About the community.

People choose an AI system, and that system increasingly chooses for them: what they watch, where they travel, which restaurants they visit, what art they discover, which papers they read, which products they buy, which tools or skills they use, and more every month. The person makes one choice — which system to ask — and the system makes the rest. The first decision quietly becomes the one that decides the others.

A recommendation has two sides. On one side is the person and what they meant. On the other are the candidates — the restaurants, the towns, the papers, the tools — and the material that describes them: their webpages, their listings, their abstracts. That material used to be written for people. It is now, quietly, being written for models. Not always, and not always badly; a clear description helps a model the way it helps a reader. But when the reward for being recommended is large and the reader is a machine, the material starts to change. A listing can be written for a reader, or written to win the recommendation. Same restaurant, two descriptions. We ask whether the system still chooses for the person when it reads that material.

Humanity’s Last Choice is an open research community studying how AI systems make choices for people. We bring together research on the systems that recommend, the information they read, and the people whose interests those recommendations are meant to serve.

We are building the Humanity’s Last Choice benchmark across the kinds of choice people delegate. Each domain contributes tasks, documented conditions, and results to a shared benchmark that grows as the community studies more kinds of choice.

SafeGEO, accepted to EMNLP 2026, is our first paper. It studies product recommendations when one candidate’s text is rewritten to influence the system. Its tasks and results contribute product domains to the benchmark, with code that others can inspect and reuse.

Our work will grow through further papers, shared evaluations, and contributions across the kinds of choice people delegate. Each study sets out its question, method, and findings so others can examine the evidence and build on it.

Open papers, reusable evaluations, and a shared body of evidence about AI recommendations. We want people to be able to inspect a result, test it on another system or kind of choice, and ask a better question from what they find.

Contributions can begin with a research question, a new experiment, a system evaluation, or a domain to study.

Bring a research question, propose an experiment, evaluate a system, or contribute a domain. We welcome work that helps us understand how recommendations are made and whose interests they serve.

Contact: info@humanitylastchoice.org