Is the extinction of Neanderthals really relevant to concerns about AI existential risk?

Last summer, the Forecasting Research Institute (FRI) published the report of their Existential Risk Persuasion Tournament (XPT), led by Ezra Karger and supervised by Philip Tetlock . It was a four-month experiment run from June through October 2022, pitting about 80 experts on nuclear war, climate change, artificial intelligence (AI), biological risks, and existential risk more broadly, against about 89 so-called “superforecasters ”, asking about their probabilistic assessments for the scenario that humanity faces an existential or catastrophic event before 2100 due to various possible causes, including (but not limited to) AI.

A few weeks ago, the FRI published the results of a follow-up research project, namely the Adversarial Collaboration on AI Risk (AC), which ran in April and May 2023, led by Josh Rosenberg and supervised again by Tetlock. The idea was to bring together two groups with distinct perceptions of AI risk, an “AI concerned” one (11 AI experts) and an “AI skeptics” one (9 superforecasters and 2 AI experts), and see if (and by how much) their adversarial collaboration would lead to changes in their respective probabilistic assessments. The result, roughly speaking, was as follows: at the beginning of the project, the concerned and skeptics groups offered probabilistic forecasts that humanity will face an AI-caused existential catastrophe by the year 2100 at 25% and 0.10%, respectively; after the end of their 2-month adversarial collaboration, these forecasts had been updated respectively to 20% and 0.12%, continuing to differ by 2 orders of magnitude.

Discussing these results and what they may imply is not my purpose here (for the interested reader, the reports of both projects offer plenty of useful information, and I recommend going through them; in full disclosure, I did not participate myself in either project). Instead, and as the title already implies, I want to focus on a specific but seemingly important part of the whole discourse.

"We killed them all off, so AI might do the same to us"

Dylan Matthews , writing about the AC project for Vox Future Perfect , notes:

But I can see where the pessimists are coming from. Their basic view is that the emergence of superhuman AI is like the arrival on earth of a superhuman alien species. We don’t know if that species would want to kill us all.

But Homo sapiens didn’t necessarily want to kill all the Homo erectus or Neanderthals hundreds of thousands of years ago, when multiple intelligent species of large apes were coexisting. We did, however, kill them all off.

Extinction tends to happen to dumber, weaker species when a smarter species that’s better at claiming resources for itself emerges. If you have this worldview, the burden of proof is on optimists to show why super-intelligent AI wouldn’t result in catastrophe.

This is an argument I see frequently in the relevant discussions and in various versions and wordings (sometimes limited to other hominids, sometimes extended to animals, too); here is psychiatrist and blogger Scott Alexander Siskind :

But I can counterargue: “There have been about a dozen times a sapient species has created a more intelligent successor species: australopithecus → homo habilis, homo habilis → homo erectus, etc - and in each case, the successor species has wiped out its predecessor. So the base rate for more intelligent successor species killing everyone is about 100%”.

Even people like Dan Hendrycks , executive director and co-founder of the Center for AI Safety , although not stating the argument explicitly, clearly and routinely hint that humanity “could go the way of the Neanderthals ” due to AI.

Truth is, we don’t actually need such anecdotal testimonies by journalists, bloggers, or executives; because the usage of this argument in the relevant discussions and assessments is now clearly documented, thanks to the meticulous research work of the FRI. On page 40 of the XPT report , we read:

AI-concerned forecasters were more likely to place the burden of proof on skeptics to explain why AI is not dangerous. They typically started from a prior that when a more intelligent species or civilization arises, it will overpower competitors.

While on page 127 of the AC report we can indeed see an AI-concerned participant, nicknamed Wesley, arguing:

No infinities needed, just the same sort of thing that happened to other Homo- species when Homo Sapiens emerged

The link leads to a ScienceAlert article from 2019 that claims indeed that we, Homo Sapiens, eliminated the Neanderthals, along with every other hominid species was unfortunate enough to cross paths with us. The article is written by Nick Longrich (notice the name, we will meet Nick again below), senior lecturer of paleontology & evolutionary biology at the University of Bath, and it was originally published in The Conversation .

But beyond the (certainly respected) opinion of any individual researcher, is there any consensus that this is really so? Did we lead the Neanderthals to extinction? And if we did so indeed, was it due to our allegedly higher intelligence?

I trust it should go without saying that, when one is confronted with such questions in any context different from a casual dinner party conversation (such as the FRI research projects discussed here), one ought to go through the actual evidence and scientific consensus, before offering any opinion - let alone before using such an opinion as a foundation for further reasoning, about AI or anything else. Even more so when such argumentation and reasoning often goes well beyond the seemingly innocent and harmless context of some research projects, and it escapes into the real world, where it tries to dictate real policy actions that will affect real people and real societies.

So, let’s proceed to do just so, i.e. check what the evidence and the scientific consensus actually say on the issue…

What exactly is the argument?

Limiting the relevant discussion to Neanderthals arguably makes sense without any loss of generality: along with the Denisovans , they are our closest hominid relatives that we currently know of; and unlike the Denisovans, which were literally unknown to us before 2010, the archaeological record on the Neanderthals is relatively rich.

The argument “we were more intelligent than them so we just killed them off” is actually decomposed into two largely distinct but interrelated ones:

1. The Neanderthals were (significantly) less intelligent and cognitively skilled than us, anatomically modern humans or Homo Sapiens
2. We caused their extinction, directly or indirectly, largely due to #1 above

Up to 10-15 years ago, such reasoning may have sounded rather plausible (and it certainly has some intuitive appeal); but today, neither argument seems to hold water…

Did we really kill them off?

A survey study was published in Scientific Reports on March 2021, titled An emerging consensus in palaeoanthropology: demography was the main factor responsible for the disappearance of Neanderthals ; here is the conclusion, quoting from the abstract:

What does the community actually believe about the demise of Neanderthals? To address this question, we conducted a survey among practicing palaeo-anthropologists (total number of respondents = 216). It appears that received wisdom is that demography was the principal cause of the demise of Neanderthals. In contrast, there is no received wisdom about the role that environmental factors and competition with modern humans played in the extinction process; the research community is deeply divided about these issues.?

Digging in the Results section of the paper, we see that, among the three factors (demographic, environmental, and competitive) thought of having a causal contribution to the extinction of Neanderthals, the participating experts ranked the competitive one as 3rd out of the 3 in magnitude and strength.

The study findings largely agree with an informal survey of 23 verified experts in Metafact ; based on their answers to the question Did Homo sapiens drive Neanderthals to extinction?, the site's aggregation method assigned an assessment of 43%, which we could loosely translate as "we don't know for sure, but from what we know it doesn't seem very likely". Interestingly enough, this survey formed the basis of a subsequent article in ScienceAlert in 2021, now concluding that "there is no clear consensus on what caused the extinction of Neanderthals, and whether our ancestors are to blame", which our Wesley above (and obviously others, too) seemed to ignore in 2023…

Were they really less intelligent than us?

Updates on our perception of Neanderthal intelligence and skills seem to have also happened in leaps during the last decade or so. Back in 2011, it was perhaps only natural for someone to claim in Nature that "there is the awkward fact that what [the Neanderthals] made and left behind is unimpressive. [...] they made no art in the form of painting or carving, just a few perforated and pigmented shells". Fast forward to 2018 , and "a study today in Science has dropped a bombshell on their field [of Neanderthal studies], by presenting the most persuasive case yet that our vanished cousins had the cognitive capacity to create art ". According to a 2021 paper in Nature Ecology & Evolution, they may even have been able to talk (the paper is paywalled, see Phys.org for a detailed presentation); while already from 2014, a PLoS ONE paper aptly titled Neandertal Demise: An Archaeological Analysis of the Modern Human Superiority Complex , concluded that:

We have found no data in support of the supposed technological, social and cognitive inferiority of Neandertals compared to their [anatomically modern human] contemporaries. The results of our study imply that single-factor explanations for the disappearance of the Neandertals are not warranted any more, and that their demise was clearly more complex than many archaeology-based scenarios of “cognitive inferiority” reviewed here seem to suggest. This has implications beyond the field of archaeology per se

The authors’ assertion that “this has implications beyond the field of archaeology per se” arguably rang a bell, already from 2014; alas, it would seem that too many AI-concerned analysts and commentators were simply not listening…

Recall Nick Longrich from earlier above? Turns out that he maintains a personal blog, where he writes non-technical pieces about his research for the general audience; here is his latest take on the subject from March 2024, i.e. literally just a few days ago:

Why did Homo sapiens take over the world while our closest relatives, the Neanderthals, went extinct? It’s entirely possible we were just smarter -and that’s the obvious, intuitive, and widely accepted explanation- but there’s surprisingly little evidence that it’s true. Neanderthals [...] were smart, and it’s possible that the average Neanderthal was about as smart as the average human. [...]

Our ideas about Neanderthals have changed radically in recent years as we’ve learned more about their tools, their lives, and their DNA. Anthropologists once saw Neanderthals as dull-witted brutes . But recent archaeological finds increasingly suggest they rivalled us in intelligence.

Beyond the peer-reviewed scientific literature, such popular non-technical articles (or even interviews) by expert researchers are arguably not without merit themselves; here is another one, aptly titled 20-Year Study Reveals: Neanderthals Were As Intelligent as Homo sapiens , from 2023:

“This confirms our observations and theories from previous studies – explains Diego Angelucci, archaeologist at the University of Trento and co-author of the study. Neanderthals were capable of symbolic thought, could create artistic objects, knew how to decorate their bodies using personal ornaments, and had an extremely varied diet. Add to that that, based on our findings, we can say with certainty that they habitually ate cooked food. This ability confirms that they were as skilled as the sapiens who lived millennia later.”

I could go on citing and quoting from references, but I trust you get the point…

Science or intuition?

So, what exactly is the point? To be clear and not misunderstood, I am not saying that the argument “we were more intelligent, hence we simply killed them off” is obviously wrong; I am just saying that, for anyone bothering to really check the status in the field instead of relying on intuitions and alleged truths, it is extremely far from obvious that it is actually correct, hence its usage as an unchallenged matter-of-fact here seems largely unjustified and arguably invalid.

Can interested and committed generalists who are not themselves experts pick a side in such cases where there is no clear and unambiguous scientific consensus? Of course they can! But we already have unwritten rules for such discourse: if one wishes to support scenario A when the expert consensus does not clearly favor it over a different scenario B, one is at least expected to explain and justify why they think so, so that others can inspect the arguments and judge their merit (recall that such persuasion and collaboration efforts were explicit requirements and objectives for both FRI projects discussed here); which of course presumes that one has already explicitly acknowledged that there is no clear consensus favoring A or B.

But you will find no such discussion here [*], and the argument that “we were more intelligent, hence we simply killed them off” just lingers on, based solely on intuition and sloppy research.

In fact, in the AC report itself (p. 72) it is noticed that, among the fundamental beliefs commonly held by the AI-concerned participants, there was the belief that “the case for extinction is intuitive”.

I am all-in for intuition; but I strictly prefer it labeled as such explicitly and upfront, and certainly not disguised as scientific facts and evidence…

[*] To be precise (and fair), on p. 101 of the AC report , we can see the anonymous AI-concerned participant #7 having lowered their probability of doom from 22.9% to 17.5%, citing as reason for this update that “Reference class of homo sapiens causing the extinction of other homo species less bleak than expected”; but given the apparent centrality of the “we killed them off” argument, this seems too-little-too-late, and I could not find anything else relevant in the ~900 pages of the reports. If I have missed something, feedback is most welcome.