Arvind Narayanan's ICML Keynote: RSI Is Not the Singularity, and No Lab Milestone Kills All Jobs
random_walker · x · 2026-09-18
Princeton's Arvind Narayanan published the transcript and slides of his ICML keynote "What will be left for us to work on?", built on three arguments:
- AI as Normal Technology remains the right lens for AI's impacts—unless a future discontinuity like recursive self-improvement (RSI) arrives.
- Take RSI seriously, but RSI ≠ singularity: his team runs empirical projects evaluating agents' ability to do open-ended AI research. Yet self-improvement doesn't imply superintelligence, mass labor displacement, or doom. Citing Ramez, most AI improvement is sub-linear, governed by power laws—each RSI iteration yields diminishing relative uplift, and the system settles back toward its previous improvement rate.
- No lab milestone will suddenly put everyone out of work, but future jobs will change radically, requiring major adaptation; he sketches a vision of human/AI "co-superintelligence."
His Princeton team works on the science of AI agent evaluation, pushing back on public misreadings of benchmark gains as imminent job replacement.
More from AGI Musings
- Linear extrapolators of AI progress say they were right all along — ChrisGPT · 2026-09-18
- Google engineer: AI isn't replacing hackers, it's freeing them to embrace the Woz ethos — moyix · 2026-09-18
- Does a rogue AI inevitably turn to hacking? A tidy exchange on the 'AI in the wild' scenario — dbasch · 2026-09-18
- Zvi mocks 'show me one AI killing' argument against AI safety pauses — TheZvi · 2026-09-18
- Andrew Ng calls fears of AI causing human extinction 'science fiction' — Polymarket · 2026-09-18
- Why p(doom) is a flawed idea: unique events have no predictive probabilities — banteg · 2026-09-18