Searching for a new material is like looking for a needle in a haystack, but here the haystack is millions of possible atomic combinations. In the past, scientists spent years trying out options in the lab, relying on intuition and luck. Now, artificial intelligence is stepping in, learning to 'understand' the language of crystals.
The ElementsClaw system works like a well-coordinated team. One part, like a meticulous accountant, calculates the behavior of electrons in the crystal lattice using a specialized model with a billion parameters. The other part, like a seasoned expert, reasons about how suitable a material's structure is for superconductivity, using conventional language intelligence. Together, they combed through a database of 2.4 million crystals. In 28 hours of GPU time, they found 68,000 promising candidates — more than all known superconductors accumulated over decades.
The most exciting part came when the predictions were tested in the lab. Four new superconductors were synthesized from common metals: zirconium, scandium, rhenium, and hafnium. One of them, Zr₃ScRe₈, transitions to the superconducting state at 6.5 degrees above absolute zero: that's about -266.5°C. At such temperatures, even air becomes solid. Another, HfZrRe₄, becomes a superconductor at 5.9 K. The numbers are modest, but they prove that the machine has truly learned to 'see' invisible patterns in the interweaving of atoms.
One of the discovered superconductors, Hf₂₁Re₂₅, was predicted by the system as 'dormant' in the database: its structure was known, but no one suspected it could conduct electricity without resistance.
In Peter Watts' novel 'Blindsight', an alien ship uses superconductors to create powerful magnetic fields — perhaps someday our discoveries will lead to such technologies.