Richard Hanania

Notes that expert extinction estimates spread from 0.1% to 100% and that superforecasters, AI specialists and Metaculus forecasters differ widely, concluding the experts are guessing. He explains the gap sociologically: Metaculus grew from the doom-focused rationalist community, while the Forecasting Research Institute comes from academic forecasting research. Free preview plus passages he quoted from it on X inspected; the rest is gated for free subscribers and was not read.

After the Hugging Face incident, an Anthropic resignation and Dario Amodei’s pacing essay, says some slowdown would be unsurprising, then asks why China, quick to suppress perceived harms, is not slowing AI. In posts linking it he adds that English-speaking countries are most negative about AI, China regulates deployment rather than models, and Western AI workers may be in a cultural bubble. Free preview and those posts inspected; remainder gated. The slowdown remark is a forecast, not a preference.

Agrees there are reasons for concern but argues optimism is undersold. Compares alarming lab incidents to crash tests with seatbelts and airbags removed, says real-world harm from misaligned agents is roughly zero so far apart from deliberate human misuse, and judges AI harms far smaller than social media’s. Only the free preview was inspected; the rest of his argument, the sponsor blurb and reader comments are excluded.

Argues that democracy works partly because elites often do not do what voters want. Citing polls showing majority opposition to nearby data centers and support for a moratorium, he calls data-center activists paranoid and unable to weigh costs and benefits; the subtitle forecasts that the anti-data-center movement will lose. Paid post: only the free preview was inspected, and his full argument and policy conclusions are unknown.

Reports a reader experiment in which Claude nearly matched his writing on one op-ed, while an AI detector identified every text correctly. Expects AI writing to win the detection arms race, credentials and track records to matter more, and more human thinkers and writers overall, since people will still want human authors and humans will always have other jobs. Full essay inspected; the survey is an informal reader poll.

UnHerd column arguing that social media is populist while AI is technocratic, so AI replacing social media could make discourse less conspiratorial and more factual. Cites research on AI debunking conspiracy beliefs and on X users’ fact-checking requests to Grok and Perplexity. Presented as a possibility, not a certainty; he grants that propaganda models are possible but expects little market demand. Full column inspected.

Older essay. Finds doomer arguments intellectually somewhat compelling but argues skeptics need only one of many objections to be right. Using probabilities he says he made up, he gets an 88% chance doomers are wrong, says other arguments would lower the 12%, and computes 4% for risk that is existential and tractable. Calls that enough for some to work on safety, opposes lobbying for government control, and warns panic could delay a better future. Full essay inspected; 2026 statements take precedence.