Artificial intelligence in research

ChatGPT, does humanity still need science, or should we just ask you from now on?

“A lovely and important question – and quite clearly: Yes, humanity needs science. Absolutely. Even though I can answer many questions, I am ultimately a product of this science – and without it, I would not even exist. Feel free to ask me – but question, examine, and continue to investigate. I am a support, not a replacement. Humanity needs science now more than ever – and ideally in combination with responsible technology.”

[Reply from GPT-4o mini]

Research into artificial intelligence has gained tremendous momentum in recent years. The technology continues to develop rapidly and offers great potential for accelerating the analysis of large amounts of data and expanding the scope of application. However, this requires enormous resources in terms of computing power, data quality and quantity, as well as the security of AI systems.

The ability of artificial intelligence to quickly analyse large amounts of data, recognise patterns within it, and identify parallels and deviations opens up new avenues for gaining knowledge and accelerating research in almost all areas of research. But here too, the following applies: Possible misinterpretations or distortions must be taken into account and controlled by humans.

In medicine, the use of AI opens up numerous possibilities in the analysis of medical images, the diagnosis of diseases, the understanding of disease development, but also in drug development and the development of personalised medicine.  

AI-supported image analysis offers a wide range of applications, for example in segmentation. Segmentation involves automatically recognising and delineating specific structures or regions in medical images (such as CT, MRI, X-ray or ultrasound). The aim is to isolate relevant anatomical areas or pathological findings (e.g. organs, tumours, blood vessels) from the image. In clinical practice, AI can assist with this analysis. Neural networks learn to recognise patterns in image data, just as a trained human eye can recognise pneumonia on an X-ray, for example.

Even though technology is constantly improving, human control remains crucial when applying AI in medicine. The results of AI must be checked and critically questioned by humans. Initial approaches to strengthening trust in AI are being made with explainable AI, i.e. AI that can demonstrate how it arrived at certain assessments when interacting with humans (see also the chapter on “Technological developments in artificial intelligence”). Explainable AI has great potential, particularly in medical applications, to improve diagnoses and the evaluation of medical images.

The ability of AI to collate and analyse large amounts of data also offers great potential for personalised medicine. This approach aims to develop tailor-made therapies for individual patients. Individual factors are taken into account, such as genetic, molecular and clinical data, environmental influences and lifestyle. AI can help to evaluate this multitude of data and influencing factors and correlate them with each other. By evaluating large amounts of data, such as multi-omics data (i.e. combined biological data from different levels such as DNA, RNA and proteins) and clinical data, biomarkers can also be found – for example, for Alzheimer's disease or rare genetic disorders. In addition, scientists hope to discover new approaches to drug development.

In the life sciences, new AI tools such as AlphaFold have greatly advanced protein research. Proteins are large, complex molecules found in all living organisms. They perform a variety of vital tasks in the body – as building material for cells and tissue, for transport, for regulating metabolism, but also as enzymes or in fighting disease. Proteins consist of long chains of amino acids linked together by what are called peptide bonds. The sequence of these amino acids (the primary structure) determines how the protein folds and what function it has. The ability to understand and predict the three-dimensional structure of proteins offers completely new insights into the processes of life and the development of diseases.

AI offers numerous opportunities for biology as a data-driven science, but at the same time, the tasks of life science research also present AI with particular challenges. Biological questions often require working with relatively small data sets and set high standards for traceability, interpretability and efficiency. Addressing these challenges therefore not only benefits biology, but also drives the further development of AI itself, with the aim of creating more robust and trustworthy systems overall.

In astrophysics, too, AI has led to new approaches and insights through the analysis of large amounts of data. AI supports research into the universe, for example. Over the past ten years, there has been a revolution in gravitational wave astronomy: Hundreds of signals have been detected that originate from the merger of binary star systems in the universe, for example the merger of two stars or black holes.

As the discovery rate is expected to increase significantly in the future with new facilities on the ground (such as the planned Einstein Telescope in Lusatia) and in space (such as the ESA's LISA mission), the need for efficient analysis methods and precise models is becoming increasingly urgent. Deep learning methods can help to speed up the analysis of gravitational wave data.

AI is making progress not only in space exploration, but also in Earth observation. Just a few years after the emergence of the first simple AI models, weather services around the world now operate AI-supported models in parallel with conventional, physical-numerical weather forecasting models. These new models drastically reduce computation times and outperform conventional approaches in several important evaluation metrics. However, the question remains as to how these models, trained using historical weather data, will perform under future climate conditions. Here, AI could enable a better understanding of physical processes: When models analyse the extensive Earth observation data, they could identify characteristic features that influence the climate. This could potentially provide new insights into the physical processes that influence weather and climate.

In the natural sciences, the combination of quantum mechanics and artificial intelligence also offers completely new approaches to research. Using quantum mechanical models, researchers aim to apply the laws of the quantum world to real chemical and physical systems. These models can be used, for example, to predict molecular structures, understand chemical reactions, design new materials or research active ingredients in medicine. Here, too, the necessary calculations were based on the approximate numerical solution of the relevant physical differential equations (in this case, the Schrödinger equation), which is extremely complex and only feasible for systems with a small number of particles. Here, too, the new machine learning-based approach bypasses the fundamental physical formulas and extracts the solution from data about systems that have already been solved or measured. This leads to extreme computing speed increases.

The application of AI is not limited to the natural sciences. AI also offers new approaches to research in the humanities. It is used for the automated recognition of images and motifs in digital art history, as well as for the transcription of manuscripts – for example, in the correspondence of historical figures – or the classification of phenomena in literary texts. As in other areas, development here is also very dynamic. However, AI methods and tools are not only applied. Their scientific application and the social consequences of their dissemination are the subject of ongoing reflection. This reflection focuses specifically on the limits of AI and, for example, the question of whether large language models, as “stochastic parrots”, can interpret a piece of music or understand the meaning of a poem.

In archaeology, AI can facilitate the analysis of historical finds. Scientists create 3D scans of cuneiform inscriptions, for example, and use AI to analyse these 3D models. AI helps to decipher the wedges, which are often difficult to distinguish, or to detect damage and replace it digitally. AI tools enable faster and more accurate classification and digital reconstruction of 3D models of shards.

The rapid development of artificial intelligence also allows conclusions to be drawn about human intelligence. Advances in the development of large language models allow for interesting comparisons with neuroscientific findings on language development and association patterns in the human brain. Advances in artificial intelligence particularly challenge the assumption that the architecture of the human brain is the only solution for building systems with general intelligence, i.e. intelligence that is not focused on specific subject areas. However, AI systems have so far fallen far short of matching the energy and data efficiency of human thinking.

Published: September 2025