"AI requires a multidisciplinary approach"

In her new book 'Man versus Machine', Mieke De Ketelaere describes three major challenges for AI. "We have about 70 years of experience with AI now," she says, "but its adoption rate is in a state of stagnation. It should be applied to AI in better ways." The first challenge concerns the energy consumption of AI. It's far too high and must be reduced urgently. We can draw inspiration from nature. A second observation is that AI's scope is often very narrow. "An AI application can perfectly detect cancer by analyzing medical images. But it can’t do more than that."

Further development of AI should focus more on reasoning skills to provide more transparency on how AI makes decisions, Mieke De Ketelaere believes. Although that inevitably leads to a third challenge. "When a system can make its own decisions, it obviously must do so in a way that takes into account all contextual information, including basic human rights." This, in itself, is enough to warrant looking at AI differently.

"Previously, every AI model focused solely on profit," says Mieke De Ketelaere. "That isn't actually wrong, although more focus should now be placed on planet and people, thus making AI accessible to everyone."

A broader and more energy-efficient view

Here, De Ketelaere is referring to so-called doughnut economics. "It's about paying attention to the different layers of the doughnut," she says. "When the automotive industry was booming in the 1950s, profit was the focus. The other layers only followed later, such as energy efficiency (planet) and safety (people)." According to De Ketelaere, AI can only be successful if it follows a similar path. "We shouldn't consider AI from too narrow an angle, as simply a story of data and technology. To achieve the right profit, a broader view is necessary with a focus on people and processes."

Mieke De Ketelaere, Director AI at Imec

How come systems are so energy inefficient? "For the past 30 years, we've been judging AI on accuracy," argues De Ketelaere. "As the amount of data we could analyze and the size of the calculations we could perform grew, we were able to build larger systems and increase accuracy." Except that, the larger the number of calculations, the more energy the systems consume. "When it comes to AI, energy efficiency has only come into play in the last year or so."

AI isn't neutral

Equally important is the need for a legal framework to regulate the use of AI. "The European Union has put together a proposal for a European AI law. It's important to have such a framework, for instance, to regulate the use of applications based on facial recognition." The primary reason for needing such a legal framework is the fact that AI can't be neutral by definition and that AI systems are context sensitive.

"Until now, it was always humans who made the decisions. AI uses that data to do its job but, inevitably, the data contains the human errors incurred when those decisions were made. Hence, on the one hand, an AI system adopts human thinking errors. On the other hand, such a system is context sensitive, and contexts can change in our world. A fact often forgotten." A legal framework will make sense of responsibility and accountability clearer.

Mieke De Ketelaere, Director AI at Imec

First the problem, then the solution

While such legislation will allow the scope of AI projects to be defined, it obviously won't be able to determine the best approach to such projects. "Data isn't the starting-point of a trajectory," says De Ketelaere. "Neither is AI for that matter. You have to start from a problem or a challenge - and find a solution to it. AI can possibly help with that, but just as easily not."

Here, it's a matter of knowing when AI is offering the right answer and then using the technology correctly for that particular case. And no, anyone who starts using AI doesn't need to be an actual specialist. "Comparing it with using a microwave is a good analogy," says De Ketelaere. "Even though we don't know the complex technology behind a microwave oven, we know the context in which we can use it and how to use it safely."

Multidisciplinary debate

Companies need to learn to look at AI in a multidisciplinary way. "AI shouldn't be the exclusive domain of data scientists, because they only think about the accuracy of the data. A multidisciplinary debate on AI is needed, including security, ethics, ease of use, and so on. All these aspects should be included in an AI application by design. This should also provide the basis for a different approach to responsibility in the context of AI. Currently, when something goes wrong with AI, it's always the engineer's fault. That's not right - and therefore needs to change."