Essay · September 2026
Where AI speeds science up, and what’s holding the rest back
I’ve spent about twenty years using data and AI to make complicated processes work better, mostly in large companies. Now I’m looking at where AI will speed up fundamental science and where it won’t, mainly in physics and astronomy.
Some parts of science are already moving faster with today’s AI. Some aren’t, and will need other kinds of work. I want to find that second set.
I’ve watched this happen for twenty years. The bottleneck is rarely the AI. It’s the part of the process AI can’t touch, and that part is usually human. Coding is the recent example. AI removed the coding bottleneck and it moved to writing better specifications and code review.
AI will take a while to move into the physical world, and longer to change how people and institutions behave. So it will help much more in some places than others.
Three guesses where this shows up.
- Getting people to agree.
- Nothing gets built until people approve it. AI can help write a better telescope proposal in a week, and the approval still takes years.
- Physical construction.
- AI in the physical world is far behind AI in language and math, so I don’t expect building to speed up soon.
- Nothing to measure against.
- Quantum gravity is the clearest example. No experiment we can run today tells us which theory is right, so AI has no answer to check itself against. Consciousness is the extreme version, where we don’t even have a measure to start from.
I’ve been exploring Convergent Research’s Fundamental Development Gap Map, which lays out the gaps a field needs to close. I am starting to break each gap into critical paths, so you can see where AI is accelerating a gap and where it isn’t, and whether more capacity would speed the field up. gapmap.sendorai.com
Two questions for you.
- How is AI helping your field, and where is it not helping?
- What investments, in AI or outside of it, would help your field move ahead so that AI can accelerate it?