Last week, a speaker delivered a keynote at the International Conference on Machine Learning in Seoul titled “What will be left for us to work on?” The talk examined widespread anxiety about AI and outlined a framework for adapting to future AI capabilities.

The speaker, who leads a research team at Princeton University, presented the “AI as Normal Technology” framework, which treats AI as a transformative technology comparable to electricity and proposes a four‑stage model of methods/capabilities, product applications, early adoption, and long‑term adaptation.

The speaker made three arguments: first, the framework is useful unless a future discontinuity such as recursive self‑improvement occurs; second, no near‑term milestone will suddenly eliminate human work; third, future jobs will differ markedly and will require substantial adaptation.

Reliability metrics were examined across 10 to 12 dimensions, showing that capability gains have outpaced reliability improvements by only a few percentage points. The speaker noted that agents used for automation face different reliability concerns than collaboration agents, which must remain flexible and creative.

An analogy to electricity illustrated the need for organizational change: early attempts to treat electric power as a drop‑in replacement for steam power failed, while the eventual reorganization of factories around power distribution took decades and required new labor practices. Similar slow, structural adaptation is expected for AI.

The speaker warned that relying on AI to replace workers could provoke stronger political backlash, and emphasized that evaluation of AI systems — rather than pure development — will become a larger share of research effort, akin to steering a ship rather than rowing it.

In personal reflections, the speaker described a balancing act of growth, productivity, and control, advocating for deliberate skill development and avoiding over‑reliance on black‑box AI tools.

The talk concluded with a call for continued focus on evaluation, alignment, and co‑superintelligence, noting that the pace of change will unfold over decades and that agency over AI deployment remains essential.