I still remember those few days in 2023 when LK-99 looked like it might actually be a room-temperature, ambient-pressure superconductor.
For a brief moment, it felt like we were watching one of those discoveries that could genuinely change civilization.
Obviously, LK-99 didn’t survive replication.
But AI has advanced enormously since then.
We now have models that can reason across scientific literature, generate hypotheses, write and run code, analyse experimental data, predict structures and materials, and increasingly interact with automated labs.
So I keep wondering:
Could AI significantly increase the probability of discovering something like a real LK-99?
Not necessarily superconductivity specifically, but a breakthrough material or physical discovery with enormous technological consequences — something humans might have needed decades to stumble upon otherwise.
It seems like materials science could be particularly well suited to this: huge search spaces, lots of existing experimental data, simulations, and relatively clear ways to test candidate materials.
Maybe the real revolution won’t be AI directly “discovering a new law of physics”, but AI exploring millions of plausible hypotheses and pointing human researchers toward the 10 experiments actually worth doing.
How close are we to that?
And what would be the best candidate field for an AI-driven “holy shit, this changes everything” discovery: superconductors, batteries, catalysts, fusion materials, drugs… something else?
I want those three LK-99 days again.
But this time I want day four to be even better.
[link] [comments]