Q&A with Sebastian Seung

What is a connectome, and why does it matter?

Look inside the mind with Sebastian Seung in this Q&A, originally published in Time. Interview by Maia Szalavitz.

First, what is a connectome?

It’s a map of all the connections between the neurons in nervous system and brain. An analogy would be the flight map in the back of airline magazines. Imagine if you had 100 billion cities and 10,000 flights from every city — a city corresponds to a neuron and a flight between two cities would be a connection between two neurons — then that would look like your connectome.

Are all connectomes pretty much the same?

Your connectome and mine are different.

What determines the differences?

There are differences because our genes are different, but even if our genes were the same, our connectomes would be different because they are influenced by experience, too.

There are many ways in which genes influence the brain. One is by controlling how it wires up when

we are in the womb or are babies. And that consists of a number of steps. A hundred billion neurons have to be born and they have to migrate to the proper place in the brain. They have to extend these branches that entangle with each and the branches have to make synapses [connections] with each other. Genes are involved in orchestrating all of these processes.

But there aren’t enough genes to tell all of them where to go and how to connect, right?

Genes set up a rough draft of the connectome and experience is required to refine that. The slogan of the book is that ‘you are your connectome,’ and that’s shorthand for the notion that all of your personal characteristics are somehow encoded as a pattern of connections between your neurons. One of the chief aspects of personal identity, for example, is your memories.

Is everyone’s neural code — what this wiring pattern means — the same? In other words, do my neurons encode “green” the same way yours do?

The neural code usually refers to how your current thoughts and feelings and perceptions are encoded in the signals that neurons are passing around — and it’s not the same. The code is not the same for every person. If it were the same between you and me, we would have to say that all neurons would be in one-to-one correlation but your code and my code might have similarities.

I might have a ‘Jennifer Anniston neuron’ [that responds to images of the actress] and you might have one, but you couldn’t say in advance exactly where those were in your brain.

The neural signals and activity in your neurons are supposed to encode your current experience, but your past experiences are supposed to be encoded in the connections and these connections are more stable. That’s a second kind [of neural code], and you asked about whether it’s the same for you and me. This is a great question and I hadn’t thought about asking it that way.

Let’s look at the example of bird song. If you look at [one] particular area, you should be able to figure out from the pattern of connections there how the neurons are activated when the bird recalls its memory of the song. That would be same for all birds, in some sense that would be a universal code.

So, if you knew my connectome and the neural code, could you read my mind? People make claims that brain scans can already do this, but they don’t seem very convincing to me because they know what people are listening to or looking at before they try to “read” it back.

We suspect that if we could measure the activity of every single neuron, we would be able to really read your thoughts, assuming we had the ‘dictionary’ [for the universal neural code]. In the case of [current brain scanning technology like] fMRI, because the images are so coarse, our ability to read your thoughts is also very coarse.

Imagine I’m really nearsighted. If I look at a book, maybe I can’t tell whether it’s Shakespeare or Dan Brown, but I might be able to tell a children’s book from an adult’s book [and that’s similar to what we can see with fMRI].

So, if we had the ability to know all the neurons and the code, could we also connect one brain to another and experience what the other person is experiencing, like in a William Gibson novel?

This is more in the realm of a thought experiment. You can imagine a really coarse one. If you are just looking at someone’s fMRI signal, that could used as a communication channel. The question is whether if a we had a really high-resolution activity map, could that be used to interface with you? It depends on the dictionary, you’d have to [know] the dictionary [and translate].

So how would you find that dictionary?

One way of finding the dictionary for perception is to show you lots of pictures and then see how your neurons get activated. You can imagine another method would be to find the connectome and guess from the connections which picture had activated each neuron [or network]. That’s a thought experiment. That really depends on answering this question of, Are you your connectome?

Would understanding the connectome be useful for treating mental illness?

The obvious challenge for connectomics is try to find abnormalities in the brain that are associated with the types of mental disorders for which no such abnormalities have yet been found. There are plenty of mental disorders for which there’s no clear and consistent pathology of the brain. Clearly in Alzheimer’s, we know something’s wrong about the brain but there are plenty of others for which nothing obvious [is wrong].

That’s basically the difference between psychiatry and neurology?

That’s a good way of saying it. People called them psychiatric diseases in the past because they were hesitant to call them brain diseases, given that you can’t see anything [that’s consistently] wrong [with people who have the same disorder]. There’s an interesting possibility that in some of these disorders, the neurons are perfectly healthy but they’re connected in an abnormal way.

There seems to be some evidence of that for autism.

Yes, there’s some evidence of structural abnormalities from studies of brain size. On average, autistic children have larger brain and head size than typical kids, but if you took every kid with a big head and said that they were autistic, that would be really inaccurate. There may be a structural difference but size alone is too coarse a measure to reveal it.

And usually, we claim that bigger is better in brains.

Yes, typically we think bigger is better and autistic people have slightly bigger brains. On the surface, that seems [strange] but it is also true that autistic people are better at certain things.

What are you currently studying and how can people help with that research?

Right now, we are focused on the retina. We are trying to find the connectome of this very approachable part of the nervous system. We have a website, Eyewire.org, where people can come help us map that connectome by playing a game that is very much like a coloring book.

We have a large number of high-resolution images of retinal tissue. In order to find the connectome, we have to trace out the branches of every neuron, because they’re like the wires, you have to trace them to find out how they’re connected. If one person were to color all that in manually, they could never finish in a lifetime, there’s so much image data. Computers can do part of the job but they make mistakes. If you come to this website, you guide the artificial intelligence (AI). The AI does most of work, but you guide it and hopefully all together, we can finish this in reasonable amount of time.

Eyewire is a citizen-science project [and is one way of asking], How can we use the Internet to make us smarter instead of dumber?

So will we one day be able to get new brains or “download” ourselves onto the Internet forever? And if we did, would that thing really be us or would it just think it was?

The general statement I would make is that regeneration of the brain is more complicated because the cells not only have to survive, they have to wire up to the rest of brain.

And that pattern is where our memories are stored?

Surprisingly, nobody really knows. One naïve idea would be that if you retain a memory for 20 years, some of the connections involved in that memory are stable for 20 years, but nobody knows.

It could be that there’s some overall pattern that remains stable even though individual connections are in flux. It’s amazing that you can pose these really simple questions that no one knows the answer to.

Answers by Sebastian Seung, Professor of Computational Neuroscience at MIT.