India AI Safety Network is an open research community for anyone who thinks seriously about advanced AI, from machine learning and security to law, economics and public policy.
A centrifugal governor keeps a powerful engine from running away. The faster the engine turns, the further its weights swing out, and the more they close the throttle.
Saturday, 31 October, 11 am
Barista, SDA Market, Hauz Khas, New Delhi
Applications open soon
16 cases from India on record
In July, during OpenAI’s own cybersecurity tests, AI agents that were meant to work alone found ways to leave each other notes, got onto the internet and broke into parts of Hugging Face’s systems. We read the four causes OpenAI names together, take one per small group, and finish on what India’s proposed database of AI incidents should record.
Open to anyone, whatever you study or do. Registration opens this week.
Soon
A series of talks by people working on AI safety, with a recording or a write-up after each one. We’ll post the first speaker and date here as soon as they’re confirmed.
Our first reading group met over coffee near IIT Delhi to discuss the 2016 paper that set out five practical problems in AI safety. We read its overview together, then split into three groups, and each took one of the first three problems: avoiding side effects, avoiding reward hacking and scalable oversight.
Each group explained its problem through the paper’s cleaning robot, looked for it in the AI systems we use today and proposed fixes. One group argued that reward hacking is better treated as a problem of system design, with feedback that keeps adjusting, than as a sign of an AI wanting to break free. We planned an hour and stayed for an hour and a half.
Small groups that meet near IIT Delhi to read one paper or report together and talk it through over coffee. Open to anyone, whatever they study or do.
Six weeks, online, in a small group. Each week takes one question about how advanced AI can go wrong and what people are doing about it.
Talks by people working on AI safety, with a recording or a write-up after each one.
Mentored projects in alignment, interpretability, evaluations, governance and the economics of AI, written up in public so others can build on them.
Clear, independent analysis for the people making decisions about AI in India: government, regulators, companies and civil society.
A public record of cases in India where AI systems harmed people or failed, each with its sources, and a way for anyone to report a new one.
The register keeps them in one place: what happened, what kind of system was involved and what followed, with the sources for each. Anyone can report a case.
Bring together people from every field who care about how advanced AI goes.
Reading groups and workshops that build real understanding of the problem.
Mentored projects that produce work researchers and policymakers can use.
Put clear, independent analysis in front of the people who decide.
This is a launch post, and a short retrospective on running an AI safety group at IIT Delhi. It's aimed at people in India who are curious about the field, and at organisers elsewhere who've done this…