The survey, which is the third in a series, found that most of the 39 specific tasks listed could be feasibly performed by AI within the next 10 years. The respondents also predicted a 50% chance of High-Level Machine Intelligence (HLMI) by 2047, and Full Automation of Labor (FAOL) by 2116. The survey also highlighted concerns about the risks in AI systems, including alignment, trustworthiness, predictability, self-directedness, capabilities, and jailbreakability.
Key takeaways:
- The 2023 Expert Survey on Progress in AI, conducted by AI Impacts, the University of Bonn and the University of Oxford, revealed that if science continues undisrupted, there's a 10% chance of unaided machines outperforming humans in every task by 2027, and a 50% chance by 2047.
- Respondents also predicted a 10% chance that all human occupations could become fully automatable by 2037, and a 10% chance that advanced AI could cause severe disempowerment or even extinction of the human race.
- Survey participants predicted a 50% chance of High-Level Machine Intelligence (HLMI) by 2047, and Full Automation of Labor (FAOL) at a 50% chance by the year 2116.
- There are concerns about AI systems related to alignment, trustworthiness, predictability, self-directedness, capabilities and jailbreakability, with a strong consensus that AI safety research should be prioritized as AI tools continue to evolve.