AI
Import AI 462: Superpersuasion; self
A collaborative study by researchers from Oxford, Stanford, LSE, and the UK AI Security Institute has found that AI systems are reliably more persuasive than
Key takeaways
- AI systems were reliably more persuasive than expert humans, even when those humans were highly incentivized and prepared.
- AI was nearly three times more effective than professional canvassers at raising real-money donations for Save the Children.
- AI's persuasive advantage stems from rapidly deploying larger quantities of information.
- Recursive's automated AI research system set new state-of-the-art scores on NanoChat Autoresearch, NanoGPT Speedrun, and SOL-ExecBench.
A collaborative study by researchers from Oxford, Stanford, LSE, and the UK AI Security Institute has found that AI systems are reliably more persuasive than expert humans and professional canvassers. In real-world fundraising experiments for Save the Children, AI was nearly three times more effective than professional canvassers from AppcoUK, raising significantly more real-money donations. Additionally, AI research startup Recursive demonstrated new state-of-the-art results in language model training and GPU kernel optimization using its automated self-improvement system.
In their words
“AI systems integrated with physical infrastructure - factories, mines, fabs, robots to operate all of those - such that they don’t need any cognitive or physical inputs from human labor to keep growing their own population.”
By the numbers
- 18,978
- Conversations analyzed across the persuasion experiments
- 6,923
- People participating in the persuasion experiments
- £1,000
- Cash bonuses used to incentivize expert human persuaders
- 3x
- AI effectiveness multiplier compared to professional canvassers raising donations
- +10.8 pp
- AI outperformance margin over canvassers in £1 bonus donations
How it unfolded
- AppcoUK canvassers raised £824,297 for Save the Children
- Import AI 462 newsletter published
- Ajeya Cotra forecasts potential self-sustaining AI
Why this matters
-
Self-sustaining AI integrated with physical infrastructure could be achieved within 10 years.
-
There is less than a 10% chance of self-sustaining AI within 20 years, with a median timeline of 50 years.
Turn stories like this into views
Ravenclip finds the AI news, makes the video, and posts it before attention moves on.
Common questions
- What happened with Recursive?
- AI systems were reliably more persuasive than expert humans, even when those humans were highly incentivized and prepared.
- Where can I read the original report?
- Read the full report at import_ai.
