long run, but that’s another story).

What the hell does this have to do with AI? Last month a team at the MIT Media Lab put out an arXiv paper entitled “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.” It’s over two hundred pages long, but here are the key findings in the abstract:

This study explores the neural and behavioral consequences of LLM-assisted essay writing. Participants were divided into three groups: LLM, Search Engine, and Brain-only (no tools). Each completed three sessions under the same condition. In a fourth session, LLM users were reassigned to Brain-only group (LLM-to-Brain), and Brain-only users were reassigned to LLM condition (Brain-to-LLM). A total of 54 participants took part in Sessions 1-3, with 18 completing session 4. We used electroencephalography (EEG) to assess cognitive load during essay writing, and analyzed essays using NLP, as well as scoring essays with the help from human teachers and an AI judge. Across groups, NERs, n-gram patterns, and topic ontology showed within-group homogeneity. EEG revealed significant differences in brain connectivity: Brain-only participants exhibited the strongest, most distributed networks; Search Engine users showed moderate engagement; and LLM users displayed the weakest connectivity. Cognitive activity scaled down in relation to external tool use. In session 4, LLM-to-Brain participants showed reduced alpha and beta connectivity, indicating under-engagement. Brain-to-LLM users exhibited higher memory recall and activation of occipito-parietal and prefrontal areas, similar to Search Engine users. Self-reported ownership of essays was the lowest in the LLM group and the highest in the Brain-only group. LLM users also struggled to accurately quote their own work. While LLMs offer immediate convenience, our findings highlight potential cognitive costs. Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels. These results raise concerns about the long-term educational implications of LLM reliance and underscore the need for deeper inquiry into AI's role in learning.

Or with less nuance, ChatGPT makes you stupid. Groups that relied on LLMs to write an essay had less brain activity, and they had trouble remembering what they wrote. Search tools were somewhere in between. In this study, outsourcing intelligence to computational infrastucture reduced the cognitive load on human beings.

This is a neurophysiological image of a longstanding theme in the history of technology of deskilling. Once upon a time, for example, gunsmiths were talented craftsmen who made beautiful bespoke weapons, but with the coming of interchangable parts, armories refocused human labor on the management of the mass production line. Technology enables people to shift the work of production away from individuals and onto the economic system. The overall system becomes more efficient and productive, even as individual labor contributions may become less sophisticated. Thus, in the great Marxist tradition, the laborers become alienated from their production, exploitable and disposable.

I’m just spitballing here, but there are glimmerings of the wild aurochs in our outsourcing of intelligence to LLMs. We can rely on LLMs to write our homework essays, recommendation letters, and coding assignments, and we do not have to expend as much brain power. Maybe we can use that brain power for something else. Or maybe it becomes an unnecessary metabolic burden. Why invest in learning difficult skills when we can thrive as hybrid human-machine agents? If cows are wild aurochs that became domesticated by hominids, then might we one day find ourselves likewise domesticated by AI?

We are careening off into the realm of speculative science fiction at this point, but that’s one of the joys of writing a substack that hardly anybody reads. One possible timeline leads to Idiocracy, where medicine is reduced to a pushbutton fastfood menu. Another leads to some sort of reproductive separation and speciation of human beings, one version who embrace cognitive outsourcing and another who only mate with nonhybridized humans. Perhaps the former drive the latter extinct, or the latter become pets of the former, or they both persist in viable niches.

The most ridiculous timeline might appeal to those who take the prospect of AGI seriously. The evolutionary emergence of a superior intelligence in the future might be analogous to the emergence of intelligent hominids in the Holocene. An AGI might construct a very comfortable low cognitive load niche for the human population. This would be mutually beneficial for AGI and domesticated humanity, at least from a population perspective. The AGI might need something from biological agents in the same way that industrial capitalists needed the small hands of children to fix their machines, so it would domesticate us for that purpose. We would become stupider as we became more dependent on the AGI for our survival, but there would be more of us, fulfilling Elon Musk’s wildest fantasies.

Vintage humans (us) would still be more intelligent than newmans (new domesticated humans). However, the AGI-newman team would be able to outcompete the surviving populations of wild humans. AGI-constructed niches would not only be more hospitable to the propagation of stupid (but tasty) newmans, they would also reduce the viable niches remaining for classic humans. And so we go the way of the aurochs, survived by our enstupified domesticated hybrid offspring.

Well this has gone on quite long enough, and there are fireworks to watch. Happy Fourth of July, everybody. Perhaps my closing charge is to consider declaring independence from machines, if not completely than at least occassionally. There is something to be said for maintaining our cognitive load, ensuring that we preserve the incentives to use the big brains that evolution has given us.

Next time you are tempted to let ChatGPT do your work, just give a thought to the aurochs.