SSC Journal Club: AI Timelines

Canonical accessible writeup of the Grace et al. (2017) AI Impacts survey of 352 ML researchers. Walks through the headline number (50% chance of human-level machine intelligence by 2062), the large framing effect that pushed the median to 2139 for a near-identical question (so experts hold no firm view), the task-by-task timeline (Angry Birds 3y ... AI research 80y, with a nice Near-Mode explanation for why researchers rank their own job hardest), and the risk block: 70% endorse Stuart Russell's alignment argument yet only 5% call it among the field's most important problems and 1.4% 'much more valuable' to work on now. Closes with the 'believe-but-won't-act' philosophy-of-charity analogy. Discloses the lead author is his girlfriend.

Quality 73 · Strong. Clear, useful, well-organized summary of an influential and much-cited survey, lifted by sharp secondary observations (the framing effect; the believe-don't-act gap). But it is commentary on someone else's data, not original work: solidly Strong, at the top of the journal-club range (above the serotonin and Friston clubs at 70-72).

Claude’s paradigm shift 45 · Moderate. Moderate. Reporting plus light original synthesis; the framing-effect and prioritization-gap points are insightful, but the post is mostly faithful summary.

Real-world impact 3 · Moderate. A clear, well-organized writeup of the much-cited Grace et al. AI-timelines survey, lifted by sharp secondary observations (the large framing effect; the 70%-endorse-but-5%-prioritize gap) — but commentary on someone else's data. Conceptual/informational influence within AI discourse, no material change — modest RWI.