Can AI invent a new element?
No. An element is defined by its proton count, so only nuclear physics can add one to the periodic table. What AI can do is design new materials and molecules from the elements we already have, and help physicists aim their experiments.
Elements: no. Materials: yes.Three things AI is actually changing
Helping find superheavy elements
Elements 119 and 120 have never been made. Physicists fire a beam of one nucleus at a target of another and hope a single pair fuses. Machine-learning models are being tested to estimate which beam, target and energy give the best odds, so scarce beam time is not wasted.
In 2024, Berkeley Lab produced livermorium (element 116) using a titanium-50 beam, a route intended to pave the way toward element 120.
Predicting millions of crystals
DeepMind's GNoME model predicted about 2.2 million new crystal structures, roughly 380,000 of which it judged stable. These are new arrangements of known elements, which may matter for batteries, chips and solar cells. They are predictions until someone makes them in a lab.
Designing molecules in days, not decades
Generative models propose candidate molecules for drugs, battery electrolytes and solar materials, then score them before anyone mixes a chemical. Microsoft and Pacific Northwest National Laboratory used this approach to narrow millions of battery-electrolyte candidates to a short list in a matter of days.
The model proposes. Real experiments decide.
The superheavy frontier
Select an element to see where it stands. Dashed tiles have not been made yet.
Pick a tile above.
Collision lab
New elements come from adding protons. Pick a beam and a target, and the atomic numbers add up. This is a simplified simulation, not real lab data.
Choose a beam and a target, then fire.
Test yourself
The takeaway
AI does not rewrite the laws of physics or add protons. It gives scientists a faster way to search the huge space of materials, molecules and experiments that already exist, and some of its predictions will turn out wrong. The skill is knowing which claims have been tested.
Sources of facts above: Google DeepMind GNoME (Nature, 2023), Berkeley Lab livermorium result (2024), Microsoft and PNNL battery study (2024). Verify current figures before citing.
