Artificial intelligence in medicine

“We see advances almost weekly”

Will artificial intelligence replace the human intellect? Prof. Stefan Wrobel from Fraunhofer IAIS tackles the numerous myths.

Everyone is talking about AI, but what exactly is artificial intelligence?

We humans see and hear, we make plans and we adapt them to changes. AI is when we replicate these kinds of cognitive achievements using machines or computers. We still don’t know exactly how the human brain actually does these things, but in some areas we can devise mathematical processes that produce similar results. A machine uses algorithms to analyze sample data from which it then derives models and thereby gradually improves its behavior.

So that’s the reality. Where do the myths begin?

Many fears stem from one assumption: that artificial intelligence thinks the way humans do. We see how AI beats humans at chess and Go, or how it translates human language from German to Italian and calls the hairdresser to make an appointment, and we conclude that if AI can do all that, then it can also do everything else we humans can do. But that’s precisely where we’re wrong. Artificial intelligence today still lacks the integrated understanding that we humans possess when it comes to certain processes.

What can AI do?

It’s definitely not a myth that AI can replace humans in certain domains. If a task is clearly defined and delimited and there is sufficient data or human knowledge available, artificial intelligence can accomplish feats that rival those of human experts. It may even exceed them.

And what can AI not do?

Each artificial intelligence system is developed for a specific purpose. It’s currently not possible to automatically apply these systems to other purposes. For example, if a system has been trained to translate sentences from German to Italian, it can’t automatically also answer questions in German or Italian. Another fundamental issue concerns self-reflection: artificial intelligence doesn’t yet know what it cannot do. If an AI system for diagnosing milling machines is suddenly asked to analyze a punch machine, it won’t refuse – but it also won’t produce any meaningful results.

When will AI overcome these limits?

Everyone wants to know what the future will bring and when. Much in the field of AI has happened faster than even we imagined possible. Other things, however, are taking longer. What we can say is that we are seeing advances monthly, almost even weekly. Even next year’s AI systems won’t be like the ones we have today. But there’s no way to reliably predict when artificial intelligence will have come far enough that it can use global knowledge and context the way we do. It may be ten or 20 years or more before AI is generally as flexible as human intelligence. It’s important not to get caught up in thinking about the future possibilities of AI. Rather, we should take advantage of the opportunities it already offers. Only by using AI today can we remain competitive!