The document is a preprint titled Thousands of AI Authors on the Future of AI. It discusses the results of a large-scale survey conducted in 2023 among 2,778 AI researchers from top AI conferences. The survey aimed to gather expert predictions on AI progress, the timeline for achieving significant AI milestones, and the potential societal impacts of advanced AI systems.

Key Points:

1. Survey Overview: The survey is the largest of its kind, involving AI researchers from six top AI venues. It expands on previous surveys from 2016 and 2022, adding more participants and questions.

2. Predictions on AI Milestones: The survey found that AI systems have a 50% chance of achieving several significant milestones by 2028, such as autonomously creating a payment processing site or producing a new song indistinguishable from a human-created one. There is a predicted 10% chance that AI systems will outperform humans in all tasks by 2027 and a 50% chance by 2047.

3. Concerns about AI: A majority of respondents expressed concerns about AI's long-term value. While many believe AI outcomes could be good, a significant portion also acknowledges the possibility of extremely bad outcomes, including human extinction. Between 37.8% and 51.4% of respondents consider a 10% chance that AI could lead to catastrophic outcomes.

4. AI and Society: The survey highlights widespread concern about AI's impact on society, including the spread of misinformation, increased inequality, and the potential for authoritarian control. There was no consensus on whether faster or slower AI progress would be better for humanity.

5. Research Priorities: There is broad agreement that research aimed at minimizing AI risks should be prioritized. Many respondents believe that AI safety research is not currently prioritized enough and should be given more attention.

6. Intelligence Explosion: The possibility of an intelligence explosion, where AI rapidly accelerates technological progress, was discussed. While most respondents did not expect this to occur within two years of achieving High-Level Machine Intelligence (HLMI), they acknowledged it as a possibility over a longer period.

7. Automation of Labor: The survey predicted that the full automation of labor (where all human occupations could be performed better by machines) could happen by 2116, much later than the predicted timeline for HLMI.

8. Uncertainty and Variability: The survey results indicate significant uncertainty among AI researchers about the future of AI. Responses varied widely, reflecting differing opinions on AI's potential benefits and risks.

He highlights that GnAI systems, which produce outputs based on patterns, can contribute to the erosion of critical human capacities like attention and questioning

• We can rebut here that social media has already achieved that point.