What Will Be Left For Us to Work On? Princeton Professor's ICML Keynote Tackles AI Job Anxiety

Arvind Narayanan's ICML 2026 keynote challenges binary AI narratives, proposing co-superintelligence as the path forward for human-AI collaboration.

Arvind Narayanan, the Princeton computer scientist known for co-authoring the influential "AI as Normal Technology" framework, delivered a keynote at the 2026 International Conference on Machine Learning in Seoul last week that confronted the question haunting the AI community: what happens to human work when AI can do more and more of what we do today?

The talk, titled "What will be left for us to work on?", made three core arguments. First, the "AI as Normal Technology" framework — which treats AI as a tool that follows familiar patterns of technological adoption rather than an unprecedented disruption — remains the correct lens, "unless and until there is some future discontinuity such as through recursive self-improvement."

Second, even taking recursive self-improvement seriously, "there is no milestone that companies might achieve in the lab that will suddenly put us all out of work." The gap between capability benchmarks and real-world deployment turns out to be wide, and Narayanan's team at Princeton has been building evaluation methods that measure factors beyond raw benchmark scores — cost, reliability, integration friction, and organizational readiness.

Third, jobs of the future "will be radically different, and a lot of adaptation will be needed." Narayanan ended with a vision he calls "co-superintelligence" — humans and AI systems working together in ways that amplify rather than replace human intelligence.

The keynote arrives at a moment of heightened anxiety. Narayanan noted that if the AI research community "simply rolls over and accepts that a lot of our work will be done by AI in the future, instead of setting clear boundaries," it will fuel a stronger political backlash against the technology. "The whole world is watching," he said.

Narayanan's framework pushes back against the binary narratives — utopian automation on one side, mass obsolescence on the other — and instead argues for a messy, negotiated middle where AI changes work rather than eliminates it. The slides and annotated transcript are available on his Substack, NormalTech.

Sources: NormalTech, ICML 2026 Schedule