Research Statement

A Reward Signal for the Academy

Research as a whole is an activity with sparse reward. A fundamental discovery or a useful material are not the result of a single experiment, but the end point of a long sequence of decisions about what to try next. So far, the science operates on the principle of peer review. This process is flawed on many levels. The technical level is something a well-motivated human can reliably do; it's a low-hanging fruit for an AI:

• Peer review is supposed to check technical validity. Humans are lazy, AI is not. AI will check the math.
• Peer review is supposed to assess the context of the related work. With a human it's a hit or miss, with Deep Research it's not.

The high-hanging fruit is the strategic level, the mystic novelty, significance, and research taste. Human reviewers are notoriously bad at it:

• Novelty and significance are subjective and not well-defined
• Echo Chambers leading to model collapse – of course I think that space groups are the most important part of generating new crystals.
• The problem is genuinely insanely hard. Answering the question "is this study useful?" requires tracing its implications all the way to the real-world applications.
• Fundamentally, the system lacks any kind of feedback mechanism. An enthusiastic, polite, well-meaning reviewer who is simply bad at significance assessment will do a lot of reviews without ever knowing he is wrong.

AI, in its text and agentic LLM form, has a unique ability at gathering and processing information at scale. It can read every paper, every blog post, every tweet, and every grant proposal. It can track the impact of each piece of work on the field and on the real world.

AI will have a feedback loop. Human reviewers are not incentivized to be right, but AI is.

Finally, elegance. For too long the scientific progress was a mysterious philosopher's journey among Platonic Forms. Let's finally make it into an RL policy.

The impact will be profound. A 10% improvement in annual grant allocation is a compounding 10% acceleration of scientific progress.