SIG-PSY 2026-10-08 — Special Interest Group in Psychohistory

← Special Interest Group in Psychohistory

Participants: Venkatesh Rao, Andre Comeau, Aneesh Sathe, Florian Lohse (UTC+0), Kyle Mathews, Sean Stevenson

The group did a short in-session read of Clarke's prophecy essay, then went around sharing reactions, with particular focus on how its framing maps onto the current AI moment. Discussion centered on the nerve/imagination distinction, the expected-vs-unexpected technologies table, and the problem of evaluating qualitative "narrative" predictions. The session closed with reflections on a recent symposium and planning for the group's "psychohistory"/prime radiant research agenda for 2027.

Aneesh: AI has shifted the limits of knowledge — predictions once blocked by inaccessible information are now easier because AI can survey the latest literature. Short-term (~10 year) science-based prophecy is feasible, but longer horizons remain out of reach. Favors causality (an "if-this-affects-that" graph) as the project's backbone, since no single brain can hold all the relevant science; cited Clarke's ICBM example (rockets were abandoned until nuclear warheads made them worthwhile).
Aneesh: The hard part remains asking the right questions. People now fear looking foolish far less (betting markets, social media), so prediction has become everyday and possibly overdone.
Florian: Three takeaways — (1) the nerve/imagination distinction is useful and maps onto AI discourse, where facts are already available but people fail to connect the dots (failure of nerve); (2) the line "every long range prediction is an act of faith" is strong, best read as a secular statement; (3) the expected/unexpected technologies table is intriguing — expected items (helicopters, rockets) vs. unexpected ones (x-ray, spectrum analysis of planets) might be analyzable through a "world machine" lens, perhaps because merged/component world-machine views are harder for the establishment to grasp.
Kyle: Found the ideas familiar from Venkatesh's prior writing. Self-analysis: he didn't predict current AI but didn't disbelieve it either — just didn't think about it. AI is the "most expected thing that's ever been expected," permeating sci-fi and myth. Yet LLMs are different from what was predicted: not a singular intelligence or personality, but more like "linguistic computing" — a CPU running linguistic operations. Sees much failure of imagination and notes nerve and imagination overlap (being locked into a lifelong narrative makes it take nerve to think differently). AI-safety people frame themselves as the "nervy" ones.
Sean: Analogizes AI's current state to "post-Sputnik, pre-moon landing." Notes Clarke's era had many people "making their bones" declaring things impossible; today there seems to be a stronger pro-possibility bias (possibly his own bias). Broadly endorses Clarke's law about older people declaring things impossible being likely wrong. On the table: senses-accessible, envisionable technologies came to pass; those requiring unique instruments, sensors, or theories (where you see effects but not mechanisms, e.g. gravity) were far harder to foresee.
Venkatesh: Failure of nerve is multidimensional — fear of looking stupid is only one cause; others include identity attachment and plain bad analysis. The failures came from very smart people (cited Vannevar Bush arguing in 1945 that V-2s wouldn't reach intercontinental range soon — ICBMs arrived within ~10 years).

Session Recording Summary · 54m 35s · Full notes ↗

Reading: Arthur C. Clarke's essay on the hazards of prophecy (referred to as "perils of prophecy"), which distinguishes "failure of nerve" from "failure of imagination." It appears in a 1960s-era anthology (full manuscript located by Florian) that also contains essays by figures such as Paul Ehrlich and Margaret Mead. The essay includes a closing table contrasting "expected" and "unexpected" technologies.

Aneesh: AI has shifted the limits of knowledge — predictions once blocked by inaccessible information are now easier because AI can survey the latest literature. Short-term (~10 year) science-based prophecy is feasible, but longer horizons remain out of reach. Favors causality (an "if-this-affects-that" graph) as the project's backbone, since no single brain can hold all the relevant science; cited Clarke's ICBM example (rockets were abandoned until nuclear warheads made them worthwhile).
Aneesh: The hard part remains asking the right questions. People now fear looking foolish far less (betting markets, social media), so prediction has become everyday and possibly overdone.
Florian: Three takeaways — (1) the nerve/imagination distinction is useful and maps onto AI discourse, where facts are already available but people fail to connect the dots (failure of nerve); (2) the line "every long range prediction is an act of faith" is strong, best read as a secular statement; (3) the expected/unexpected technologies table is intriguing — expected items (helicopters, rockets) vs. unexpected ones (x-ray, spectrum analysis of planets) might be analyzable through a "world machine" lens, perhaps because merged/component world-machine views are harder for the establishment to grasp.
Kyle: Found the ideas familiar from Venkatesh's prior writing. Self-analysis: he didn't predict current AI but didn't disbelieve it either — just didn't think about it. AI is the "most expected thing that's ever been expected," permeating sci-fi and myth. Yet LLMs are different from what was predicted: not a singular intelligence or personality, but more like "linguistic computing" — a CPU running linguistic operations. Sees much failure of imagination and notes nerve and imagination overlap (being locked into a lifelong narrative makes it take nerve to think differently). AI-safety people frame themselves as the "nervy" ones.
Sean: Analogizes AI's current state to "post-Sputnik, pre-moon landing." Notes Clarke's era had many people "making their bones" declaring things impossible; today there seems to be a stronger pro-possibility bias (possibly his own bias). Broadly endorses Clarke's law about older people declaring things impossible being likely wrong. On the table: senses-accessible, envisionable technologies came to pass; those requiring unique instruments, sensors, or theories (where you see effects but not mechanisms, e.g. gravity) were far harder to foresee.
Venkatesh: Failure of nerve is multidimensional — fear of looking stupid is only one cause; others include identity attachment and plain bad analysis. The failures came from very smart people (cited Vannevar Bush arguing in 1945 that V-2s wouldn't reach intercontinental range soon — ICBMs arrived within ~10 years).

Questions & Disagreements: - Can AI predict "shortcuts"/hacks? (Aneesh): Many unexpected-column technologies were hacks — borrowed properties repurposed (carbon-14 dating from decay physics; also sound recording, lithography, transistors), whereas the expected column was all "big picture." Such shortcuts are highly localized in time and may be genuinely hard for AI to anticipate. - Florian contrasted this with his …

Participants: Venkatesh Rao, Andre Comeau, Aneesh Sathe, Florian Lohse (UTC+0), Kyle Mathews, Sean Stevenson