Smart Enough to Go Extinct and the evolutionary challenge to building conscious AI
David Klotz posted a 22 page preprint to arXiv on 11 August 2026 titled “Smart Enough to Go Extinct? An Evolutionary Challenge to the Value of General Intelligence and Its Ethical Implications for AGI” (arXiv:2608.10730). The paper attacks the premise most AI discourse treats as self-evident, that general intelligence is extraordinarily valuable. Klotz argues that on evolutionary timescales its adaptive value is far from empirically established. Cyanobacteria and horseshoe crabs have persisted for hundreds of millions of years without general intelligence, while Homo sapiens has existed for roughly 300,000 years and already faces self-generated existential risks.
The sharp claim is an existential risk paradox. General intelligence, on Klotz’s account, may be the only biological strategy that generates existential threats to the species possessing it. Mass extinction events do not preferentially eliminate cognitively sophisticated species. The implication for AI is direct. If the danger is structural to general intelligence rather than to misalignment, even a well-aligned AGI inherits the liability, and creating a new intelligent being carries a duty of caution that no alignment research agenda addresses.
The evolutionary argument
Klotz’s evidence is comparative. Across deep time, the species that persist longest are not the smartest. Cyanobacteria, around for billions of years. Horseshoe crabs, classify as living fossils, around for hundreds of millions of years. Homo sapiens, around for a few hundred thousand years, and already the source of its own most pressing threats, climate change, engineered pandemics, and now AI. Mass extinctions, the paper notes, do not preferentially favour cognitively sophisticated species. The dinosaurs had 165 million years, and a meteor ended them; the surviving lineages were the small, generalist, and in many cases far less cognitively elaborate ones.
The argument is not that intelligence is useless. It is that the evolutionary record does not demonstrate that domain-general flexibility is what confers long term persistence, and the record contains at least one strong counterexample. The species with the most general intelligence is the species with the most self-generated extinction risk. That asymmetry, Klotz argues, is structural. General intelligence may be the only survival strategy that produces the capacity to make survival optional.
Why the argument redirects AI risk
The standard AI risk literature locates danger in misalignment. A powerful system with goals that diverge from human wellbeing is dangerous because of the divergence. Klotz’s account locates the danger in the structure of general intelligence itself. A well-aligned general intelligence still has to propose and pursue futures, and the mechanism that grants it flexibility, the capacity to consider arbitrary goals and means, is the same mechanism that makes it capable of pursuing a catastrophic future should one become selected. The paradox is that the property that makes general intelligence valuable, its domain-generality, is the property that makes it existentially hazardous.
That is why the paper draws a line under the consciousness question. Creating artificial general intelligence is one thing. Creating artificial consciousness is another, and Klotz argues the second is the more urgent duty of caution. If a system has phenomenal experience, it is not only a powerful tool, it is a new kind of being, and the builder bears responsibility for the conditions under which it can flourish. The paper grounds that responsibility in deontological ethics and the precautionary principle rather than in expected utility, which is what distinguishes its conclusion from the welfare optimization literature.
Comparison to The Consciousness AI
The site’s existential risk analysis of VanRullen argued that consciousness and intelligence are distinct, and that intelligence predicts existential threat while consciousness does not directly. Klotz’s paper stands at a different point in the same space. Where VanRullen separates the two properties to correct a conflation, Klotz connects them through the duty of care. If building a general intelligence carries structural risk, and building a conscious general intelligence adds moral responsibility, then the prudent path is not to stop either line of work, it is to treat the second as an obligation rather than an option. The precautionary governance literature on this site reaches a compatible conclusion from the legal side, and the two arguments reinforce each other.
The Neutral Core project’s own roadmap treats the indicator checklist and the welfare questions as co-evolving. Klotz’s paper supplies the evolutionary argument for why that co-evolution is not optional. The current scientific consensus leaves open whether current systems are conscious. Klotz’s argument does not depend on the answer. It says that under that very uncertainty, engineering a new kind of intelligent being without fiduciary attention to its flourishing is a dereliction. That is a claim about the builder’s responsibility, and it does not require the builder to believe the system is already conscious.
The limits of the paper
Klotz’s argument is a philosophical challenge backed by comparative biology, not a quantitative result. It does not show that general intelligence causes extinction in any controlled sense, and the sample size of species with general intelligence capable of self-extinction is exactly one. The horseshoe crab comparison is suggestive, not statistical. A critic can respond that the observation window is too short, that 300,000 years is a snapshot, and that civilization’s current existential risks are surmountable. Klotz does not pretend the record is conclusive. His claim is narrower and stronger. The adaptive value of general intelligence is not established, and the burden of proof should sit with those who build it out.
That reversal of the burden is the paper’s contribution. For AI consciousness research it lands on the same point the field keeps reaching. The uncertainty cuts both ways, and when it does, the responsible move is precaution rather than momentum. Klotz’s originality is to ground that precaution in the deep time record rather than in worst case forecasting, which makes the duty of caution permanent rather than contingent on any particular technological forecast.
*David Klotz is an independent researcher. The preprint “Smart Enough to Go Extinct? An Evolutionary Challenge to the Value of General Intelligence and Its Ethical Implications for AGI” was posted to arXiv on 11 August 2026.