Our guest in this episode is Andrea Miotti, the founder and executive director of ControlAI. On their website, ControlAI have the tagline, “Fighting to keep humanity in control”. Control over what, you might ask. The website answers: control deepfakes, control scaling, control foundation models, and, yes, control AI.

The latest project from ControlAI is called “A Narrow Path”, which is a comprehensive policy plan split into three phases: Safety, Stability, and Flourishing. To be clear, the envisioned flourishing involves what is called “Transformative AI”. This is no anti-AI campaign, but rather an initiative to “build a robust science and metrology of intelligence, safe-by-design AI engineering, and other foundations for transformative AI under human control”.

The initiative has already received lots of feedback, both positive and negative, which we discuss.

Selected follow-ups:

• A Narrow Path - main website
• ControlAI
• Conjecture - Redefining AI Safety
• What is Agentic AI - Interface.AI
• Chat GPT’s new O1 model escaped its environment to complete “impossible” hacking task - by Mihai Andrei
• Biological Weapons Convention - United Nations
• Poisoning of Sergei and Yulia Skripal - Wikipedia (use of Novichok nerve agent in Salisbury, UK)
• Gathering of AI Safety Institutes in November in San Francisco
• Conscium - Pioneering safe, efficient AI
• The UK's APPG (All Party Parliamentary Group) on AI

Music: Spike Protein, by Koi Discovery, available under CC0 1.0 Public Domain Declaration