AI microscopes could become more useful research partners by learning which observations scientists consider worth pursuing. A study listed by ACS Nano on October 7 introduces an approach that brings expert feedback into autonomous experiments at the nanoscale.
Many automated laboratories select their next experiment by trying to improve a predefined numerical score. That works when the target is straightforward, but subtle patterns in complex materials can be difficult to reduce to one number.
The researchers instead let expert evaluations guide the system. From those judgments, the algorithm learns an underlying measure of scientific interest and uses it to direct subsequent microscopy experiments.
The team demonstrated the approach in investigations of ferroelectric thin films and their domain walls, where intricate polarization patterns can escape conventional scoring methods. Human knowledge therefore becomes part of the experimental decision process, helping the system choose what to examine next.
The work points toward automated laboratories that can accommodate research goals scientists recognize visually but struggle to express mathematically. Its scope remains specific: the demonstrations concern nanoscale materials investigations, while broader applications will require testing in other experimental settings.