In newspapers, on TV, at the pub or by the coffee machine: the words artificial intelligence are on everyone’s lips. By now, AI has become an umbrella term for a wide range of technological applications. And that’s leading to unnecessary confusion, according to Vice-Rector of Research and Impact Maarten Weyn. High time, then, for a practical overview.
If it were up to Weyn, we’d stop talking about artificial intelligence. Or at least stop talking about it the way we do now. ‘Today, when you open a newspaper or watch a talk show, you hear statements such as: “AI will save our jobs”, “AI will kill our jobs” or “AI is a threat to humanity”,’ the vice-rector says. ‘The debate is fascinating, but it starts from a flawed assumption: AI doesn’t exist as a single, indivisible technology.’
Let’s stop asking what we think about AI and start asking: Which system, for what purpose, and under what conditions?
‘By lumping everything together under the label of AI — from Netflix recommendations to autonomous drones — we lose sight of reality. We’re using a catch-all term where precision is needed. Imagine a national debate about the future of transport. One person warns about traffic deaths, another wants more cycle lanes, and a third argues for road pricing. The absurdity becomes obvious immediately. Yet this is exactly what we’re doing with AI.’
Breaking AI down
According to Weyn, the risks, benefits and regulations differ fundamentally depending on the type of AI system involved. He argues that the technology should be divided into functional categories and offers the following starting point.
- Agentic AI: The doers. The next step, which we’re currently in the middle of. Systems that not only provide answers, but also independently carry out actions: booking a trip, fixing a software bug or managing a complex logistics process without constant human intervention.
- Artificial General Intelligence (AGI): The horizon. This is the theoretical form of AI that would outperform humans in every domain. It’s important to think about this, but crucial not to confuse it with the tools we already use on our smartphones today.
- Fake AI: The marketing gloss. Not every application carrying an AI label actually contains AI. Traditional statistical models or ordinary databases are now frequently sold as AI-powered. This isn’t a harmless issue: it clouds the debate, misleads customers and investors, and makes it harder to distinguish genuine AI applications from empty hype.
Curious to find out how AI can add value to science?
The responsible use of specific AI models can facilitate and accelerate medical breakthroughs in healthcare. In this way, they offer enormous added value and make personalised medicine possible.
- Foundation models: The foundations. Large, versatile base models on which many other applications rely. A single model such as GPT or Llama powers thousands of products, from chatbots to agentic systems. Whoever sets the rules here also indirectly codetermines what becomes possible. It’s no coincidence that the European AI Act treats them as a separate category.
- Generative AI: The creative assistants. Think of Large Language Models such as ChatGPT or image generators such as Midjourney. Built to recognise patterns in language and images and generate new content. Useful as brainstorming partners, but unreliable as sources of facts.
- Physical AI: The embodied systems. AI that leaves the digital world and acts physically: self-driving cars, warehouse robots, drones, surgical assistants, etc. Software errors have real-world consequences, ranging from incorrectly packaged parcels to accidents on public roads. Questions surrounding safety, liability and certification are therefore of an entirely different nature from those surrounding a chatbot.
- Predictive & classification systems: The calculating powerhouses. These systems identify trends in enormous amounts of data. They predict when a factory machine will break down, detect tumours in medical scans or filter spam from your inbox.
- Recommendation systems: The curators. The algorithms behind YouTube or Spotify that determine what you get to see. This is where the major questions about polarisation and filter bubbles arise — questions that are entirely different from those asked about, for example, industrial robots.
Generic fear leads to generic policy — and that’s precisely what we cannot afford, Weyn argues. Policymakers therefore need to be aware of the many different facets of the technology. ‘Someone who fears AGI shouldn’t block a local entrepreneur from optimising stock management with a simple model,’ he explains. ‘Let’s stop asking what we think about AI and start asking: Which system, for what purpose, and under what conditions?’
‘Only then can we give technology the place it deserves: as a collection of very different tools, each with its own challenges, ethics, opportunities and responsibilities. The real debate only begins once we stop saying “AI” and start being specific about what we’re actually talking about.’
Interested in AI?
Fancy finding out about the latest developments in artificial intelligence? Curious about its applications, opportunities and pitfalls?