Human roles & AI

Industry perspective & forecast

Inside the Event Horizon: Why Everyone Should Take AI Progress Seriously

Jack Clark · Co-founder and Head of Policy, Anthropic, as listed by the IMF for this event
Per Jacobsson Foundation Lecture · IMF–World Bank Annual Meetings, Bangkok · Event: 17 October 2026

Synopsis

Clark argues that advances in AI could accelerate scientific discovery and change the organisation of research, work and infrastructure. He describes a possible shift from individuals performing every task themselves towards people directing and evaluating the work of many AI systems. His central human question is therefore about choices: which activities to delegate, how much understanding to retain, and where to insist on oversight and restraint. He calls for scientific infrastructure and governance that could translate technical capability into public benefit. For this collection, the useful focus is the tension between expanding capability and responsible use: deciding that a system can perform a task does not settle whether it should perform it, who sets its goals, or who answers for its consequences. His claim that a technological singularity has begun is his interpretation, rather than an established conclusion adopted by the Observatory.

The human role

• Set worthwhile questions and goals, including whose needs and values count; computational capability does not decide those priorities.
• Exercise judgement over delegation: decide what requires human understanding, review, approval or intervention rather than assuming oversight is meaningful merely because a person is present.
• Maintain accountability and restraint: consider consequences, distribute benefits fairly and retain the ability to question or refuse a technically possible action.

Why it matters for UX and research

UX practitioners can make these decisions tangible through permission boundaries, understandable evidence, uncertainty disclosures, escalation and recovery. Researchers and students can strengthen their work by documenting what they delegated, what they checked themselves and why a decision remained theirs. These are our editorial connections to the talk, not findings from a UX study or guarantees about employment.

Questions to take into practice

• Which decision in your next project should remain human, and what evidence supports that boundary?
• How would someone detect an error, challenge an AI recommendation or stop a delegated action?
• What understanding must a researcher retain to justify a result, rather than merely report an output?
• Who benefits from the proposed automation, who bears its costs, and whose voice is missing?