TB: AI Discourse Is Haunted by Presentism Bias (Bentham's Bulldog)

Most public discussion of AI is anchored to today's capabilities and today's failure modes, and so it systematically misweights the field — overweighting near-term harms (water use, job displacement, hallucinations) and underweighting catastrophic risks from systems that will exist in a few years. The argument is structural: if your prior is fixed to current performance, you cannot reason properly about a technology whose capacity is doubling on a four-month timescale. The fix is to argue about future systems on their own terms.

• The dominant framing of AI in op-eds, academic critiques, and social media treats current chatbots as the object of evaluation; that framing is dated almost as soon as it is written.
• Hallucinations, art-imitation, and similar near-term issues are likely to diminish as systems advance; treating them as the central problem locks the discourse to a moving target.
• John's annotation in Blogs.org: "Over weighted on near-term risks, underweighted on long-term."
• The author's preferred concerns are existential: weapons enablement, concentration of unprecedented power in a few hands, and loss of meaningful human control over superintelligent systems.
• Current AI capacity is "doubling every four months" — historically unprecedented — so reasoning about AI without modeling this growth is reasoning about a stationary process that does not exist.
• Comparing AI to past technologies (electricity, the internet) understates the discontinuity because no prior technology compounded on this timescale.
• Society is "radically underprepared" for systems that exceed human cognitive ability across all domains.
• Concerns about near-term water and energy use are not trivial in absolute terms, but on the relevant policy axis they are comparable to the impacts of any other transformative tech and crowd out attention from the existential layer.

• "If you are only thinking about current AI capabilities, then your overall assessment of AI will be biased by the present moment."
• "We're building technology that could allow AI company heads to launch a coup or proliferate trillions of digital minds."
• "We are summoning digital aliens into our world, and we are not ready."

The article is a methodological critique, not a doomer pitch — it diagnoses a specific epistemic failure (anchoring to current performance in a fast-moving domain) that distorts both alarmist and dismissive takes. The complement to John's recurring theme on AI urgency in academia: it is the same anchoring failure that lets tenured faculty argue from GPT-4-era anecdotes about systems that no longer exist.

• Tyler Cowen on AI and Academia — the same temporal-misperception failure, applied to academic incentive structures
• Academics Need to Wake Up on AI (Kustov) — Kustov's "jaggedness explains polarization" is presentism's sibling
• Something Big Is Happening (Shumer) — same urgency thesis from a builder's vantage

AI Discourse Is Haunted by Presentism Bias — Bentham's Bulldog, Bentham's Newsletter (Substack), 2026-05-21