An Engineer in the Brain, a Neuroscientist in the Machine

I am a biomedical engineer who fell for the brain, and a neuroscientist who fell for the machine.

For most of my career those felt like two different people doing two different jobs. The engineer wanted to build things that work and to understand a system by taking it apart. The neuroscientist wanted to know how the brain does what it does. Lately the two have become one. That is the real reason I moved toward AI.

The science I kept coming back to

For years my work was about turning brain signals into something a person could actually read: EEG rhythms, MRI scans, measures of how the brain ages. Over time it narrowed toward language and cognition, and what happens to them when the brain is injured.

I care about that for a personal reason too. When I moved to the United States and started living in a second language, I felt my own speech thin out, and it still has not fully come back. In my first language I am quick and confident, sometimes funny. In English I am often none of those things, and it is strange to feel a part of me go quiet while everything I want to say is still there, unspoken. It made me pay attention to how the brain produces speech, not just how it understands it, and it is a large part of how I ended up studying aphasia, the loss of language after a stroke.

There is more to say about that, and it deserves its own post. For now it is enough to say that language, and how the brain builds it, is the science I keep coming back to.

Pointing at the same thing

This year I started working with language models, the systems behind the AI tools that are now everywhere. I did not do it to leave the brain behind. I did it because, for the first time, the engineering I was trained in and the science I love were pointing at the same object. A language model is an engineered system, something people designed and can inspect. It also does something that used to belong only to brains: it produces language. That overlap was too interesting to walk past.

Two things pulled me in, and I want to be honest about both. The first is simple. AI is a young, fast field. After years in neuroscience, where a good answer can take most of a decade, the pace felt like fresh air. The second matters more. In my current project I switch small parts of a language model off, watch how its ability to name things breaks down, and ask which internal piece was holding that ability up. The model is not the point. The point is to connect where a function lives inside a system to where its loss shows up in a person, and to let that inform how we help patients get their language back after a stroke. Engineering in the service of the science, and the science in the service of someone.

What AI is to me

These models play two roles in my work at once. They are tools that make the science sharper, helping me test ideas, analyze data, and design cleaner experiments. And they are systems worth studying in their own right, whose inner workings I want to explain.

It is the same curiosity I brought to the brain, pointed somewhere new.

The meeting point

I think this is a good moment to be standing here. Neuroscience and AI grew up in different buildings, with different vocabularies and different heroes. They are now close enough to lend each other tools and questions, and that distance keeps shrinking.

I did not plan to spend my days turning parts of a model on and off in order to understand something about people. But it fits the shape of what I always wanted, which was never to choose between engineering and science. It was to work in the place where they meet. Right now there is a lot of light on that place, and the work is only starting.

This is a personal reflection on how my research has shifted lately. The specifics of the work will come later, once they are ready to share.