Viewpoint: Will AI help Big Tech dominate science?

Since the advent of large language models such as GPT-4, artificial intelligence has caught the eye of the public and policy makers alike. The investment boom into the development of ever more advanced AI systems is breathtaking. However, only a very few global actors like Microsoft, Google and Apple may end up dominating the market, due to their scale and market power. How this will shape the future of science, the source of AI’s power, has only begun to be discussed.

Cases in which algorithms based on machine learning have been able to solve problems like protein folding are well known. But what will be the result when advanced AI systems are sourcing all relevant scientific literature in a single field like medicine? In this fashion, a single platform could become an ‘intelligent’ scientist that has acquired knowledge beyond the reach of any human. One can easily imagine that this AI-acquired knowledge can generate new cures and medical treatments, that may enable patenting. This development will help science and the scientific community to advance knowledge and possibly support new avenues of research.

But who will own this AI-created ‘meta science’ and ‘meta knowledge’? Who will reap the benefits? Will it be in the hands of a few private companies, or do we need to safeguard our acquired knowledge, such that it will benefit society and the public as a whole?

Those are questions policy makers must start considering, for the impact of AI on science will be tremendous. On one side, we encounter the use of AI/ML-supported algorithms that are capable of solving specific problems hitherto beyond the reach of human scientists. Much of scientific development occurs in incremental steps, implying that accumulated previous knowledge paves the way for future development. At the same time, the tremendous accumulation of scientific knowledge today is such that individual scientists cannot possibly keep up to date in their own fields. Advanced language models and AI-supported analysis of scientific data and knowledge can be a great help to scientists designing their own research. As reported recently in Nature, it is already being used by many researchers for literature searches. Evaluators and referees will also have new tools at their hand to judge proposals and the value of publications.

The new ‘meta’ science

But an obvious question for the use of AI-supported scientific development is ownership. The most advanced AI systems will be at the hands of a very few actors that may claim ownership of this ‘meta’ science – in other words, the science of how science is conducted.

Given that much research now is published as open access, there are few barriers to penetrate and gather relevant information to create AI-based meta knowledge. Although many research publications are protected by copyright, it will not be difficult to either circumvent it, or access the information in different ways. It may even become impossible to directly trace the sources of AI-generated new knowledge, rendering the notion of copyright somewhat obsolete. And in practical terms, accessing the scientific literature will never be a question of copyright so much as of technical capacity – the software tools, databases and analysis. For that, big developers like Microsoft and Apple, with their massive R&D budgets and global market dominance, have a head start.

We can be certain that in the footsteps of the rising, privately developed AI platforms, meta knowledge and meta science will emerge. Step by step, these platforms will penetrate medicine, material science and eventually the entire world of natural science and humanities.  We are on the verge of science powerhouses owned by one, two or three companies, and possibly some governments.

Unless the scientific communities and public bodies create policies for the use of AI-based scientific meta knowledge, and unless they regulate access to the scientific literature, we may end up in a very precarious position. Global actors could use their leadership in AI technologies to dominate scientific development, based upon their priorities, far from public control.

What is to be done?

So far, governments have only started to consider such issues. In Brussels, the new head of the European Commission’s research directorate-general this summer developed plans for a unit dealing with AI-in-science issues, but details are not yet public.

Still, given the regulatory power of Brussels, there is plenty of room to create a specific task force with the science community that can take up that role. The EU is probably the only global actor that can assure that science remains within the domain of the public and for the public.

In my view, a few guidelines are key to safeguard both AI-powered scientific advance and to warrant its service to all citizens.

1. Initiate the creation of an independent platform with the task to develop meta knowledge and meta science.
2. Ensure that services from the platform are open to the science community, the public at large and commit to open science and open innovation.
3. Create a regulatory framework for how private actors may access and use the bulk of scientific literature.

Ramon Wyss is professor emeritus in theoretical nuclear physics at Stockholm’s KTH Royal Institute of Technology, and former vice president of the university.