Interview, Fireside Chat
How to pivot before the intelligence explosion
- AI R&D is projected to become automated within a couple of years, potentially triggering an algorithmic feedback loop where AIs design better AIs, though this timeline could be delayed to the early 2030s or extended for several decades if compute scaling becomes too expensive.
- Automated AI research could expand the research workforce by a factor of one thousand, compressing five years of progress into a single year as algorithmic efficiency doubles annually and AI revenue grows tenfold to allow for equivalent tenfold increases in compute power.
- Once automated, AI systems could evolve into general digital workers capable of performing almost any virtual job, leading to societal changes moving faster than human tracking capabilities and creating a scenario where a single company employs a workforce larger than the entire human population.
- Specific risks include a 60% probability of AI R&D automation by the end of 2028 according to external forecasts, alongside potential "loss of control" risks if companies deploy hundreds of millions of AI copies simultaneously.
- Economic forecasts present a bifurcated view: short-term mass unemployment of 10 to 20 percent may occur within five years, while other models predict human wages could increase tenfold and continue rising even after full automation due to 99% automation thresholds creating new economic demand.
- Partial automation is expected to initially raise wages and employment in fields like radiology by increasing productivity, but wages for fully automated tasks are predicted to fall while wages for non-automatable tasks rise, likely increasing inequality as labor becomes less valuable relative to capital.
- Future labor dynamics may shift toward "relational jobs" such as luxury crafts and management, with individuals potentially acting as "mini CEOs" overseeing thousands of AI agents, though organizations may initially become more top-heavy.
- Career transitions into AI roles are described as achievable within one to two years for those starting in fields like philosophy or business, with high competition for immediate offers necessitating alternative strategies like working at startups or government for skill acquisition.
- High-value skills are identified as those difficult to automate, including AI engineering for code review, management capabilities, and rapid learning, while other skills may see diminished value as the pace of change accelerates.
- Strategic recommendations include prioritizing AI safety and power concentration prevention for high impact, with policy and government roles being accessible within a couple of months, while individuals may also choose to pursue personal interests if advanced AI systems arrive soon.
- The speaker anticipates that if AI risks do not materialize, global wealth could increase by a factor of 100 to 1,000, and the book "80,000 Hours" aims to reach a standard audience by becoming a definitive career guide.