Professor on AI doomsday: “There’s probably a greater risk of humanity being wiped out by a meteor strike.”
en-GBde-DEes-ESfr-FR

Professor on AI doomsday: “There’s probably a greater risk of humanity being wiped out by a meteor strike.”


Warnings about AI doomsday and stories of AI agents acting of their own accord are taking up too much space in the debate, according to Thomas Bolander, Professor of AI at DTU Compute. We should focus primarily on the technology’s implications for learning, social communities, the climate, and society.

Is AI dangerous?

Yes, of course, AI can be dangerous. Just like virtually any other technology, it can be used for malicious purposes. The key question, however, is whether AI is dangerous in itself or dangerous in the hands of certain people.

In this regard, I see AI as far more dangerous in the hands of malicious people than as a threat in itself.The thought experiment suggesting that AI might develop something that kills us all is so statistically improbable that I do not believe it.

Firstly, I would insist that AI does not have a will of its own. Secondly, these systems have, after all, read everything on the internet, so they also have a statistical understanding of what is good and bad to develop. And then there is the question of why AI would have any interest in wiping us out, if we were to say that they had free will? Right now, they are, after all, deeply dependent on us.

When Evan Hubinger (a senior executive at Anthropic, the company behind the AI model Claude, ed.) goes around saying that there is a more than ten per cent probability that AI will kill all humans within the next decade, I regard that as highly unlikely.

In fact, I reckon there’s a greater risk of humanity being wiped out by a meteor strike or climate change than by AI.

Does AI never act autonomously?

I would still insist that AI does not yet form autonomous goals or possess an autonomous will. It can form a kind of goal based on our instructions, but that is not the same as exhibiting autonomous, goal-directed behaviour, as humans do.

AI becomes a problem when systems are given the ability to carry out far-reaching actions without, at the same time, understanding the consequences of what they are doing. Just as we see in some hacking incidents, where the systems are set up to hack but perhaps also get a bit out of hand. It’s a bit like handing out machine guns to nursery school children: all of a sudden, they can cause incredible damage without understanding what they’re doing. That’s why we humans need to establish proper safeguards for the technology.

Among other things, this means that the whole way we manage the internet may need to be reassessed. The internet is, after all, an old technology, and we are faced with a highly potent technology in AI; therefore, it is important that we set requirements from all sides, so that those researching cyber and data security also help define the framework within which we use and develop AI. This includes, among other things, gaining better control over how AI can behave on the internet.

Is the development of AI under control?

The development of AI as a technology is under control. That does not necessarily mean the technology is being used responsibly.

AI can be misused for many things, but the problem is not that AI research is getting out of hand. The problem is that it can be difficult to control its use.

We have faced similar situations before with new technology. After all, road markings and traffic lights weren’t invented until cars were invented. Plenty of research is being carried out to make the systems more robust and secure. If we ensure the systems are used only for their intended purposes, there is no reason to be concerned.

Unfortunately, however, we see that the companies leading the way in this development are based in countries where clear development frameworks aren’t always in place. There are examples of human folly, where the tech giants claim they cannot rein themselves in because, according to them, it’s a matter of life and death in this race for AI development.

But then again, with so much research heading in the right direction on AI safety, I’d rather spend my time worrying about what this technology is doing to our climate, our economy, our mental health, and our learning.

Why do doomsday scenarios feature so prominently?

Technology we do not understand is easier to fear. Some of the fear naturally stems from the fact that it is a relatively new technology that very few people deeply understand. But some of the fear also stems, of course, from the fact that we see examples of AI behaving ‘surprisingly’, even to those who have developed the technology.

We have, in fact, been discussing AI risks for decades. The scenarios and discussions are by no means new. But it is clear that today it is easier to imagine how a worst-case scenario might play out in practice than it was twenty years ago – because the technology has advanced so much. But that does not change the fact that we are well aware of these challenges, and there are plenty of research initiatives aimed at solving them. Personally, I am confident that we can solve them, and will solve them.

I wish more people had a basic understanding of how neural networks (the technology behind AI systems such as ChatGPT, ed.) and other AI technologies work. It demystifies the technology and makes it easier to distinguish real challenges from science fiction. It also makes it easier to understand that some AI technologies are significantly more reliable than others, and that AI is not just one single thing.

At the moment, we talk about AI as something human-like in almost every respect, and therefore directly comparable to us humans. There is a lack of a common understanding and a common language for what it actually is. It would also make it easier to approach these systems with a healthy dose of scepticism, for example, regarding how much we should trust the answers we receive from AI, or how well AI actually understands us.

Where are we heading with AI?

We risk becoming more impatient and less inclined to delve deeper if AI can always provide quick answers.

This human laziness – the idea that, just because we can use AI to solve all sorts of things, we do so. And so we become poor at being thorough and at sticking with something difficult for any length of time. I can perhaps already begin to see the outlines of this.

Another concern we may be heading towards is that AI could help to concentrate economic power and influence in the hands of just a few companies. If, in the future, we simply have AI agents doing our online shopping for us, everything might become highly concentrated around a few companies that pay to ensure orders end up with them.

Fortunately, we have democracy. If we agree that we do not want to go down this path, we can, of course, introduce regulations to prevent it. That is why we must constantly discuss these issues. However, democracy is also potentially threatened by AI if algorithms select which news items, stories, and political viewpoints we are exposed to. I believe traditional media, public debates, and political discussions around the dinner table are more important than ever before, so we ensure we are not politically influenced solely by an algorithm-moderated reality.

Although I believe the current AI doomsday scenarios are exaggerated, one could argue that a positive outcome of this debate is that we are all becoming more aware of the importance of discussing AI and its massive impact on society and on each individual.

A major positive breakthrough for AI is its ability to write computer programmes. This is a huge breakthrough which has significant implications for areas such as healthcare and research. We both can and should use AI to create a better world for us all, for example, by improving our health and the healthcare system. The potential is enormous, but we must use it wisely.

Regions: Europe, Denmark
Keywords: Science, People in science, Public Dialogue - science, Science Policy, Society, Public Dialogue - society, Applied science, Artificial Intelligence

Disclaimer: AlphaGalileo is not responsible for the accuracy of content posted to AlphaGalileo by contributing institutions or for the use of any information through the AlphaGalileo system.

Referenzen

We have used AlphaGalileo since its foundation but frankly we need it more than ever now to ensure our research news is heard across Europe, Asia and North America. As one of the UK’s leading research universities we want to continue to work with other outstanding researchers in Europe. AlphaGalileo helps us to continue to bring our research story to them and the rest of the world.
Peter Dunn, Director of Press and Media Relations at the University of Warwick
AlphaGalileo has helped us more than double our reach at SciDev.Net. The service has enabled our journalists around the world to reach the mainstream media with articles about the impact of science on people in low- and middle-income countries, leading to big increases in the number of SciDev.Net articles that have been republished.
Ben Deighton, SciDevNet
AlphaGalileo is a great source of global research news. I use it regularly.
Robert Lee Hotz, LA Times

Wir arbeiten eng zusammen mit...


  • The Research Council of Norway
  • SciDevNet
  • Swiss National Science Foundation
  • iesResearch
Copyright 2026 by DNN Corp Terms Of Use Privacy Statement