AI in Science and Mathematics
The field of science, tech, and mathematics is advancing at an unprecedented pace today. Often, there is news that cogent solutions to old unsolved problems of math, physics and chemistry are being accurately found with the help of AI. Until the last few years, many top mathematicians and researchers were heavily skeptical about usage of AI in cutting-edge research. Traditionally, doing deep thinking, writing papers and collaborating with others has always been the way to go.
But all this is changing now. Even great mathematicians like Terence Tao have attested to and are actively promoting AI tools to do math faster and on a scale that was not possible earlier. That does not mean that the age of mathematicians is over and AI is going to take over very fast. Society will still need great mathematicians for a long time because despite AI being able to do math, Terence Tao points out that there are certain things it is still not good at such as:
- Coming up with creative ideas on unsolved problems.
- Understanding deep context that experienced mathematicians are able to do.
- Judging if a suggestion is good or bad.
What it is good at is:
- Doing repetitive mathematical or scientific tasks with great efficiency.
- Breaking big problems into tiny pieces so that multiple people can work on them at the same time.
- Verifying math proofs with great accuracy.
In fact, Terence Tao has categorically stated that AI should not replace mathematicians but should be used as a powerful assistant. Similarly, in big science projects such as CERN, James Webb Telescope, Vera C. Rubin Observatory, Google DeepMind, AlphaFold, weather forecasting, etc., AI is assisting tremendously in unprecedented ways such as:
- Searches for unusual signals that might point to new physics.
- Tracking hurricanes/typhoons and extreme events.
- Real-time detection of moving objects, supernovae, asteroids, etc.
- Discovering new crystal structures and materials.
- Discovered 2.2 million new crystal structures, including over 380,000 stable ones.
- Predicting how proteins interact with DNA, RNA, and other molecules.
What Remains
The common pattern that is emerging now is:
| Task | Who Does It Best | Why |
| Asking big scientific questions | Humans | Creativity & judgment |
| Handling massive data | AI | Speed & scale |
| Spotting patterns | AI | Tireless & consistent |
| Making predictions | AI (increasingly) | Fast approximations |
| Validating important results | Humans + experiments | Need for truth & understanding |
| Interpreting discoveries | Humans | Meaning & next steps |
Basically, the scientists need to stay in the loop to do course corrections. The benefits are real as discoveries are happening much faster, costs have been reduced tremendously and access to science has increased exponentially. Some of the key challenges that remain are availability of good training data, reducing hallucinations, significant human oversight, etc. Overall, the future has become more interesting, and we can expect more disruptive changes in the future.