You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$52,700 left to reach the funding goal.
I want to spend spring 2027 at Harvard writing my first AI safety paper. The question: what happens to us when AI starts planning our lives? AI assistants already handle our calendars, messages and shopping, and soon bigger decisions. Every single step is convenient. But if we hand over one decision after another, we may slowly lose the habit of deciding for ourselves, and nobody notices because no single step looks bad. I think this matters for safety too: humans can only oversee AI if they still form their own goals.
The paper connects this to the gradual disempowerment idea (Kulveit et al., 2025), but looks at it from the individual person, not the whole economy.
Goal: a published paper that measures how AI planners change people's sense of control, and turns that into concrete goals for how assistants should be aligned.
Jan–Feb: reading (alignment, philosophy of autonomy, research on habits and delegation) and 15–20 interviews with people who already let AI plan a lot of their life.
Feb–Apr: an online study with about 200 people. Half plan real decisions with an AI planner, half without. Then I take the AI away and see how well they decide on their own and how in control they feel.
Apr–May: a simple framework (which decisions are fine to hand over, which ones hurt us), design ideas for assistants that protect autonomy, and a small open test labs can run on their own assistants.
Output: paper submitted to NeurIPS, FAccT or AIES, posted on arXiv and the Alignment Forum, data and test open. I'll post updates here.
Next to this I take graduate courses in ML and AI safety, and I'm part of Prof. Boaz Barak's AI safety course.
Harvard admitted me for the full year, but non-degree students get no aid, and my fall scholarship doesn't cover spring. Costs for Jan–May 2027 (Harvard rates):
Tuition (4 courses): $29,524
Health fee and insurance: $3,449
Rent and living in Cambridge (5 months): $17,700
Paying study participants: $2,000
Total about $52,700. With $20,000 I can go ahead if I find the rest elsewhere. Harvard needs proof of funds for my visa in early December, so I need commitments by the end of November.
Just me for now, and I'm looking for a co-author who has run user studies before.
I grew up on a small farm in Lower Saxony, Germany, and trained as a farmer before going to university. Then WHU (BSc) and London Business School (MSc Financial Analysis, merit scholarship), with internships in investment banking and private equity. I co-founded a social app where creators sell content directly in the feed. We got to over 30,000 monthly users and more than 1,000 transactions, and I pitched it at Harvard to Spencer Rascoff. In October 2026 I shut it down because I want to work on AI.
This is my first research paper. What I bring is that I learn fast and I'm used to shipping things without anyone telling me to.
I don't raise enough by December: then I can't stay for spring, and any money pledged here goes back.
The paper ends up too broad because I have no formal supervisor. I'll keep the scope narrow and get feedback early.
Peer review rejects it: it still goes on arXiv and to a workshop, and I resubmit.
The effect in the study is small: a clean null result is still useful.
My fall 2026 semester at Harvard is paid by a scholarship from the Hasso und Katharina von Etzdorf Foundation. For spring I have nothing committed yet. I've applied to the Transformative AI Fund and written to several foundations.