The big puzzle

Published in Punkt Magazin, Friday, 21.04.2017, Text by Pascal Hügli & Rino Borini, Photos: Boris Gassmann

Everyone talks about artificial intelligence – and yet, it doesn’t even exist, according to Pascal Kaufmann. He doesn’t mean to provoke or to put a halt to the general euphoria: the neuroscientist only wants to steer research into the right direction. In his opinion the top priority should be to crack the brain code. Only once we understand how the brain works can we create artificial intelligence that deserves its name.

Pascal Kaufmann focuses his interest on the interface between technology and the human being. Working along the principle of neural networks, his company, Starmind, wants to map the existing knowledge reservoir within a company and make it available to employees. It is a type of learning real time Google that will always find the person who can best answer the question. In the next step the network is to be made available also to individuals outside of a company: They, too, should find the best possible answers to their questions – in the future even before they are even aware of their questions. The 39-year old Zurich native was inspired by his studies of biology at the University of Zurich and the many years of research work at the Laboratory for Artificial Intelligence, where he worked on the humanoid robot Roboy in collaboration with Rolf Pfeifer, who in the meantime has retired. In 2010, Kaufmann founded Starmind together with Marc Vontobel in the town of Kuesnacht near Zurich. In 2013 the company received the Swiss ICT Award as the newcomer of the year. In the meantime, Starmind's services are in demand by a whole range of customers from different industries in more than 40 countries.

Mr. Kaufmann, as a neuroscientist you are much in demand. Hardly any other subject stimulates people’s imagination as much as artificial intelligence. Why is that?

Artificial intelligence is indeed a dream long entertained by mankind. It is interesting that “artificial people” have been around for centuries, in the form of dolls or robots – however what they have always been missing is life, some would call it a soul. It is that indefinable something that makes a human being a human being. The reason why everyone talks about artificial intelligence today is because many people feel that we are just at the threshold of discovering that indefinable something.

What exactly is artificial intelligence?

Intelligence has no consistent definition; everybody has his or her own definition. At the Laboratory for Artificial Intelligence at the Zurich University our motto was: Machines have become intelligent if eleven of them can challenge a human soccer team and the robots win the match. In soccer the machines would have to work as a team, tackle the unforeseeable and, if necessary, be able to break the rules. The machine should possess all facets that define a human being. Only then we consider a machine intelligent.

So intelligence is always the entire package?

If you only look at certain aspects a machine will invariably be superior to a human being. A microscope will be much more sharp-sighted than a human eye. A car will always be faster than human legs. And a computer computes faster and better than the human brain. You can take any facility of a human and improve it artificially. But to unite all aspects in an artificial being – that has not been managed so far.

When Google, IBM or Uber talk about creating artificial intelligence (AI), then they always have only one of these aspects in mind?

That is right. In March I attended the Start Summit of the University of St. Gallen together with lots of AI experts. At the start of my presentation I wanted to clarify something from the get go: “Do we all agree that artificial intelligence does not even exist yet? If you advertise your products with the moniker artificial intelligence it is just a sales pitch. Actually and in reality your products don’t possess any artificial intelligence.

You must have been booed off the stage as the kill-joy of the year…

Not really. The participants on the panel even agreed with me.

Interesting. But why then is the term artificial intelligence such a compelling sales story?

What we term artificial intelligence today is only the intelligence of the programmer preserved in the form of codes. He is the developer who packages his intelligence into a source code and that code is then presented by smart sales people as artificial intelligence in order to win investors and customers.

So what has to happen for us to really talk about artificial intelligence? We must decode the principle by which the brain works, i.e. crack the brain code.

Is that even possible? I don’t see any reason why the brain should be indecipherable magic; even today various organs are manufactured artificially. I am convinced that the day will come when the brain code is cracked. This will be one of the most important discoveries ever made by mankind.

Are there indications that this might soon be the case?

I believe that 95 percent of all AI investments flow into so-called brute force methods, by which computer processes are made faster. It is as if one wished to build an artificial bird and instead was throwing stones into the air. What you notice is that the stone, too, flies. So you just keep on building larger and larger canons in order to catapult the stone into the air even faster.

So instead of hurtling stones into the air it would be smarter to study the principle of how a bird flies.

Exactly. One would, for example, have to examine the shape of an avian wing or the principles of aerodynamics more closely. In brain science, however, comparatively few scientists deal with the principle of the brain code. Similar to the example with the stone the focus is too much on computers. If, instead, work would be done on the “wing profile of the brain,” science would arrive at a break-through much faster. The computer brain analogy was not very helpful in furthering the understanding and knowledge about the underlying principles of the brain very much, on the contrary.

Even today, consciousness remains a mystery. Science has some tentative explanations, but they are often so abstruse that one can’t help but to conclude: We know rather little about our consciousness.

Thus we would have to understand the principle by which our brain works?

Personally I do not believe that the brain code is very complicated. The unknown always appears big.

Not complicated? Could this be the hubris of the overly smart researcher?

Not at all. We should also not underestimate the challenge. The brain has a little less than 100 billion brain cells, each with up to 10,000 connections to other brain cells. This network is conceivably complex. Some scientific theories argue that a brain could probably function according to two or three simple basic principles. Some scientists believe, for example, that the brain is a cellular automaton.

A cellular automaton?

These are artificial structures that repeatedly implement a simple rule sometimes resulting in highly-complex structures.

Would you have an example? Yes, an ant colony. Nowadays it is simple to simulate their behavior because an ant, simply put, follows three rules: First of all: If I encounter a cube of sugar I will take a bite. Secondly: Once I have eaten the sugar I will lay a pheromone trail. And thirdly: If I sense such a pheromone trace, I will follow it. Using these three rules it is actually quite easy to duplicate the complexity of an ant hill quite impressively and to simulate or predict many patterns with amazing accuracy.

And where is the connection to the brain?

Viewed from above brain cells are like ant colonies in a way. With an ant hill you also have small ant paths that branch off from the center and grow and shrink. According to some neuroscientific theories a brain could be composed of innumerable ant hills corresponding to the brain cells with their many dendrites. The assumption that the brain is a cellular automaton that functions according to simple basic rules is quite plausible. If these basic principles are discovered the brain code will be cracked and one day brains can be artificially built and duplicated.

Will there come a day when we produce brains on a conveyor belt? If I don’t like my natural brain, I will simply exchange it for an artificial one?

I believe that one day it will be possible to feed information into a brain, for example in order to make learning processes more efficient. So different types of characters or feelings and experiences could be simulated too.

Will we be able to connect the brain to a computer in order to live forever?

The idea that our consciousness is like computer software is probably wrong. If I affirmed your assumption that would mean suggesting that the consciousness is tantamount to computer software. I consider this improbable.

You just said that our brain is based on a few basic principles. That is not different from software.

Even today, consciousness remains a mystery. Science has some tentative explanations, but they are often so abstruse that one can’t help but to conclude: We know rather little about our consciousness. However it is the nature of science that in some areas we can make reasonable statements, in others we operate with a certainty of 50%, and other areas remain completely closed to us. It is important to know what is unknown or not certain.

Are there no certain facts at all regarding consciousness?

Consciousness, individual thinking, has always been a hot subject in science. There is no uniform theory specifically because we don’t completely understand how the brain works. That however is the prerequisite because the awareness of a human being is in a way like a derivative and produced by the brain. We now know that a person loses his or her consciousness, his or her individual thinking, if certain regions in the brain are eliminated. Thus there must be a so-called neural correlate of consciousness - that is what we can say for sure given what we now know.

So the consciousness issue doesn't need be clarified completely in order to be able to crack the brain code?

That is correct. Most living beings are only reflexive machines. Scientists agree that in a human being many things happen unconsciously. In my view there is no such thing as the often-hailed free will. Instead, the consciousness retroactively makes us believe that what we just did was intended to be done. We humans are so-called deterministic machines with the illusion of having a free will. However, in our everyday life we better forget this fact. If we were not living in this illusion of free will, the implications and consequences would be rather decisive.

It would probably go too far to discuss these consequences. Could you instead explain to us the principle of a brain cell?

Each brain cell is a small living organism and has the urge to process information. If it cannot do that, it will die. For example, if your right arm were to be amputated, the brain cells that used to connect with your right arm will no longer receive information. They then attempt to make themselves useful elsewhere, for example by trying to contact the left arm or other targets. The brain is very plastic and is always subject to changes.

Google is like a long-term memory that knows the entire past of the internet. Starmind, by contrast, is a short-term memory that can look into the immediate future and tap knowledge that is still in people’s heads and has thus not been documented.

So the brain cell leads a constant battle for survival?

You could formulate it that way. The future of neuroscience, too, faces a battle. Again and again it is a matter of uncovering errors and wrong assumptions.

Could you give us an example?

There are many perceptions that from today’s view have turned out to be wrong. For example, at high school I was taught that if you shoot a header in soccer, brain cells will be irretrievably lost. Today we know that every day 20,000 to 30,000 new brain cells are born into the spinal fluid. If you learn something new these cells become stronger and survive. Brain cells can definitely be replaced.

Where can our readers learn more about such errors?

I would check YouTube. Put in “Brain Myths Exploded” and you will probably yield quite a lot.

Is it true that you have not read any books for the last ten years?

Yes. Information transfer on the basis of reading is comparatively inefficient. If you read 30 pages per hour each day for 24 hours for 120 years, the information content of what you’ve read, measured by letters consumed, roughly corresponds to the information content of two DVD’s. I don’t read books to merely absorb two DVD’s.

How do you stay up-to-date?

It is much more efficient to talk to an expert who has read 50 books and pose my questions to him. He or she can give me a condensed answer. I want knowledge on demand and not have to read 1,000 pages in order to by happenstance find the two relevant points in which I am interested.

This is how the idea for your company, Starmind, was created?

Exactly. I wanted to give people the possibility to ask any kind of question and get the answer right away. I am thinking of a kind of “Google for the Brain.” You pose the question and our algorithm finds the employee anywhere in the world who can give the best answer. Many people are experts in certain fields, but know little about other areas. It would be excellent if I had a map outlining the knowledge of all other experts – and that is precisely what Starmind does. An artificial brain for a large number of employees, driven by artificial intelligence and consciously by human intelligence.

Who uses Starmind?

Customers in more than 40 countries, among others, Nestlé, Bayer, Swiss Re, Swisscom and car manufacturers. For Starmind to work we have to have access to a sufficiently large human population of at least 1,000 brains.

Who benefits most by Starmind in these companies?

The system is very helpful above all for newcomers in the company. They often have a lot of questions but are too timid to ask their new colleagues. With our technology they can ask their questions to the corporate brain. Even when you type in the question, the inquiring party will be made aware if such a question has already been posed. The corporate brain provides the best answer. Furthermore, knowledge that usually gets lost once an employee leaves a company can also be stored in the corporate brain.

Can private individuals take advantage of Starmind?

Our vision is an algorithm that can think into the immediate future. When reading an article, the consumer is to get relevant answers to questions about which he didn’t even know that they interested him.

Starmind is going to be Google 2.0?

Google is like a long-term memory that knows the entire past of the internet. But a lot of the knowledge that people have is unknown to Google. Starmind, by contrast, is a short-term memory that can look into the immediate future and tap knowledge that is still in people’s heads and has thus not been documented.

Answers to questions that we have not even posed. Do we even want that?

I think that we won’t have any alternative. Using our established behavioral patterns, our outdated strategies that have been anchored in our brain for millenniums, we will face more and more difficulties as a society on our planet. If we crack the brain code we have the chance, for example, to fight illnesses on a sustainable basis or to move on to other planets. We have an opportunity as mankind to grow beyond ourselves and to find important answers in the symbiosis between human and machine.

This article was first published in German on April 21st 2017 at Punkt Magazin. Here