scripod.com

TIME100 AI Scientist: The Next Era of AI Has Already Started | Richard Socher

Richard Socher, a leading AI researcher and founder of Recursive Superintelligence, discusses his ambitious goal of creating an AI that can improve itself, predicting this breakthrough within two years. He explores the implications of such intelligence, from the challenges of reward hacking to the future of work and the next bottlenecks beyond computing power.
Socher defines superintelligence as a multi-dimensional concept, where AI already surpasses humans in specific areas like protein folding and translation. He identifies metacognition—an AI questioning its own goals—as a key unexplored frontier. For the job market, he applies an elasticity rule: jobs with finite demand, like illustration, will suffer, while those with elastic demand, like software engineering, will see increased opportunities as costs drop. He argues that hardware, not software, is the primary bottleneck for home robots. Socher advises studying AI fundamentals alongside a passion for another field, and advocates for industry-specific AI regulation rather than intelligence-based rules. He predicts that while AI will tackle complex problems like aging, people will still find meaning in human-centric experiences and luxury goods.
00:00
00:00
Superintelligence within two years.
08:40
08:40
Superintelligence is a volumetric concept with multiple dimensions.
20:24
20:24
AI's job impact depends on demand elasticity
25:40
25:40
Hardware is the bottleneck for home robots
33:49
33:49
No company is working on AI goal selection
47:22
47:22
Biological research is slow.