Microscopy AI stops getting stuck on the same answers, starts finding rare materials instead

A new system lets automated microscopes explore broadly for interesting materials rather than narrowing in on whatever looks optimal first. In practice, this means researchers can discover unexpected materials and properties they weren't specifically hunting for, rather than having the AI converge on familiar answers and miss rare cases.

For years, machine learning in materials discovery has worked like a GPS with tunnel vision. Point it at an objective, and it finds the local maximum and stops, missing the interesting outliers. This system flips that: it balances hunting for what's good with hunting for what's unusual. That matters because the most scientifically valuable discoveries often live in the weird spaces the optimizer would skip entirely. Now a human can stay in the loop, steer the microscope, and actually find new things instead of getting better measurements of old ones.