AI searches for the research ideas no human community is positioned to find
The alien space of science: Coherent but cognitively unavailable directions
Scientific discovery is limited not just by what's true but by what researchers can imagine. Many viable directions never get proposed because no existing community combines the needed concepts and methods. This NeurIPS 2026 paper decomposes papers into "idea atoms," then scores combinations for coherence and for how unlikely any current author group is to produce them. On 16,068 LLM papers, the sampler yields more diverse portfolios than frontier LLM baselines while matching or beating their per-idea quality, and recovers 2025 combinations at 35.9 times chance.
We target the complementary region, the alien space of science: directions that are scientifically plausible but unlikely under the distribution of existing researchers.
- martincsweiss
Co-author here - first paper from my postdoc accepted to NeurIPS main conference.
We built a way to sample diverse ideas in machine learning - ideas that make scientific sense, but that no existing research community is positioned to propose.
In short, the way it works is like this:
1. Take 16,068 LLM papers (NeurIPS, ICLR, ICML, ACL, EMNLP, 2017 to 2025) are broken into 273 "idea atoms". Every paper, and every candidate idea, becomes a set of atoms.
2. Train a Coherence Model: a 42.8M-parameter transformer trained on papers as sequences of atoms. Basically a language model whose tokens are ideas.
3. Train a Cognitive-Availability Model: a dual encoder trained on the publication histories of 39,791 researchers, which predicts where each person could plausibly go next.
4. Use those models to score sampled tuples using reciprocal rank fusion (you want ideas that go together well, but that are not cognitively available to anyone).
We tested it in a bunch of ways. One interesting one was that we took all the papers at the end of 2024, then checked which high-scoring tuples showed up in 2025 papers. It worked well - one examples is an ICLR 2025 oral, goes from rank 1,489 with coherence alone to the top 25 of 1.14M samples.
A funny one: if you ask Claude to come up with an ML research idea, it is often about interpretability. If you then run an autoresearch pipeline with some A100s on the idea, the result tends to be interpretability 8/10 times. Our Alien Sampling method stayed at […]
- unsupp0rted
I love this. My personal completely unsupported theory is we’re cognitively capable of conceiving of less than 1% of what’s true/accurate about the universe, and cognitively capable of understanding 1% of that 1%
- tauntz
A kind-of similar concept made the rounds yesterday but for games, instead of science.
- smallerize
Damn, this was the killer app for semantic spaces I'd been waiting for since Word2vec.
- Error6571
Semantically similar papers: https://papermatch.me/?q=2603.01092