AI can't automate science – a philosopher explains the uniquely human points of research

In line with the final trend of Inclusion of artificial intelligence in just about all areasResearchers and politicians are increasingly using AI models trained on scientific data to derive answers to scientific questions. But can AI ultimately replace scientists?

The Trump Administration signed an executive order announcing this on November 24, 2025 the Genesis missionan initiative to construct and train a spread of AI agents using federal scientific datasets “to check recent hypotheses, automate research workflows, and speed up scientific breakthroughs.”

So far, the achievements of those so-called AI scientists have mixed opinions. On the one hand, AI systems can process huge data sets and recognize subtle connections that humans cannot. On the opposite hand, their lack of rational pondering may result in unrealistic or irrelevant experimental recommendations.

While AI can support tasks which might be a part of the scientific process, it continues to be a good distance from automating science – and should never give you the option to achieve this. As a philosopher Those who study each the history and conceptual foundations of science see several problems with the concept AI systems can “do science” without humans and even higher than humans.

AI models can only learn from human scientists

AI models don't learn directly from the actual world: they need to be “tells” what the world is like by their human designers. Without human scientists overseeing the development of the digital “world” by which the model operates – the info sets used to coach and test its algorithms – the breakthroughs enabled by AI wouldn’t be possible.

Consider the AI ​​model AlphaFold. Its developers were awarded the Nobel Prize in Chemistry in 2024 for the model's ability to infer the structure of proteins in human cells. Because so many biological functions depend upon proteins, the flexibility to quickly generate protein structures for testing through simulations has the potential to speed up drug development, track disease evolution, and advance other areas of biomedical research.

However, as practical as it could be, an AI system like AlphaFold alone doesn’t provide recent insights into proteins, diseases or more practical drugs. It simply allows existing information to be analyzed more efficiently.

As philosopher Emily Sullivan put it, AI models must succeed as scientific tools retain a robust empirical connection on already established knowledge. That means the predictions a model makes have to be based on what researchers already know in regards to the natural world. The strength of this connection depends upon how much knowledge already exists on a given topic and the way well the model's programmers translate highly technical scientific concepts and logical principles into code.

Without this, AlphaFold wouldn’t have been successful existing body of human-generated knowledge about protein structures trained the model with the developer. And without human scientists providing a foundation of theoretical and methodological knowledge, nothing AlphaFold creates would have amounted to scientific progress.

Science is a uniquely human enterprise

However, the role of human scientists within the strategy of scientific discovery and experimentation goes beyond ensuring that AI models are properly designed and anchored in existing scientific knowledge. Science as a creative achievement derives its legitimacy, so to talk, from human abilities, values ​​and lifestyles. These, in turn, are based on the unique way people think, feel and act.

Scientific discoveries are greater than just theories supported by evidence: they’re Product of generations of scientists with diverse interests and perspectives working together through a shared commitment to their craft and mental honesty. Scientific discoveries are never the product of a single visionary genius.

For example, when researchers first proposed Double helix structure of DNAThere were no empirical tests that might confirm this hypothesis – it was based on the pondering skills of highly qualified experts. It took nearly a century of technological advances and several other generations of scientists to go from seemingly pure speculation within the late nineteenth century to a discovery recognized with the Nobel Prize in 1953.

In other words: science is one decidedly social companywhere ideas are discussed, interpretations are offered, and differences of opinion are usually not all the time overcome. As other philosophers of science have noted, they’re scientists more like a tribe as ““passive recipients” of scientific information. Researchers don’t gather scientific knowledge by recording “facts” – they create scientific knowledge through skillful practice, debate, and agreed standards based on social and political values.

AI is just not a “scientist”

I imagine that the computing power of AI systems will be used to speed up scientific progress, but provided that done rigorously.

With the lively participation of the scientific community, ambitious projects comparable to the Genesis mission might be of profit to scientists. Well-designed and thoroughly trained AI tools would make the more mechanical parts of scientific research smoother and even perhaps faster. These tools would compile details about what has been done previously in order that they will more easily inform how future experiments will be designed, measurements collected, and theories formulated.

But if the guiding vision for using AI models in science is to interchange human scientists or to totally automate the scientific process, then for my part the project would only turn science right into a caricature of itself. The very existence of science as a source of authoritative knowledge in regards to the natural world depends fundamentally on human life: on shared goals, experiences, and aspirations.