Artificial intelligence: “Our computers were appallingly stupid”

Mario Sela, mobility expert at Bitkom, Germany’s digital association, on the technological and social dimensions of artificial intelligence, the potential for optimization that AI offers for mobility, and the question of who should have the final say in automated traffic systems: the machine or the human?

by Peter Rosenberger

Mr. Sela, Tesla founder Elon Musk sees artificial intelligence as “the greatest risk to our civilization”. Physicist Stephen Hawking warns that artificial intelligence could replace humans. Do you think that the concerns of these two well-known sceptics are understandable?

Such concerns are discussed in public repeatedly, and are shared by some distinguished thought-leaders in the technology scene. However, what is being ignored here is the fact that the vast majority of experts see things differently. The risk that the automatic parking system in your car or an intelligent training assistant at work will achieve world domination is very small. We are seeing better and better AI systems, each of which has been trained for a particular task and in some cases can do it better than we humans can. But a machine that is particularly good at diagnosing illnesses can only do just that. It will not be booking trips for me or cleaning my apartment.

How do you rate such warnings that tend to elicit a strong echo in the media? Are they simply cases of counterproductive panic-mongering that hampers development – or reasonable appeals to politicians to think about introducing regulations as soon as possible?

We need to educate people in a better and more broad-based way so that they understand what artificial intelligence can and cannot do. Artificial intelligence is a key technology that will profoundly change our economic, political and social structures in the future. What we urgently need is a strategy at the federal level for the development and use of artificial intelligence. In the area of AI research, Germany is in an excellent position. But if we spend the next few years simply debating the pros and cons of the technology, we will lose out internationally. Both China and Russia are setting an example by placing AI at the top of their agendas and providing adequate financial resources.

In your view, what kind of regulations would be needed?

In addition to having an AI strategy and promoting the subject itself, it is also critical that we look into the risks it may generate. A strategic process to assess the potentials and risks involved in AI should be set up. At the European level, Germany should also campaign for the further development of data privacy and protection in connection with AI. Here a “regulated co-regulation” may be the approach of choice when it comes to implementing data protection rules for AI applications. Under this approach, companies would use best practices to develop codes of conduct that the EU Commission or the supervisory authorities would then recognize as compliant with the applicable laws. AI is of fundamental importance to the development of our future digital society. That is why we need a set of cornerstones for digital ethics. These should be developed in partnership between society, business and politics.

In which areas of intelligence are machines already superior to humans – and in which ones do they still have some catching-up to do?

The idea of intelligent machines whose performance is equal or even superior to that of humans has been inspiring our imagination for a long time. Whether clever on-board computers, human-like androids or the merciless Terminator, science-fiction literature and the movie industry have explored the theme of artificial intelligence again and again. By comparison, the computers on our desks used to be terribly stupid for the longest time. In recent years, all this has changed. Intelligent private assistants on smartphones or in our living rooms who can engage in a real dialog with us, automatic translators that do not simply produce gibberish, or cars that drive from A to B all by themselves without causing an accident – all these are examples for AI systems that have made something approaching a breakthrough into our everyday reality.

What is new about AI, compared to classical electronic data processing systems, is the learning component. It’s no coincidence that terms like ‘machine learning’ or ‘deep learning’ are closely related to AI, as learning is central. There is a huge benefit in this. It’s almost impossible to describe to a software program what a chair really is and the many shapes and designs it can take. A learning system that has been trained with a few hundred photos of chairs is then able to decide quite well as to whether a specific object is a chair or not. Or think of autonomous driving. Explicitly explaining to a conventional system how to respond in any thinkable traffic situation is almost impossible. In contrast, a learning system can master this task.

And how will that all look in 20 years’ time?

In just a few years’ time, many products and services will be equipped with machine intelligence or even have been fundamentally shaped by it. Self-driving cars, resource-efficient logistics processes or effective police work – AI will support people in all kinds of areas and relieve them of routine tasks. It’s not about an AI system replacing a human doctor. Rather, it’s about intelligent systems that work for the human doctor.