How did you come to choose AI as the topic for the Leopoldina‘s annual conference?
Thomas Lengauer: This is the third annual conference that I have organized. In 2009, it was about computer models in science, and in 2019 we had a conference on the phenomenon of time in nature and culture. For me, this was and is a fundamental question of all sciences. This time the topic was a natural choice. For me, the impetus came from two recent technical developments: the language model ChatGPT and the solution of the protein folding prediction problem with the AlphaFold algorithm. I used to work on the protein problem myself – and even months before AlphaFold, I would not have thought that I would live to see the solution. These are not continuous developments, something completely new is being created and we still don‘t really understand how it happens. We didn‘t have to convince anyone at the Leopoldina that these breakthroughs were important. And, of course, AI is also on everyone‘s mind in society in general.
Klaus-Robert Müller: For me, one of the important breakthroughs was that AI models based on large amounts of data can now create new knowledge, for example in quantum chemistry or mathematics. The system learns something that nobody knew before, and this is happening more and more often. And with techniques like “explainable AI,” you can actually watch the system learn.
Does this mean AI is outsmarting humans?
Müller: Not at all. These are supporting tools. Mathematicians can use them to find even better proofs, doctors are supported in their routine diagnostics and can also make better decisions about rare diseases if an AI tool is sitting at the table, so to speak.
These are positive effects of AI, but the conference will also discuss the potentially negative social consequences. Is this a clash of two different views, of two scientific camps?
Lengauer: We are not dealing with two camps at all. In my view, the age of techno-optimism is long gone. We all know that AI is an incredibly sharp weapon that can also be misused in catastrophic ways. So we have to show this side of the coin as well. The ethics of science is always a focus for the Leopoldina, and I can personally back every presentation given at this conference.
Müller: In the Leopoldina, we are always working together, not against each other. It’s a constructive cooperation where we try to study a phenomenon in its scientific and social depth and come to a shared position.
The technical development of AI is progressing very rapidly at the moment. Is an organisation like the Leopoldina agile enough to react to these processes?
Müller: The question is whether a policy institution or an academy has to proceed at the same speed as the tech companies? It‘s all about reflection and fundamental ideas. And those don‘t get better if they are fast.
Lengauer: The Leopoldina is definitely caught between the poles of depth of analysis and speed of reaction. This became particularly clear during the pandemic, when the Leopoldina presented an agile format comprehensively for the first time, called “ad hoc statements”. However, this was also due to the incredible acute pressure caused by the pandemic. Despite all the warnings and the scenarios of threat, I don’t believe that we are under the same kind of time pressure when it comes to AI today.
But for example, the European AI regulation was almost ready to go when ChatGPT came out. And then it had to be revised again to apply to the new technologies.
Müller: For me, it‘s similar to the field of cybersecurity. If someone claimed to have solved the cybersecurity issue once and for all, they would be laughed at. It‘s a process, you always have to adapt to new circumstances.
The new American administration has made it unmistakably clear that it does not want to set any barriers to AI.
Müller: It‘s always a bit tempting to pit the academic viewpoint against the industry viewpoint. But there are many people in the industry who also have ethical principles and want their products to be safe.
Professor Müller, you are working on brain-computer interfaces. Will we soon be able to connect our intelligence directly to an AI?
Müller: Definitely not. We are still working non-invasively and we won’t be plugging anything in anytime soon. One of our developments was, for example, to monitor the mental workload of pilots or car drivers via EEG, which has ulitmately led to these warning lights in cars that tell you to take a break. Related to invasive implants, surveys have been carried out among patients with amyotrophic lateral sclerosis (ALS) to see whether they could imagine using implants for communication. And they tell us that the brain is the last functional organ that they have – so please don‘t mess with it!
How is your collaboration and the cooperation with the Leopoldina going in preparation for the annual conference?
Lengauer: We are well on track. The two of us discuss the event regularly and react quickly to problems, and we have great support from the Leopoldina. The meeting is ready to go!
The interview was conducted by Christoph Drösser.