Toygar Tanyel

I am a visiting researcher at the Centre for Industrial Software, University of Southern Denmark, in the group of Serkan Ayvaz, where I work on explainability and the philosophy of machine learning. I studied computer science at Yıldız Technical University, then did a master's at Istanbul Technical University, where my thesis, supervised by Ilkay Oksuz, was on counterfactual explanation.

I was already publishing on explanation when I joined Hevi AI, and there it became a practical need. I built the AI stack in hBreastMMG, their breast-cancer detection and diagnosis product for hospitals, which is part of the first radiology AI suite from Turkiye to be CE marked under the MDR. A radiologist who acts on an explanation needs it to be more than convincing. These days I design and build the multi-agent stack at Promake.

Everything I work on comes back to two questions. What does it take to understand something, and what does it take to explain it to someone else? I look for answers in cognitive science and the philosophy of language. What makes an explanation more than a description, what does it need from the person listening, and why does it sometimes leave behind only the feeling of understanding?

Now AI systems explain themselves. What they give back is an account, not a record. An account does not have to be true of what it describes, and it may not survive someone pushing on it, so this is a question of trust and security too. These systems also take over the very work that used to make a person good at something. The person stays, watching, and still looks like someone who could do it. It is the same failure again. The understanding leaves, and the look of it stays.