Somebody asks on Mastodon why LLMs haven’t solved cancer, why LLMs haven’t solved climate change yet. They argue that contrary to for instance nuclear energy that was a civil application of nuclear science that also gave us the atom bomb, LLMs seem to only have negative side effects.

Interesting question.

If LLMs can (help) solve millennium prize problems, surely they should be able to cure cancer too, right? Why haven’t they yet?

The question about climate change seems easier to answer: the solution is not in smarter AI, the solution is in all of us making sacrifices. Not taking a car, not taking a plane, not eating meat, not putting on the air conditioning, not using bitcoin, etc… we can blame a part of climate change on the data centers, but the missing solution to climate change is on us, not on them.

I can think of a few reasons we haven’t seen a cancer solution yet:

  • the medical field is not like the mathematical field where you can just present your solution on the internet, and everybody accepts it. Validation of a cancer treatment would take multiple years, and we don’t see a solution at the moment, but maybe we will in ten years time?
  • solving cancer is really solving a lot of different diseases. The individualized therapeutic vaccines we are seeing emerge today have been developed using machine learning techniques that are ten years old. The industry is only now starting to use the new and developing LLM techniques, who knows what they will find?
  • solving cancer diseases may be more difficult than the millennium prize problems? I am not knowledgeable enough in either field to compare.

LLMs have been largely used for tasks like helping programmers code faster, helping people write texts that are more correct, translating texts into languages they don’t understand, helping people solve millennium prize problems in mathematics, … These successes are certainly not negative.

In the end, suppose my spouse or my parents have a rare cancer, would I want to use an LLM to treat it? No, probably not. Would I want my heavily educated and trained cancer researcher to use LLMs to get to a solution ASAP?

I think yes… The early evidence we have seems to point to the fact that these LLMs can help humans solve bigger problems faster.