Algorithms help to explain our behavior when it is a mystery to us

I am shocked: The Bonn political scientist Ulrike Guérot recently claimed in a panel discussion that algorithms (and thus AI) would fuel the polarization of society because they only operate with zeros and ones, i.e. the binary system, and therefore the mediating "third party" missing

No, dear Ms. Guérot, please don't talk about things you have no idea about: algorithms can do more than 0/1, much more. They even help us to explain our behavior and choices that are a mystery to us.

Would you like an example? Well, it's Friday night and you want to go out to eat. As so often, you are faced with the question: Favorite restaurant or try a new place? – Do you follow cherished habits like Pablo Picasso and Ernest Hemingway? Picasso preferred to paint in the morning, and Hemingway wrote his novels and short stories standing up.

Some base their decisions on rules of thumb, such as those we know from driving school for braking distances: speed in km/h divided by ten squared. Others like to rely on the judgment of others and buy a book because the literary critic Denis Scheck recommends it.

If you follow the psychologist Gerd Gigerenzer, former director of the Max Planck Institute for Human Development, in situations of uncertainty and cognitive overload we like to use simple methods to solve problems, rules of thumb, estimates or empirical values, and we do well with them. According to Gigerenzer, some behavior that initially seems unreasonable to us can turn out to be a sensible way to deal with complexity.

He illustrates this by catching a baseball: when a player catches a ball hit high in the air, he behaves as if he had solved a series of complicated ballistic formulas in a matter of seconds to predict the ball's trajectory. But he probably doesn't know what the formulas are, nor does he care.

He just follows a simple "pi times thumb" rule, also called heuristics: "Fix the ball and always run so fast that the angle of view of the ball does not change!" And still catches the ball. Or as the Nobel Prize winner for economics put it As Milton Friedman put it, “People only pretend to act rationally.”

Well, Friedman and Gigerenzer have been refuted in a way—at least as far as catching a ball goes. Here, heuristics that solve problems in a practicable but not necessarily perfect way not only pretend to be optimal ball-catching machines, they actually are, as Constantin Rothkopf and Jan Peters, my colleagues at the TU Darmstadt, were able to prove.

For the first time, a robot answers questions in the British Parliament. Ai-da, an "ultra-realistic humanoid robot artist" developed at Oxford University in 2019, told members of a House of Lords committee how it works.

Source: WORLD | Lore Schulze-Velmede

In AI, heuristics are used for a wide variety of tasks, from chip design to car navigation to machine learning. If a machine were to learn to recognize sweet grapes, one could proceed like a mountaineer who is looking for the summit in dense fog, directing his steps uphill and not downhill as much as possible.

Initially, the machine follows the rule "The harder, the sweeter." It quickly finds out from the data: "Hm, the hard ones aren't that sweet!" So it changes the rule a bit: "The softer, the sweeter the grapes." What a treat. The change has worked and the rule is adopted. If every further change leads only to worse results - like at the North Pole every step leads only to the south - then the machine must have reached a peak: It has learned!

Sure, we can get lost in the fog like Hans Castorp in The Magic Mountain, and heuristics don't always lead to the summit - but heuristics in AI are always and always algorithms. Therefore, I unfortunately have to disagree with the computer scientist Katharina Zweig, Rhineland-Palatinate's AI ambassador since September 2020, when she tweets that "in most cases, no algorithms are used in machine learning, only heuristics".

In short, the power of algorithms is almost uncanny in that they can map even seemingly inexplicable behavior. But only the right basic understanding of their probably unlimited possibilities opens up completely new perspectives for all of us, the economy and science.

Kristian Kersting is a professor for AI and machine learning at the TU Darmstadt, co-director of the Hessian Center for AI (hessian.ai), author of the book (“How Machines Learn”) and winner of the “German AI Prize 2019”.

"Everything on shares" is the daily stock exchange shot from the WELT business editorial team. Every morning from 7 a.m. with our financial journalists. For stock market experts and beginners. Subscribe to the podcast on Spotify, Apple Podcast, Amazon Music and Deezer. Or directly via RSS feed.