THE ANIMAL HOUSE IS CLOSED

What do you think of rebooting the plasticity in adult brains to behave more like children&rsquo;s brains? Like, chemically?

That&rsquo;s what everybody would hope for, especially after injury.

What about technology improving to allow us to observe our own brains more regularly and in more detail? Like fMRI machines in our phones or something? Do you think spatiotemporal resolution will improve so much in the near future?

Well, you can&rsquo;t carry around an fMRI machine in your pocket, you need a 3-Tesla magnet. You can do EG, very lousy spatio-resolution, because you have all this volume production. You can plant electro chips. You do this with para- lyzed people for them to control a robot arm, for example. I am more on the skep- tical side.

Why?

First of all, we have not understood the essential principles of the brain. Silicon Valley people produce these good-looking machines and neural networks, and they have fantastic performances in classification, but that&rsquo;s it. Playing Go is nothing more than that. You just have to learn from examples. If you have enough time and enough speed, you iterate these trial and error things until you get a stra- tegy that&rsquo;s super good. So they outper- form us on particular tasks. My phone outperforms me when I do a numerical calculation. Let them do it. It&rsquo;s fine. Nice servants. I come from a time when I still used the ruler to calculate logarithmic numbers or tables.

So these tasks are abstractions of biological processes?

Abstractions, approximations, guess work. We don&rsquo;t really understand how the cerebral cortex does what it does &ndash; and it only takes 30 watts of energy. Compare that with what computers use, it&rsquo;s like a city in order to do the calculations that we can do in our heads. We have much to learn from it. They will have to learn from us as soon as we understand more and then try to implement those principles in... probably not silicon. My guess is that it will have to be another substrate, because much of this stuff is analog computing, and this you can&rsquo;t do well in silicon. So far they have no technical implementation of a clever learning rule. It all has to be calculated, embodied in a chip. So I&rsquo;m very relaxed. And I know I&rsquo;m in good company, because everyone who doesn&rsquo;t make big money with machines but who instead try to really get at the essence of what generative computing means, they share my skepticism.

We&rsquo;ve encountered a lot of optimism in tech. Do you think they&rsquo;ll catch up to your skepticism?

 
Certain computer people, those who really made the advances on the theoretical level before all the limitations came, now detect and realize that they get the same problems we have: it&rsquo;s the binding problem, the question of how you represent message relations, how you get a representation of a leaf on a branch from a tree in an environment. You have these many brackets, and in language construction you have the same thing. The way Google Translate does it is it compares the world literature in the ori- ginal language in the input with world literature in translations in the output. And they match it until it fits, but this is not how we do it. We try to get the meaning, we search the right vocabulary, it&rsquo;s a completely different science. And we call these functions generative functions. These machines cannot do it because they lack essential features of organization that we have. But unfortunately these are features that engineers hate. They hate the recurrent network.

Why?

Because it&rsquo;s not controllable. You cannot analytically analyze it, not mathematically either because it&rsquo;s too complex. It&rsquo;s too non-linear so you can&rsquo;t really predict what it&rsquo;s going to do. So it has this runaway kinetics that must be very well controlled or else it gets epileptic or it dies out. And all these problems make them find other solutions.

You could argue that airplanes don&rsquo;t flap their wings like birds.

But this is aerodynamics. The cognitive principles used by the brain, in my opinion, are still in some respects radically different from those used nowadays in supercomputers. It will take quite some time until we have done our job, and we can build little machines that only consume 30 watts, and start to behave a little bit like a fly. If you look at a mos- quito, and the intelligence of this mosquito &ndash; I&rsquo;m sure you&rsquo;ve tried to catch one at night &ndash; you start to admire these little machines: there&rsquo;s nothing in the artificial world that can approximate this in terms of energy efficiency and cuteness.

What are the challenges in the next ten years in neuroscience?

Cope with the dynamics and the complexity. We now have the tools, and this is really new in the field. Until we could look at more than one node of a network at the same time, people used to observe one node in different stages of the brain, one after the other. This precludes you from seeing relational constructs. You cannot clap with one hand. As soon as you start with this, you see relations, and you start to see what looks like noise, because A is always doing this and B is doing something else. As soon as you see that these two things are related, it&rsquo;s no longer noise. The more of these nodes you record simultaneously, the more you see that everything is coordinated with everything in a very subtle way.

So if you look at a single place it looks like noise, but if you look at many places it looks like a pattern?

We can finally do this. With modern technologies, optimal recording, we can look at thousands of nodes at the same time. We get this extremely high-dimensional data, you can&rsquo;t see anything when you look at it, it&rsquo;s just dots and curves &ndash; you can&rsquo;t make any sense out of it. So you need machines in order to detect patterns in there &ndash; machine learning &ndash; and you need mathematics to cope with these complex, high-dimensional vectors. It&rsquo;s not only vectors, it&rsquo;s trajectories, and the trajectory of vectors, because activity changes in time all the time. It must, and only because it does do we have a concept of time flowing. If it always stayed the same, time wouldn&rsquo;t move.

It&rsquo;s like we need new mathematics.

Yes. We need much more conceptual work to make sense out of the data. We can collect it much better than interpret it. New technologies have opened the field up. We are able to record from thousands of neurons at the same time, we have anatomical methods to see the whole network. You could eventually really trace it, but that doesn&rsquo;t really help you. What you see is complexity and very high-dimension dynamics. And somehow this goes well together. A complex system will develop such dynamics. The real problem now is what to do with all these facts &ndash; how to put them together, what sort of concepts do they develop, how to test them, how to make good predictions for further research, because obviously there&rsquo;s no point in just collecting whatever you get, as we said initially. You never know if what you have is a side effect not worth pursuing or if it&rsquo;s the real thing. Before you have a concept you don&rsquo;t know.

It&rsquo;s like the more we are able to observe the more we know how little we know.

This is exactly what my feeling is. 20 years ago, I thought I had understood more than I now know I understand today. There is a lot of progress but the insight into not knowing has grown more rapidly than the insight into knowing has.

What about neurological diseases? Do you have any insights into slowing the ageing brain?

There are different aspects. Obviously with degenerative diseases we get more and more a handle on the mechanisms, as well as the genetic causes. Therapy is a big problem. It&rsquo;s not easy to interfere with these processes. We know roughly what&rsquo;s going on but we are not yet able to stop it. That may change rapidly with technology, since we can really hunt down genes and manipulate gene expression, but we aren&rsquo;t there yet. But I do think that we will have a cure for certain degenerative diseases, whether it&rsquo;s Alzheimer&rsquo;s I&rsquo;m not sure. Maybe Parkinson&rsquo;s. ALS is about to be solvable &ndash; at least in the near future.

And what about psychiatry?

It&rsquo;s very different, because there we don&rsquo;t understand what the problem is, where it resides. All we know now &ndash; and we have a number of conferences on it, the Ernst Str&uuml;ngmann Conferences, which used to be in Dahlem. We had three or four on psychiatric diseases. The bottom line is the taxonomy, the diagnosis, is very coarse. What we call schizophrenia probably has a very different result, a very different mechanism to something else we call schizophrenia. It&rsquo;s probably very different diseases, and the same with autism and so forth. So we need a better classification and taxonomy of it before we can do systematic research. We have certain hyimpotheses of what&rsquo;s going wrong but if you look at all of them, they are not coherent yet. This is partly a reflection that we don&rsquo;t understand very basic principles of cortical functions sup- porting higher cognitive functions.

Does that also mean that there&rsquo;s no pro- gress pharmacologically?

Very little. All the drugs that we have nowadays were serendipitously discovered 50 years ago, with added modifications to alleviate certain side effects. There&rsquo;s no new principle so far. Lithium for depression. So the field is failing, and the field is searching for solutions, and the field doesn&rsquo;t quite know where they will come from. It&rsquo;s a big problem. We are helpless here.

What does lithium do for depression? I just watched Homeland recently, and Carrie&rsquo;s prescribed lithium &ndash;

It acts like sodium in the brain, in terms of binding, but it works in some patients because it changes excitability levels. But we haven&rsquo;t really come to grips with it. Deep brain stimulation has also been developed.

For Parkinson&rsquo;s?

We don&rsquo;t know how it works exactly for Parkinson&rsquo;s. It&rsquo;s been examined in animal experimentation, and there&rsquo;s a good concept behind it, and people have realized that when they got it wrong, when they stimulated places that they didn&rsquo;t want to stimulate, that it had effects on mood. So there was this revival of psychosurgery, which we had already condemned 50 years ago.

Is it bad for the brain?

Stimulation is thought to be reversible, but I doubt it, because if you stimulate the brain over weeks and weeks, it must change something. It&rsquo;s an active field, trial and error, ethically questionable sometimes, because these interventions, unlike prescribing a drug, are not sub- ject to the same ethical criteria as drug development.

So the FDA is something we should hold on to?

They require endless trials, double-blind and so forth, before treatments are approved. With deep brain stimulation it is enough, because it is a method, if the patient and the psychiatrist agree that they should have an intervention. If they find a neurosurgeon to do it, they can do it. They don&rsquo;t have to ask an ethical committee and so forth. And of course money is involved. It started in patients that are so-called helpless, who can&rsquo;t be helped pharmacologically. So desperate cases, who consent because they see it as the last resort. And if you look at what they do, they try here and they try there, and stimulate here and stimulate there. I was directing the Academy of Sciences for a while, asked to analyze the situation, stop it, and do what you have to do on ethical committees. You have to be there for a longterm examination of the development of these patients, follow them for a long time, and do it in a systematic way, and publish, and also publish negative findings. I hope that this will stop this hazardous, aleatoric playing around with brains.

We were in touch with DARPA-associated institutions &ndash; the Lawrence Livermore Institute in San Francisco. They&rsquo;re implanting these chips &ndash; electronic devices &ndash; in the brain, with the hope of being able to sort of control them remotely. It&rsquo;s not there yet, but proof of concept is. But for Parkinson&rsquo;s it seems to work.

There&rsquo;s something to it. I can imagine that certain forms of major depression can be treated that way. By stimulating the reward centers in the brain. But so far there is no canonical recipe. 
We did transcranial magnetic stimulation.

Ah yeah. We have these machines here.

The idea is that early artforms made by prehistoric humans &ndash; from the north and the south &ndash; resemble each other because trance states activate the visual cortex in the same way that TMS does. It sounds dangerous when you talk about it.

No, TMS is not dangerous. Maybe it can trigger epilepsy.

He just triggered it a little above the neck.

You see phosphenes. You&rsquo;d have to bring it higher up in the cortical area, and all of a sudden you&rsquo;d see faces appearing. Imagery. The question is why do we draw stick figures all the time? This might be a genetic imprint. Because the body scheme is so similar in all mammals. A head, a trunk, and four paws. Either you could take the stance that it&rsquo;s a very high degree of abstraction, or it is the most primitive representation of a mammal. That&rsquo;s always the discussion, right?

What do you think?

I think it&rsquo;s both.

What do you think about virtual reality?

Ah, great opportunity for art. I&rsquo;ve been to several symposia recently on the chances of using virtual reality and augmented reality to embed the observer much more in the piece of art. Because it can really absorb you completely, which looking at a painting can&rsquo;t as much. I know Daniel Birnbaum, the former director of the St&auml;delschule.

Yeah, I studied there. He taught a philosophy seminar.

Ah, I know him well. He is moving now from the modern art museum in Stockholm to a company that does virtual reality, because he wants to make this technique available to artists. It&rsquo;s certainly something that one should keep in mind. Cinema started to outperform theater to some extent. This will certainly replace the current video monitors in exhibitions.

In terms of simulating experiences, it can also work much better with emotions like empathy.

Yeah, yeah. Because you can fool the brain if you simulate the sensory evidence. You can also take a flight simulator at the airport. They have all the noise and the vibrations. 
There you can really embed it. I saw these pilots sweating. There were sitting in a simulator, and they really felt they had to do it right. So you forget very quickly that you are in a simulator.

We saw Star Wars in 4DX. The seat was rattling, and there was a plastic tube that tickled your legs, and water was squirted in your face. I put my jacket on, it was freezing.

VR becomes reality again. It capitalizes on the knowledge that the brain has about the world. You give it a few things to eat and it will reconstruct the rest.

&copy; Yngve Holen 2019. ETOPS IIII, edited by Yngve Holen and Mathew Evans.

Jaguar photographed by Raffaella Russi in the Mamiraua Sustainable Development Reserve, 23rd of August, 2015

OF COURSE NO ONE PAYS AMAZONAS FOR PROVIDING THAT FUNCTION

Yngve Holen

Recommandé pour vous

People

Qui était à la soirée Moët & Chandon x L’Officiel?

Fashion

Chaque look du défilé Michael Kors printemps/été 2022

Fashion

Jessica Chastain en couverture du numéro anniversaire de L'OFFICIEL

Woman

Sydney McLaughlin revient sur son chemin vers les Jeux olympiques

People

Moment d'or de Sydney McLaughlin

Woman

Elsa Majimbo pour Valentino: le livre "L'alphabet pour enfants et adultes"

Woman

Anna Andres: facettes des femmes modernes

Food

Ruinart, la plus ancienne maison de champagne, s'engage pour l'environnement