German computer scientist Jürgen Schmidhuber is one of the world’s top experts on artificial intelligence. He’s excited about self-learning programs that imitate the human brain and expects a revolution in the financial world and beyond – even an upheaval in human history. He talked recently with Handelsblatt during a portfolio management conference in Frankfurt.
Professor Schmidhuber, has there been increased interest in your work?
Yes, an enormous increase. It’s the big companies – like Google, Facebook, Baidu, Apple or Microsoft – that are suddenly showing great interest in our algorithms, some of which are quite old, for machine learning to solve problems. These involve language processing, picture recognition, individualized advertising and information searches in big data pools.
How long have you been involved in research?
I have been working on artificial intelligence since 1985. Relevant algorithms, many of which were first developed in my laboratory, are more popular today than they used to be. Because in two decades, computers have become about 10,000 times more efficient by price unit.
Where is the overlap with financial markets?
Forecasting stock prices, for example. Looking back, many stock price developments seem very simple and predictable. I like to use the example of Cisco Systems. In the 1990s, the stock price increased quite consistently in value, achieving an incredible 85,000 percent gain. That was a very regular series of financial data.
Is that realistic?
Often it isn’t, and that’s the point. Many stock price data seem to have no pattern, but perhaps they do. There can be regularities hidden in a series, hardly recognizable to the naked eye, although they are relatively simple. Most people overlook the patterns but our neural networks are able to recognize it and that way make better predictions.
Who’s using this already?
Since the 1990s, people have been trying to find patterns in financial data with the help of machine learning processes. We’ve seen a lot of success recently with comprehensive feedback neural networks, based on the human brain, in many applications. There are now more of these being developed for the financial world. We have set up our own company, using processes like this to find stocks of undervalued companies.
How should a layman imagine artificial intelligence and neural networks?
Artificial intelligence based on neural networks learns how to recognize patterns and solve all kinds of problems. That already works quite well. Our neural networks have already won many competitions, for example, for handwriting recognition. Also for languages that we don’t understand, like Chinese and Arabic. Medicine is also a big future niche, for example where our neural network can learn from microscopic pictures of breast tissue, how to recognize early stages of cancer nearly as well as an experienced histologist.
And how will things look in the future of the financial world?
We feed all the data we can get into the learning feedback neural network. Not just past stock prices, but also newspaper articles – and in the future, perhaps video transmissions from stock trading rooms, where we can capture the atmosphere through face recognition. If traders are looking aghast, then clearly something is going on that will affect stock prices.
Are there are many such systems in circulation?
Many asset managers at least have small teams for machine learning. But hedge funds in particular often don’t want to talk about their success models. I assume that learning programs are being used more often than many people think.
Will this field continue to expand?
We will see an explosion of applications in all areas of pattern recognition. Perhaps we will recognize the best ones by their success. So we’ll see the end of the classic fund manager – either completely or partially. In the future, he or she will work with the support of artificial intelligence – or the intelligent machine will even work alone.
Do the explosion of knowledge and accelerating speed of information-processing also mean dangers in the future?
We know that every 10 years the performance of our computers increases 100 times over, per euro. If you extrapolate this further, the picture looks pretty intense. By the next century at the latest, we will be in a situation hardly imaginable today: A machine smaller than a pinhead could theoretically simulate a community of 10 billion agents comparable with the human race. Each would be equipped with an artificial feedback neural network with the raw computational power of a brain, located in virtual reality, itself 10 billion times more complex than all the neural networks put together.
What would that mean?
One thing seems clear: It’s the most important development of our time. In coming decades it will change nearly every aspect of our civilization – very quickly and very radically. Humans will soon no longer be the most important decision-makers.
So will machines then rule, as we have seen in the movies?
Most of these movies are far-fetched – like the film “Matrix,” where for unfathomable reasons machines are dependent on the negligible energy of human brains. What nonsense.
So do we have a chance?
It is likely that human beings will not be able to control the much cleverer artificial intelligence. But we can hope there will hardly be any conflict of interest between “us” and “it.”
Human interests are mostly restricted to a very thin biosphere around the third planet of the solar system. The biosphere is full of oxygen, which makes many robots rust. The rest of the solar system is not made for humans, but for suitably constructed robots.
Does that mean machines will inhabit outer space in the future?
Robots or parts of them will travel in the most elegant and rapid ways – for instance, by radio, from transmitter to receiver, right through the solar system and beyond. What would be considered life-threatening from a biological point of view, offers incredible new possibilities for robots and software. Why should advanced examples of artificial intelligence attach such importance to the puny surface of planet number three? That was the optimist speaking.
One last question: You always wear a cap? Do you ever take it off?
Yes, under water. But it’s become my trademark during lectures. Besides, many of my colleagues are mathematicians who have difficulties recognizing faces. The cap helps them.
Video: Jürgen Schmidhuber says we may be part of a simulation.
Ingo Narat is an editor with Handelsblatt’s finance section. To contact the author: firstname.lastname@example.org