What Jim Carroll Saw Coming
Original archive: 2013-11-06 — Keynote: Sandia National Laboratories – Rethinking Innovation in a World of Accelerating Science.
In a keynote for Sandia National Laboratories about rethinking innovation, Carroll examined the challenge of adapting scientific and organizational practice to a changing world. This voyage explores how automated experiments and AI-assisted research might change the pace of discovery. Those methods should not be presented as exact historical predictions unless supported by the keynote record.
What has changed in the real world
Modern laboratories can generate far more possible experiments than researchers can physically perform. Algorithms may predict structures, search combinations and rank which trial is worth doing next. Automation can execute repetitive preparation and measurement, while scientists interpret whether an unexpected result is meaningful. The most interesting consequence is the feedback loop: a model proposes, an instrument measures, and the next proposed experiment changes accordingly. This is not the same as a machine discovering truth without human judgment.
Carroll’s exploration of accelerated scientific work anticipated an important institutional challenge: researchers have more data and more powerful tools, but still face finite time and equipment. The AlphaFold database offers an extraordinary example of computational scale, making predicted protein shapes widely accessible. That resource can guide biological investigation. It does not prove that a chemical works in a living organism or that a proposed medicine is safe. The physical experiment remains the arbiter.
A fact that changes the scale of the question
The AlphaFold Protein Structure Database makes over 200 million predicted protein structures freely available to researchers. The original source is EMBL-EBI / Google DeepMind. The distinction behind the number or capability matters just as much as the headline: A predicted protein shape is not a validated medicine, clinical trial outcome or automated discovery of every biological mechanism. The point of introducing the fact here is to show the actual scale of the machinery, knowledge, data or infrastructure involved in this particular voyage. It is not evidence that the imagined 2036 destination is inevitable.
2036: one possible destination
An imagined scene, not an account of an event that has happened.
Imagine a 2036 materials laboratory running overnight. An automated bench prepares a carefully chosen set of samples, measures the results and flags a surprising failure that the model did not expect. The next morning, a researcher asks whether the instruments were calibrated, revises the experiment and investigates the anomaly. The laboratory never stops collecting evidence, but it still needs people to challenge what the evidence means.
One image captures the possibility: An automated lab proposes the next experiment overnight; scientists choose what to believe. The conditions behind that image deserve attention. A future built around this capability would also change work, trust, operating costs or personal choices for someone outside the technology industry. That is why this voyage is less about predicting the arrival date of a device and more about understanding the environment that would let it matter. Different communities and businesses could arrive at the same destination at different speeds, or choose a different route altogether.
The detour that could change everything
An automated system can repeatedly make the same mistake if its instruments drift or its assumptions are wrong. Proprietary datasets may make findings hard to reproduce. Expensive robotics can be justified in some fields and uneconomic in others. Laboratories also need safety practices and human review when chemicals, pathogens or high-risk processes are involved. Faster work does not automatically mean better science.
There is no requirement that technical progress and adoption advance together. People may welcome one application and reject another. The most persuasive future is the one that survives comparison with cheaper, simpler or better-understood alternatives. This is also where the archive has value: an old idea can be supported in direction while the timetable, commercial model or surrounding institutions evolve in ways that could not have been obvious at the start. That history deserves to be told rather than edited into perfect hindsight.
The next chapter is still being written
Look for independently replicated discoveries where closed-loop systems reduce the number of physical experiments needed for a useful result. The evidence should include failures as well as wins. The breakthrough is trustworthy acceleration, not a claim that scientists have become unnecessary.
The practical test is this: Do not imply autonomous systems replace peer review or replication The question for the people making choices today is not whether the imagined 2036 scene is visually convincing. It is whether emerging evidence alters a real decision: where to invest, what to build, which safeguards to demand, and which assumptions should be checked again next year. Does accelerated experimentation shift science’s bottleneck from doing to validating? The answer is not locked in. That is what makes tomorrow a voyage worth taking.