Tool-making is one of the defining traits of our species. The earliest stone tools, the Oldowan industry, date to approximately 2.6 million years ago, well before the emergence of Homo sapiens. Our ancestors were shaping the world with their hands long before they looked like us.
What is striking is that tool-making is not only about survival. It reflects foresight, planning, and the capacity to imagine an object that does not yet exist and then bring it into being. Humans are unique in how we build tools to make other tools, and then build on those, generation after generation. This cumulative culture is what eventually produced the wheel, the printing press, and the smartphone.
Each tool encodes the knowledge of everyone who came before.
The lineage of instruments
Each wave expanded what scientists could see
Chemistry has been transformed by successive waves of tool-making. Mass spectrometry, the technique at the center of our laboratory, sits within a long lineage of instruments, each of which opened a dimension that had been invisible the day before.
- Late 1700s
The analytical balance
Lavoisier turns chemistry into a quantitative science and ends the phlogiston era.
- 1860s
The spectroscope
Bunsen and Kirchhoff identify elements by the light they emit, and eventually read the composition of stars.
- 1953
X-ray diffraction
Franklin, Watson, and Crick visualize the structure of DNA.
- 20th c.
NMR spectroscopy
Molecular structures mapped atom by atom.
- 20th c.
Mass spectrometry
Molecules identified and quantified at attomole sensitivity, sorted by mass-to-charge ratio.
- Today
Multi-omics
Lipidomics, metabolomics, and proteomics map whole molecular systems at once, much of the chemical space still unannotated.
The newest instruments are made of mathematics and data
Predictive models such as AlphaFold for protein structure, and AI-collision cross section predictors, such as the ones we develop, represent a new kind of tool. They extend human perception into spaces too vast or too expensive to probe experimentally. The tool-making impulse is the same. It has simply been applied to information instead of matter.
When the instrument begins to think
The most compelling feature of the present moment is the blurring line between instrument and intelligence. Historically a tool was passive. A microscope showed what was there, and the scientist did the interpreting. The next generation of scientific tools will increasingly do the interpreting alongside the researcher. Self-driving laboratories are already operational at Argonne, the University of Toronto, and elsewhere, where robotic platforms design experiments, run them, analyze results, and decide what to do next.
The bottleneck shifts from running the experiment to asking the right question.
Closer to our own work, predictive models are beginning to close the loop with experimental instruments in real time, guiding mass spectrometers mid-run toward precursors that are likely to be biologically meaningful. Lipidomics and metabolomics are particularly ripe for this, because the chemical space is vast and most of it remains unannotated.
A new kind of instrument
Foundation models trained on the chemistry, biology, and physics literature are beginning to function as reasoning tools rather than lookup systems. Combined with cheap automation, better sensors, and democratized access through platforms such as Hugging Face Spaces, a graduate student in a small laboratory can now accomplish what required a national facility a decade ago.
The risks are real. Black-box models can mislead as easily as they illuminate. The laboratories that will thrive are those that treat AI tools the way Lavoisier treated the balance: with rigor, calibration, and a clear sense of what the tool can and cannot tell us.