What Happens When Intelligence Becomes Cheap

Summary

The discussion argues that superintelligence already exists in narrow, measurable domains, such as chess and Go, where AI can simulate outcomes and verify whether a result is correct. In many areas, raw reasoning ability is no longer the main obstacle to practical AI. More important challenges include connecting models to reliable information and existing software, enabling longer-term planning, and building systems people can trust.

Levels of AI

The speakers distinguish several levels of AI:

• AI: Systems people can interact with conversationally.
• AGI: A general-purpose model that can respond usefully to many kinds of tasks.
• Superintelligence: Performance beyond humanity along a particular dimension.

Automating the Scientific Method

In Richard Socher’s view, a major opportunity is moving beyond automating routine work toward automating parts of the scientific method. AI could generate ideas, design experiments, validate results, and improve hypotheses—helping address problems in medicine, climate, business, and software.

Researchers would still need judgment and “taste” to choose worthwhile problems, guide AI agents, and assess their findings.

Potential Effects on Work and Society

The speakers examine how cheaper intelligence could reshape professions and expand access to services:

• Legal work: AI can handle document review and other routine tasks, but adoption depends on incentives, trust, and the market. Clients may resist paying for work AI can do cheaply, while law firms’ reputations and human negotiation skills remain valuable. AI-native firms may combine AI efficiency with human expertise and credibility.
• Journalism: Transcription is an example of work AI can readily automate. Reporting, interviewing, and understanding events firsthand are presented as more human-centered activities.
• Art and creative work: AI lowers the cost of producing images and music. Human creativity may shift toward choosing concepts and directing execution, though some people value the process of making art, not just the finished output.
• Physical work: Industries with little digital training data, such as plumbing, may be harder to automate soon than information-heavy fields such as law.
• Access to services: As the cost of intelligence falls, services once limited to wealthy people—such as legal advice, personal assistance, and help interpreting health information—could become more widely available.

Human Agency and the Future of Work

The speakers question whether intelligence will remain a reliable way to rank people if AI makes it widely accessible. Socher argues that people are multidimensional and that agency, creativity, and the ability to manage AI may become more important. People will need to know how to delegate, when to trust AI, and when to verify its work.

They also discuss the possibility of universal basic income (UBI), while noting that work can provide meaning and a sense of contribution.

Main Speakers

• Richard Socher: Introduced as a former Stanford professor and Salesforce chief scientist who is now working on AI ventures.
• Robert Mahari: Introduced as Stanford CodeX associate director and founder of Akiva, a company providing AI solutions for legal work.

Scope

No product reviews, step-by-step guides, or tutorials are presented. The content is a broad discussion of AI’s capabilities, adoption barriers, and potential effects on work and society.