Should we fear AI?

Since the beginning of the 1990s, Professor Ole Winther from DTU Compute has conducted research in machine learning—a field combining statistical modelling, intensive computation, and artificial intelligence. His vision of the future is not quite as grim.

q: What are the most common misconceptions about artificial intelligence?

a: It’s a misconception that we’re very close to having a so-called strong artificial intelligence—i.e. an intelligence on a par with human intelligence. Or higher. We’re very far from that scenario. Maybe one day we’ll see it. In a thousand years Another misunderstanding concerns the speed of development. There are some—including Elon Musk—who insist that we’re witnessing an exponential technology development which is more or less out of control. But this assumption is incorrect. The rapid development we’re currently witnessing is mainly due to the fact that we’ve now succeeded in getting computers to use some of the models we developed more than 25 years ago. Put another way, we shouldn’t spend time developing algorithms that enable a computer to learn. We just have to open our drawers and pull out the existing models.

q: So the old models still work?

a: At the time, we actually abandoned the models and thought they didn’t work. But it turns out that—with a few adjustments—they work just fine. Back then, we simply didn’t have the computer power to use the algorithms or handle large data sets. Nor was the technology in great demand. So, in reality, current AI development is riding on the back of more than two decades of fundamental technology development.

q: Should we fear AI?

a: There is—in my opinion—an element of hysteria in the current debate, and I don’t believe that we should fear artificial intelligence. In reality, artificial intelligence is just another technology which most of us are already using—e.g. in our smartphones.

q: In what direction is the development heading?

a: The main part of the development is taking place in private companies and is therefore profit motivated. In the coming years, therefore, the development will primarily take the form of narrow artificial intelligence—e.g. voice and facial recognition for new smartphone services. In the latest iPhone, Apple recently introduced facial recognition to unlock the phone.

q: What are the biggest challenges facing the global community in connection with artificial intelligence?

a: The immediate challenge is the real danger that artificial intelligence is used in weapons. Here, we need to tread carefully, as—in a few years—the hardware will be quick and light enough to make it possible to build, for example, inexpensive autonomous drones with facial recognition software and weapons. Therefore, there is a need for international conventions like those we have for chemical weapons, to prohibit this kind of weapon. But artificial intelligence also has the potential to make a positive difference in many areas. The challenge is to roll out the technology to the greatest possible benefit for all of us. For example, it can also be used for self-driving cars. When this succeeds, it would change society significantly, because it can make children, youngsters, and old people mobile in a completely new way. Artificial intelligence is also a technology that can help people make the right decisions, e.g. doctors who are to make diagnoses and choose between different types of treatment.

q: Will we not end up making ourselves superfluous?

a: If we believe that artificial intelligence will make people superfluous, we’re underestimating human creativity in finding something to do. Sometimes, technology can surpass human abilities. But when Kasparov was beaten in chess by a computer, we did not, after all, stop playing chess. On the contrary, people began studying how the computer had done it. We can suddenly learn something new. This is an example of how technology can expand the human repertoire. It can find solutions that humans have never even considered.