Dr. Adrien Courtois

When did you first become interested in AI?

My interest started when I was in engineering school —AI, and especially Deep Learning, were not as big of a thing then as they are now. Large Language Models (LLMs) didn’t exist. I saw a lot of promise in this field, and as I love riddles, I couldn’t stop thinking about this challenge, one that even math couldn’t fully explain.

What problem are you trying to solve with your research?

Deepfakes are very easy to spot when you already know about the generator, but new generators are challenging. At GetReal Labs, we’re trying to determine what makes a deepfake a deepfake, and what is the nature of the traces they leave so we can design robust methods. This involves delving into the inner workings of deep neural networks and building upon our expert knowledge of traditional computer vision techniques and natural images.

What are the biggest breakthroughs or challenges in this area?

Anyone with the right data can design a deepfake detector that performs well against known attacks. What we aim to achieve at GetReal Labs is developing a robust detector that can combat attacks that have never been seen before. AI methods are getting easier to use, and malicious actors now have the knowledge to build their own deepfake generators. It is critical that we’re able to detect any kind of deepfake generator.

How could your work change the way people interact with digital media?

I hope the products we build and the cybersecurity services we provide will one day be incorporated into all social media platforms, where vulnerable people always believe every piece of content they see, even when it’s AI-generated. The work we’re doing today can help regulate generative AI technologies.

What’s the biggest lesson you’ve learned as a researcher?

Trying to understand and guess what a deep neural network does or learns is impossible. Only principled experimentation protocols and strong evaluation protocols can help you understand these tools. In that sense, AI research is extremely close to physics research: the only way to truly understand AI is by observing it.

What do you do outside of research? Any surprising hobbies or passions?

I am a semi-professional table soccer player, or at least I try to be. I train as hard as I can, and I am currently ranked in France’s top 100.

What’s one piece of advice you’d give to aspiring researchers?

What matters is not writing the best paper or determining the best method that will revolutionize the world. What matters is the knowledge you get along the way. A well-done experiment that you understand everything about is worth much more than the latest trendy paper. In the end, trends fade away, and the only thing that remains is what you’ve learned.