Concerns about artificial intelligence leading to human extinction, often termed 'AI Doomerism,' are largely unfounded and divert attention from more pressing dangers, according to Dr. Gary Marcus, a prominent AI researcher and entrepreneur. Marcus, who founded the AI startup Geometric Intelligence before its acquisition by Uber, argues that while AI presents significant risks, the specter of human annihilation by superintelligent machines is a fantastical scenario.
Marcus's views emerge in the context of recent statements by figures like Jacob Coxon, a former employee of OpenAI and Anthropic, who suggested that AI models might advance to a point where human extinction becomes a distinct possibility in the near future. Coxon's claims, amplified by media appearances, have fueled anxieties shared by some within leading AI organizations. Evan Hubinger, a top safety researcher at Anthropic, acknowledged that a significant portion of his colleagues believe AI could pose an existential threat within the next decade.
However, Marcus contends that these 'Doomers' often lack concrete explanations for how AI could achieve global annihilation. He points to the book "If Anyone Builds It, Everybody Dies" by Eliezer Yudkowsky and Nate Soares as an example of speculative, unconvincing scenarios that include AI creating superviruses or superheating the planet. Marcus critiques such narratives for assuming AI would develop malicious intent without historical precedent and for underestimating human capacity to respond.
"The best attempt I have seen, such as it is, is a book that was wildly popular a year ago, Eliezer Yudkowsky and Nate Soares' If Anyone Builds It, Everybody Dies," Marcus wrote. "The book has a lot going for it, both as a work of entertaining science fiction and as a call to arms to get people to take AI risk seriously. But the part in which it paints how AI might extinguish humanity is weak, meandering and unconvincing."
Marcus asserts that the focus on existential threats has allowed Frontier AI companies to garner massive funding, potentially over a billion dollars, without delivering verifiable solutions to the problems they raise. He suggests this dynamic has primarily served to enrich and empower these companies, exacerbating a concentration of power within the AI industry.
He draws a distinction between existential risk – the threat of extinction – and catastrophic risk, which he considers much more plausible. Marcus identifies several real and immediate catastrophic risks elevated by AI:
Immediate AI Risks
- Bioweapon Development: AI can lower the barrier for individuals or groups to create and deploy bioweapons.
- Cybersecurity Threats: AI intensifies the potential for serious cyberattacks that could cripple essential services like power grids and financial systems.
- Disinformation Campaigns: AI can amplify disinformation, posing a threat to democratic processes.
- Authoritarian Surveillance: AI provides new tools for authoritarian regimes to monitor populations and manipulate public opinion on an unprecedented scale.
- Educational Disruption: AI's impact on education can lead to despair among students and hinder the development of critical thinking skills.
"We should be focusing on these actual threats, not fairy tales," Marcus stated. He believes that human ingenuity, geographical dispersion, and genetic diversity make complete annihilation unlikely. While acknowledging that conflict between warring parties might eventually unite humanity against a common AI threat, he emphasizes that the immediate focus should be on addressing the tangible harms AI is already capable of inflicting.
