The science of our age is computational. Without models, simulations, statistical analysis, data storage and so on, our knowledge of the world would grow far more slowly. For decades, our fundamental human curiosity has been sated, in part, by silicon and software.

The late philosopher Paul Humphreys called this the ‘hybrid scenario’ of science: where parts of the scientific process are outsourced to computers. However, he also suggested that this could change. Even though he began writing about these ideas more than a decade ago, long before the rise of generative artificial intelligence (AI), Humphreys had the foresight to recognise that the days of humans leading the scientific process may be numbered. He identified a later phase of science – what he called the ‘automated scenario’, where computers take over science completely. In this future, the computational capacities for scientific reasoning, data processing, model-making and theorising would far surpass our own abilities to the point that we humans are no longer needed. The machines would carry on the scientific work we once started, taking our theories to new and unforeseen heights.

According to some sources, the end of human epistemic dominance over science is on the horizon. A recent survey of AI researchers offered a 50 per cent chance that, within a century, AI could feasibly replace us in every job (even if there are some we’d rather reserve for ourselves, like being a jury member). You may have a different view about whether or when such a world is possible, but I’d ask you to suspend these views for a moment and imagine that such artificial superintelligences could be possible eventually. Their development would mean that we could pass over the work of science to our epistemically superior artificial progeny who would do it faster and better than we could ever dream.

This would be a strange world indeed. For one thing, AI may decide to explore scientific interests that human scientists are unincentivised or unmotivated to pursue, creating whole new avenues of discovery. They might even gain knowledge about the world that lies beyond what our brains are capable of understanding. Where will that leave us humans, and how should we respond? I believe we need to start asking these questions now, because within a matter of decades, science as we know it could transform profoundly.

Though it may sound like the stuff of science fiction novels, Humphreys’s automated scenario for science would be yet another step in a centuries-long trend. Humans have never really done science alone. We have long relied on tools to augment our observation of the world: microscopes, telescopes, standardised rulers and beakers, and so on. And there are plenty of physical phenomena that we cannot directly or precisely observe without instruments, such as thermometers, Geiger counters, oscilloscopes, calorimeters and the like.

The introduction of computers represented an additional step towards the decentring of humans in science: Humphreys’s hybrid scenario. As one prominent example documented in the book Hidden Figures (2016) by Margot Lee Shetterly (and subsequent film), the first United States space flights required computations to be done by human mathematicians including Katherine Johnson. By the time of the US lunar missions, less than a decade later, most of those computations had been passed off to computers.

Our contribution to science remains critical: we humans still call the shots

The following decades witnessed continual, logarithmic growth of computational processing and power, and a correlated decrease in the price of computation. We are now at what we might call an advanced hybrid stage of science with an even greater reliance on computational systems. As one example, the philosopher Margaret Morrison explained how computational simulations were essential to the discovery of the Higgs boson – helping scientists know what to look for and sorting through the data from high-energy collisions.

And now AI has begun to have a large impact on science. AlphaFold, for example, is an AI designed to help predict how proteins will fold, given their chemical makeup. While humans can do this work independently of computers, it is time-consuming, labour-intensive and expensive. The creators of AlphaFold – Google DeepMind – claim that it has saved ‘hundreds of millions of years in research time’. Similar benefits can be seen across the sciences: the analysis of extremely........

© Aeon