Essay on alien space of science
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
A nine-author paper on the 'alien space of science' reached its third arXiv version on Oct 21.
It argues discovery is limited by what researchers find cognitively available.
Background
The paper, led by Alejandro H. Artiles with co-author Bernhard Schölkopf among eight others, targets what it calls the 'alien space of science': research directions that are scientifically plausible but unlikely under the distribution of existing researchers. It models this with a coherence model and an availability model over 'idea atoms', and argues that scientific discovery is constrained not only by what is true but by what is cognitively available to researchers — a bias it says modern language models inherit. It is published in the NeurIPS 2026 Main Track (a major machine-learning conference).
How it unfolded
- Mar 1, 2026The paper is submitted as arXiv:2603.01092 v1.
- May 15, 2026A second version (v2) of the paper is posted.
- Oct 17, 2026Harvard's Avi Loeb publishes a Medium essay, 'Take Me to the Nearest Alien Spacecraft', on a separate topic.
- Oct 21, 2026A third version (v3), the current one, is posted.
Who’s saying what
- Authors
- Scientific discovery is constrained not only by what is true but by what is cognitively available to researchers, and language models inherit this bias.
- Authors
- On 16,068 peer-reviewed LLM papers from NeurIPS, ICLR, ICML and major NLP venues, their sampler produces more diverse research portfolios than frontier LLM baselines while matching or exceeding per-idea quality in blind human, LLM and downstream experimental evaluation.
- Authors
- Selecting less than 1% of the candidate space recovers 2025 combinations at 35.9 times chance, and weighting unavailability raises recovery of strictly novel combinations to 3.24 times chance.
Still unverified
Avi Loeb cites whistleblowers' contention that an alien spacecraft is held by the U.S. government or corporations like Lockheed Martin, framing such tech as a chance rather than a confirmed fact.