In health research, artificial intelligence can do more than answer questions: it can estimate biological age, suggest research ideas and help interpret personal health data. At the closing panel of the Portfolio Deep Tech 2026 conference, participants discussed those opportunities alongside the technology’s limitations and its effects on human thinking.
One of the panel’s closing observations was that people should consider collecting their own health data now, including information about their eating habits. Such records could later become important for personalized advice.
Biological age and autonomous research ideas
Csaba Kerepesi, a researcher at HUN-REN SZTAKI, said automating scientific work offers substantial potential. In his view, an AI researcher could speed up science by as much as a factor of 1,000. His own goal is to extend the period of life spent in good health, and he would use tools that outperform humans at certain tasks to help achieve it.
His research group uses machine learning to estimate biological age from sources including laboratory results, medical histories and facial photographs. Using follow-up data, the researchers examined whether people identified by the models as aging faster had a higher risk of death. Such methods could support personalized medicine and the study of interventions targeting aging.
The group is also testing autonomous research. In one experiment, an AI system reproduced the analysis in one of the researchers’ studies and then proposed a new direction: examining indicators associated with individual organs rather than estimated whole-body aging in athletes. Kerepesi called the idea promising, but said the results were still being verified.
Fast searches and broad knowledge
András György, a senior researcher at Google DeepMind, traveled from the United Kingdom to attend the conference. He said machines are strong at processing large volumes of data and rapidly surveying possible solutions. If a system learns which steps can lead to a result, the speed of its search can give it an advantage in areas such as games or mathematical problems.
He was more cautious about originality. In his view, models mainly build on existing ideas, but can draw on a broader body of knowledge than an individual person. That can allow them to connect a method already used in another discipline to a problem whose researcher was previously unfamiliar with it.
A health assistant built on personal data
Longevity researcher Bertalan Thuróczy used his own experience to illustrate how AI can complement efforts to maintain health. He feeds a system he calls his “third brain” with physiological data, medical histories and journal entries. Alongside data from sensors measuring sleep and activity, he uses blood glucose readings, genetic information and microbiome test results.
Combining those sources can enable more personalized responses about nutrition, exercise and stress management, for example. Thuróczy stressed that the system does not replace a medical consultation, though it can complement one. He sees its value in its ability to process information about lifestyle and health together.
The user’s role and the limits
Dávid Farkas, co-founder of Principle Zero, said AI’s effectiveness cannot be judged solely by the tool’s capabilities: it also depends on collaboration with the user. Rather than handing over entire packages of tasks, he favors an iterative approach that breaks work into steps.
He also warned that outsourcing too much thinking could affect people’s own skills. In his view, users therefore need to learn not only how to use the technology effectively, but also when it is better not to use it.
Other panel topics included the hard-to-predict societal consequences of AI development, the difference between problems humans choose and tasks assigned to machines, emotional intelligence and machine analysis of emotions, and AI-based evaluation of cellular rejuvenation experiments. The discussion was moderated by Márton Baranyi, project manager at Portfolio Csoport.



