E-MailSucheenglishdeutschTwitterYouTubeFacebook

Logo

 

Algorithms & Artificial Intelligence: Genuine Progress or Just Digital Alchemy?

 

“When we talk about artificial intelligence and algorithms,” said Prof. Dr. Vincent Heuveline, “we touch on a matter that is not only highly topical for us, but which – for good reason –involuntarily gives rise to emotions in many people.” Some 290 visitors attended his lecture on “Algorithms & Artificial Intelligence: Genuine Progress or Just Digital Alchemy?” at Stuttgart’s Mercedes-Benz Museum on the evening of December 3. This was part of the “Dialogue in the Museum” series jointly organized by the Daimler and Benz Foundation, Daimler AG, and the Mercedes-Benz Museum.

First of all, Heuveline said, it is important to acknowledge the achievements brought about by AI in recent years. He still remembers when his teacher from his school days said that a computer would probably beat the world chess champion one day, but that this could definitely be ruled out for the much more complex Asian board game “Go”. But in 2016, he said, the world’s best Go player finally had to admit defeat to a computer in a competition, with 3 games to 0. “Only a year later, it all went up to the next level; it was quite astounding: This time, the computer program had learned the game by itself, without being programmed with any prior human knowledge, and had reached a much higher level of skill,” Heuveline explained. Two AI systems had played several million games against each other within only a few weeks, based solely on the rules of the game, and had constantly improved in this way.

Many people, he said, feel an unsettling sense of competition when they hear of a machine endowed with such extraordinary capabilities. “It’s important that we ask ourselves what role AI should play in future in our society and its organization, and where democratic values could be impinged on by its use,” he said. If we venture to ask what human intelligence actually is, it can be described under many different aspects: as action intelligence, social intelligence, emotional intelligence, or cognitive intelligence. When asked to define AI, Heuveline understands it as the capacity of an IT system to exhibit characteristics similar to human cognition. In certain applications, such as image recognition, AI can well exceed human capabilities. Situational learning and perception are further possible aspects in this regard.

Nevertheless, the current discussion fundamentally overlooks the difference between “weak” and “strong” AI, he continued. All systems we use in everyday life today are weak AI systems that master specific application problems – such as speech recognition in cell phones, navigation devices or, in future, self-driving cars. Strong AI, which is to be on a par with humans in their ability to act or to even surpass them in all areas, on the other hand remains a futuristic dream, he said. “Neither the threat of a Terminator nor a robot that speaks eloquently with subtle humor are something we will see in the foreseeable future,” said Heuveline. He is particularly enthused by applications in which weak AI in the form of electronic expert systems offers support for humans already today. “Especially in medicine, for example when surgeons have to make decisions based on highly complex information in the course of an operation, imaging AI systems are enormously helpful.”

The current rapid advances in AI systems can be explained above all by two factors, Heuveline continued: computer processing power, which is growing at a breathtaking rate, and increasingly cheap hardware. “In combination with suitable software, we then get these amazing results that we already experience in our everyday lives.” Most important is a fundamental understanding of the way data is processed. “It’s all based on algorithms – simple instructions that are processed step by step by the computer.” This can best be compared to a cooking recipe, he said, in which the various ingredients, the order in which they are added, and the way they are processed are precisely defined. Even without detailed knowledge – about butter, sugar, or oven temperatures, for example – the outcome is still predictable.

“The problem for us now is: When we look at a high-performance computer with 2.4 million parallel processor cores, how can we coordinate their computational work? Imagine you’re building a house and 2.4 million workers are helping you,” Heuveline said. “The most important thing is to organize the individual tasks so that they’re not carried just out one step at a time; the helpers’ actions all have to be coordinated.” For this reason, modern artificial neural networks are modeled on a way of working that resembles that of the human brain. Like the brain, he continued, they are capable of processing several items of information in parallel. “An intriguing point here, however, is that although we mathematicians know that highly complex AI systems are based on algorithms, we ultimately cannot say exactly why they work, where information is stored, or how the result is finally reached. For us, too, the system is still something of a black box.” As in alchemy, some very special results sometimes come about – without the scientists being able to explain them in detail.

Heuveline concluded by demonstrating the research results that he has already been able to realize together with scientists from Heidelberg – from a self-learning pacemaker to individualized asthma therapies, to a detailed reconstruction of the body structure of insects that lived some four million years ago. “It’s crucial to the current discussion that we, as responsible citizens, understand what the term AI really means and what applications could be appropriate for self-learning systems. Ultimately, it’s up to us to engage in democratic and enlightened discussion about what kind of society we want to live in and what roles we should assign to such algorithmic systems.”

Speaker
The mathematician and computer scientist Prof. Dr. Vincent Heuveline received his doctorate from the Institut National de Recherche en Informatique et Automatique, France, in 1997. Since 2013, he has been the professor in charge of the Engineering Mathematics and Computing Lab (EMCL) at Heidelberg University; he is also Director of the University Computing Center and head of the research group “Data Mining and Uncertainty Quantification” at the Heidelberg Institute for Theoretical Studies (HITS). He has received several international awards for his research work.
 

Dialog in the Museum
December 3, 2019
Mercedes-Benz Museum
70372 Stuttgart

Speaker:
Prof. Dr. Vincent Heuveline
Heidelberg University