The respondents expressed significant uncertainty about the long-term value of AI, with 68.3% believing that good outcomes from superhuman AI are more likely than bad. However, 48% of these optimists still gave at least a 5% chance of extremely bad outcomes, such as human extinction. Over half of the respondents suggested that substantial or extreme concern is warranted about several AI-related scenarios, including misinformation, authoritarian control, and inequality. There was disagreement about whether faster or slower AI progress would be better for humanity, but there was broad agreement that research to minimize potential risks from AI should be prioritized.
Key takeaways:
- By 2028, AI systems are predicted to achieve several milestones, including autonomously constructing a payment processing site, creating a song indistinguishable from a new song by a popular musician, and autonomously downloading and fine-tuning a large language model.
- Unaided machines are estimated to outperform humans in every possible task at 10% by 2027, and 50% by 2047.
- All human occupations becoming fully automatable is forecast to reach 10% by 2037, and 50% as late as 2116.
- There is substantial uncertainty about the long-term value of AI progress, with respondents expressing concern about potential AI-related scenarios, including misinformation, authoritarian control, and inequality.