Import AI 462: Superpersuasion; self-sustaining AI; paths to ASI

· Source: Import AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Advanced, long

Summary

A recent analysis highlights several critical advancements and future considerations in AI. Researchers from Oxford, UK AI Security Institute, Stanford, and LSE found AI systems, including Opus 4.1, 4.6, GPT-4o, and Gemini 2.5 Pro, are significantly more persuasive than expert humans. Across 18,978 conversations with 6,923 people, AI was nearly 3x more effective at raising real-money donations, exceeding professional canvassers by +10.8 pp. This advantage stems from AI's ability to rapidly deploy large quantities of information. Separately, discussions on self-sustaining AI, defined as systems integrated with physical infrastructure needing no human input, project timelines from 10 to 50 years, contingent on humanoid robot development and tacit knowledge automation. Google DeepMind also explored pathways from AGI to ASI, considering scaling, algorithmic shifts, recursive self-improvement, and group agent formation. Finally, startup Recursive demonstrated early success in recursive self-improvement, achieving state-of-the-art results in language model training speed and GPU kernel optimization using an automated research system.

Key takeaway

For policymakers and AI ethics researchers weighing the societal impact of advanced AI, you must prioritize monitoring and regulating AI's persuasive capabilities. The demonstrated ability of AI to significantly outperform humans in persuasion, even for real-money donations, implies a profound shift in influence. Your focus should be on establishing frameworks that prevent power consolidation and ensure equitable access, while also tracking progress toward self-sustaining AI and artificial superintelligence to proactively manage future risks and opportunities.

Key insights

AI's persuasive power and potential for autonomous growth necessitate proactive societal and technological foresight.

Principles

Method

The Recursive startup's automated AI research system proposes ideas, implements them, runs experiments, validates results, and uses learning to choose next experiments.

In practice

Topics

Best for: CTO, VP of Engineering/Data, Executive, AI Scientist, Director of AI/ML, Policy Maker

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Import AI.