by Rodion Shishkov
Most of us do experimental research to test a hypothesis or characterize a sample. We design a measurement, acquire data, and analyze it to decide whether the evidence supports our goal. Techniques steadily improve—faster detectors, sharper optics, smarter software—but the hardest problems often remain: heterogeneous samples, tight time budgets, and the gap between a few striking images and results that are statistically solid.
Automation is a powerful lever, yet truly automatic experiments are rare. Routine protein crystallography at synchrotrons is a notable success because the workflow is standardized: mount crystal, center, collect, repeat. It resembles an assembly line and needs little judgment about the sample’s future value. By contrast, “full autonomy” would mean giving a system a sample and a scientific objective, and having it plan measurements across many options to reach that objective. That level of generality matches what people call artificial general intelligence (AGI); we are not there yet.
What works today is multi-technique thinking. Pair a well-established, accessible method with a second technique that has high potential but practical limits. In our lab, we combined optical fluorescence imaging of cryogenically fixed human cancer cells with nanoscale X-ray fluorescence (XRF) microscopy. Two neural networks first analyzed the fluorescence data: one segmented cells, and another ranked regions of interest (ROIs) worth the costly XRF scan. The XRF instrument then auto-navigated to those ROIs without operator input. This fusion preserved the strengths of both methods—speed and context from optics; elemental sensitivity from XRF—while lifting a major constraint: throughput. Most importantly, it shifted our conclusions from “a few compelling examples” to statistics drawn from large, representative datasets.
My message is simple: when you feel stuck, look sideways. Ask which complementary technique could guide, filter, or accelerate the one you already use. If no such pairing exists, try to build it; the act of linking methods can become a new methodology in its own right. Multi-technique design is not a detour—it’s often the shortest path to reliable, relevant science.