Understanding Artificial Intelligence from a Philosophical Perspective

The group led by Florian Boge is concentrating on deep neural networks (DNN). These are used in machine learning, for example in programs such as ChatGPT. DNN are complex mathematical functions that could also be described as a rough imitation of a brain: They consist of various small units and layers that are interconnected and therefore often compared to biological neurons. These networks have helped to achieve significant breakthroughs in research in recent years, for example in protein structure biology: The AlphaFold2 DNN developed by Google’s DeepMind team, for example, can meanwhile predict protein structures on the basis of amino acid chains: A challenge that human researchers had been unable to solve for 50 years.

New ways of understanding

“In theory, DNN can solve almost any task – which means that in some cases they can already do more than humans,” says group leader Boge. “The problem is that we are sometimes unable to understand why DNN can do something and why they do it in a certain way.” For example, the networks are often faster, more accurate and above all more flexible than researchers would in fact expect. To track these processes and understand how important it is to make AI more transparent, the researchers in the new group are working with Explainable Artificial Intelligence (XAI) models. Here, they are pursuing two important questions: How do DNN learn and what do they learn? In this context, the group led by Assistant Professor Boge is focusing on philosophical knowledge gain: “The main goal of science is to understand. We want to learn how and why something happens and derive findings from this that we can apply elsewhere,” says Boge.

For science, AI therefore means a radical change on a philosophical level as well. “It’s important that we do not simply continue blindly with our research and rest on the breakthroughs in AI, but instead reflect on every success,” says Boge. “We need to understand science in a new way and re-think it. Understanding can mean finding an explanation. But it can also mean finding a new direction, a new structure. And that’s one thing which artificial intelligence does for us: It gives us a map, as it were, that we can use to guide us in unfamiliar terrain. If we additionally learn to better understand AI itself, it can also at least inspire us in our own search for scientific explanations. These are new ways of understanding that we must learn to work with.”

About Florian Boge

Florian Boge has been Assistant Professor for Philosophy of Science with a focus on Artificial Intelligence since the start of the 2023 summer semester. He is now pursuing his main interests at the Institute of Philosophy and Political Science: The impact of AI on scientific understanding, scientific (anti-)realism, the philosophy of models and simulations, and the foundations of quantum physics. Before coming to Dortmund, he was a postdoctoral researcher in the interdisciplinary Research Unit of the German Research Foundation “The Epistemology of the Large Hadron Collider”, especially in the sub-project “The Impact of Computer Simulations and Machine Leaning on the Epistemic Status of LHC Data”.

UDNN – Emmy Noether Independent Junior Research Group

Contact person for queries:

Back