It may sound strange, but future Nobel Prizes, and other scientific achievement awards, one day might well be given out to intelligent machines. It could come down just to technicalities and legalities.
Swedish chemist Alfred Nobel established the prestigious prizes in his will, written in 1895, a year before his death. He created a fund whose interests would be distributed annually “to those who, during the preceding year, have conferred the greatest benefit to humankind”.
Nobel explained how to divide those interests in equal parts, to be given, “one part to the person who made the most important discovery or invention in the field of physics… the most important chemical discovery… the most important discovery within the domain of physiology or medicine”.
He also created prizes for the person responsible for the most outstanding work of literature and for the person who did the most to advance fellowship among nations, oppose war and promote peace (The Peace Prize).
What should we draw from the use of the term “person” in Alfred Nobel’s will? The Nobel Peace Prize can be awarded to institutions and associations, so could it include other non-human entities, such as an AI system?
Whether an AI is entitled to legal personhood is one important question in all this. Another is whether intelligent machines can make scientific contributions worthy of one of Nobel’s prestigious prizes.
Grand Challenge
I do not consider either condition to be impossible and I am not alone. A group of scientists at the UK’s Alan Turing Institute has already set this as a grand challenge for AI. They have said: “We invite the community to join us in… developing AI systems capable of making Nobel-quality scientific discoveries.” According to the challenge, these advances by an AI would be made “highly autonomously at a level comparable, and possibly superior, to the best human scientists by 2050”.
Such a milestone may be closer than we think. But it will depend on what we are prepared to consider as worthy scientific contributions. These can range from standard data analysis to generating whole new scientific explanations for observed phenomena. There is a whole spectrum between these two conditions, which is already being explored.
In a few weeks, the computer scientists Demis Hassabis and John Jumper of Google DeepMind will be presented with their Nobel medals (they won this year in the chemistry category). The prize was awarded for the development of AI that can predict the structures of proteins from the order, or sequences, of their molecular building blocks, called amino acids.
This had been a notoriously difficult problem in biology, with a history going back to at least the 1970s. But, in 2020, Hassabis and Jumper unveiled an AI system called AlphaFold2, which has enabled researchers to predict the structures of virtually all the 200 million proteins that have so far been identified.
The success of AlphaFold2 is no isolated case; there are analogous situations in other sciences.
In 2023, researchers from the Massachusetts Institute of Technology (MIT) used AI to discover a novel class of compounds that can kill drug-resistant bacteria. Then, in 2024, major archaeological discoveries – in South America and in the Arabian Peninsula – were made using machine intelligence.
Also this year, a study at Massachusetts Institute of Technology (MIT) tested the impact of AI in materials science research. It concluded that “AI-assisted researchers discover 44% more materials, resulting in a 39% increase in patent filings and a 17% rise in downstream product innovation”. The study found that these new materials possess relatively novel chemical structures and lead to more radical inventions.
There is even recent evidence that drug candidates discovered by AI may be of better quality than those discovered by traditional means.
Should we consider these as “scientific contributions”? AI generally makes such discoveries through a process of systematic screening of different possibilities. It’s a highly structured process that’s just the kind of thing we would assume machines are good at. But humans come up with scientific breakthroughs through the kind of innate creativity that a machine can’t emulate, right?
Well, without trying to diminish the roles of great scientists, systematic screening – this time carried out by humans – was involved in the discovery of artemisinin as an important antimalarial treatment, and the discovery of prontosil – a crucial antibiotic. These also led to Nobel prizes. So we should remember that tasks such as screening can make important contributions to science and are not something carried out only by machines.
So, can we imagine a machine going one step further, generating scientific hypotheses with a high degree of autonomy? Hypotheses are preliminary explanations for natural phenomena that can be tested by means of experiments. A hypothesis is a key stage in the scientific method, a kind of educated guess pending evidence from real testing. Furthermore, could the AI then go on to test its hypothesis and present the results to us in our own language?
It may surprise you to know that this has been attempted already, within the domain of computer science research. In August, an international research group demonstrated an AI system that was able to carry out a scientific investigation, and even write a scientific paper describing the results.
Read More: The unspoken rule of conversation that explains why AI chatbots feel so human
It seems very likely that AI will one day take an active part in scientific investigations. But will it be able to compete for Nobel prizes, perhaps as junior partners to humans? That remains to be seen.
Even if a machine could one day win one of the science prizes, the literature prize should remain safely in the hands of humans. Or will it too be opened up to artificial intelligence? A recent scientific study compared human reactions to poetry generated by machines and poetry produced by humans. Its main finding was that people cannot distinguish between them, and “AI-generated poems were rated more favourably in qualities such as rhythm and beauty”.
If there is a limit to what AI can achieve in what had been exclusively human fields of endeavour, we’re currently struggling to find it.
- is a Professor of Artificial Intelligence, University of Bath
- This article first appeared in The Conversation
