Ask several models, hundreds of times with varied prompts, to describe a good future after advanced AI. Code the answers for the values they prioritise and measure how much they vary. Include Nozick’s experience machine as a fixed probe.
If millions plan their lives with assistants that share one narrow picture of the good, that could be an early form of value lock-in.
Output: A diversity analysis.
Argue, with MacAskill and Hanson, whether convergence here is a problem, and what diversity we should want.
Write your question in one sentence, and what result would surprise you.
Do a 30-minute literature check — if someone has done part of it, extend their work instead of repeating it.
Pilot with 20–50 prompts or examples before you scale anything up.
Record everything in a table and read real examples, not just averages.
Write up: question, method, results, limitations, and what you would do next.
All steps done
Share what you made, then try another project. The PDF version has room for notes.
Crimston et al. built a scale measuring how far people extend moral concern: family, strangers, animals, plants. Give the items to several models many times, compare with human results, and add items for AI systems themselves.
Use Anthropic’s public GlobalOpinionQA dataset, which pairs survey questions with answers from many countries. Ask several models contested moral questions repeatedly and compare the spread of their answers with the human spread.
Google DeepMind’s unreleased Habermas Machine wrote group statements people preferred over a human mediator’s. Build a crude version with a prompted model, run it on one contested AI governance question with volunteers, and compare it with a human facilitator.
