We are talking to Dr. Mihai Petrovici, a physicist and computational neuroscientist at the University of Bern. He leads a team that is working at the intersection of biological and artificial intelligence.
Dr. Petrovici, could you tell us a bit about what you're working on day-to-day, what your goals are, and what gets you out of bed in the morning?
The questions we have are some really big ones. What are the brain circuits that ultimately make us who we are, and that give us our cognitive capabilities, such as intelligence? In our day-to-day work we start by addressing a bit more modest questions. We’re looking at particular skills that humans have, for example, recognizing an image, or a song, or another person.
I’ll give you an example. When you asked for this interview, I took a look at your website and saw a picture of you. I must have seen your picture maybe two or three times, and I looked at it each time only for a couple of seconds. Still, I'm pretty confident that I could recognize you, if I saw you on the street. Compare this to the image recognition networks that are typically used by giants in the field, like Facebook and Google. They need millions of pictures to tell apart a dog from a car. So clearly, we have an advantage there, but how?
Looking to the future, what sort of goals do you have in mind? What would success look like to you, in terms of practical applications of your work?
I think that we would measure success along two dimensions. One of them would be to better understand how the circuits of the biological brain operate. For example, to correlate the neuronal signals that we observe like particular sequences of spikes, to cognitive phenomena such as recognizing a particular pattern.
On the other hand, we're also interested in transferring these ideas and results to artificial systems, their algorithms and hardware. Here, one measure of success would be to observe how these brain-inspired algorithms, instantiated in the artificial system’s hardware, allow the hardware to perform better than existing solutions. This would be measured along various metrics, such as speed or power consumption, or just pure computational accuracy or strength of these artificial systems when they’re solving a particular problem.
What sort of common questions do you get asked by lay people when you describe what it is that you're working on? Is there something that people constantly demand to know from you?
When I give talks for the public, people find it fascinating that the machines that we are building, which use completely different physics than the physics of the brain, ultimately are capable of carrying out very similar computations as in the brain. People then often tend to ask: do these machines have consciousness like us? And then the discussion quickly goes into the direction of artificial general intelligence, and of the hopes and fears that people have in relation to this topic.
We don't have an artificial general intelligence at the moment. Predictions about the future are difficult. Perhaps it will be a matter of decades before there is such a system, but sometimes things work out faster than we expect, as we have observed over the last decade, and sometimes slower. I think we still have some way to go. That’s mainly because the systems that we have right now are very specialized. They're extremely successful, even superhuman, at particular tasks such as playing games. Other capabilities like safe automatic navigation of vehicles, and aid in diagnostics for medical purposes (which may be even better than experienced doctors) all will hopefully get there soon.
But what is missing in all this, is the general in artificial general intelligence. The capability of these systems to perform multiple tasks, and to transfer their knowledge and experience from one task to another. This is something that is very difficult to achieve, because the circuits of these systems are by construction very specialised.
Could you tell me what the added value of the Human Brain Project has been for you? How was the interaction with the other members, what were the key standouts?
I think it's very important, especially in our field, to have continuous interaction with experts in different subfields. There are a lot of components to the big quest of understanding the brain. You need to understand molecular biology, you need to understand the circuitry of the brain, you need to understand systems biology, and you need to be able to simplify all of these insights into mathematical models. This is also where the contribution of physicists come in. One of the core aspects of physics is to observe phenomena and then, with help of mathematicians, put these phenomena into a simple set of equations, which can then predict how they will evolve in the future.
The Human Brain Project is unique in offering a forum of expertise by bringing people from all these various disciplines together, and encouraging them to communicate with each other. You could argue that communication with other researchers could be easily achieved otherwise. Clearly, I could just make a quick phone call and have a video conference with any expert in any field. But being together in a room makes a huge difference. That’s how strong collaborations arise, and that’s how, I would say, quite a few beautiful results have been produced.
Listen to an extended version of this interview below: