Artificial intelligence is everywhere — in the news, in our phones, in the workplace, and increasingly, in our imaginations. Headlines warn that AI might soon replace human jobs, write bestselling novels, or even surpass human intelligence altogether. Between fascination and fear, it can be difficult to tell what’s real progress and what’s pure hype.
As part of this effort, Integreat scientists are challenging the myths that shape public understanding of AI. They argue that behind the buzzwords lies a technology that is powerful and useful, but not magical — a tool that depends, above all, on the people who design, instruct, and apply it.
“Modern day AI (LLMs) are not sentient intelligent subjects”,
says Daniel Kaiser, PhD fellow at UiT, The Arctic University of Norway and a member of Integreat. “It has no feelings, intrinsic goals of its own, or agency. It’s trained to pursue the goals of the person instructing it. How you instruct it matters and turns it into what you want it to be.” As Daniel says, “AI is a tool like a hammer. What it does, depends on what you use it for”. Understanding that distinction is essential to separating fact from fiction in today’s AI debates.
Myth #1: “AI Will Take Over the World”
Few ideas capture the public imagination quite like the notion of AI taking over — outsmarting humans, seizing control, or even causing extinction. These stories make for gripping films and headlines, but they are far removed from the scientific reality.
Daniel points out, modern AI systems have no independent will or purpose. “They will not take over the world or cause human extinction”. He says, “In fact, it’s far more likely we will avert the extinction of our species when we have AI available as a tool, than not.”
AI models don’t have motives or consciousness — they don’t want anything. What they do have is the ability to identify patterns in vast amounts of data and respond to human instructions. This means their impact depends entirely on human choices: how they’re trained, what they’re used for, and who controls them.
Yet even as AI systems grow more capable, they remain imperfect. Kaiser reminds us that
“AI is not perfect and makes mistakes just like we do — it will do so for the foreseeable future.”
Like any human-made system, AI reflects both our intelligence and our errors. Recognising that is the first step to using it wisely.
Myth #2: “AI Always Knows the Answer”
Another common misconception is that AI systems always know what they’re talking about. After all, when a chatbot produces fluent, confident responses, it’s easy to mistake that confidence for understanding.
But appearances can deceive. “I would like to change the fact that people tend to think that AI is certain about what it proposes,” says Ida Scheel, Associate Professor at the University of Oslo and member of Integreat. “AI is not as confident as you think — it’s like the worst know-it-all you know.”
This misplaced confidence stems from how AI models are built. They are designed to predict what comes next — not to reason, verify facts, or grasp truth in a human sense. “AI doesn’t ‘think’, but rather predicts,” explains Youssef Wally, Doctoral Research Fellow at UiT. “It’s just a fancier way of unconditionally following patterns in human behaviour and opinions.”
Jonas Nordstrøm, PhD research fellow at the University of Oslo, adds that even the term AI hallucination can be misleading. “It suggests that certain situations cause flaws or lapses in the model, when in fact, untruthfulness is inherent to how these systems work. AI models don’t truly understand what ‘correct’ or ‘incorrect’ means; they simply predict what seems most likely or plausible. In that sense, they’re more like improvisational artists: generating responses that sound convincing, and that often happen to be accurate, but are ultimately performances of plausibility.”
That’s why users must interpret AI output with care. “Modern generative LLMs are just stochastic parrots,” cautions Aliaksandr Hubin, Associate Professor at the University of Oslo. “One needs to take responsibility for what the parrot says and not take for granted anything it says without double-checking it with human intelligence.”
The bottom line: AI can be remarkably persuasive — but not necessarily right. Its apparent confidence reflects probability, not certainty. Using it responsibly means knowing when to trust it, when to verify, and when to say, “that doesn’t sound quite right.”
As T.S. Eliot once asked,
“Where is the wisdom we have lost in knowledge? Where is the knowledge we have lost in information?”
In the age of AI, these words feel more relevant than ever — reminding us that understanding cannot be automated.
Myth #3: “AI Is Cheap and Effortless”
If AI can now write essays, generate images, and answer questions instantly, it might seem like a tool that’s cheap, effortless, and endlessly scalable. But beneath that apparent simplicity lies a complicated — and costly — reality.
“‘AI is easy and cheap to use’. Not quite,” says Johan de Aguas, postdoctoral research fellow at Integreat. “Extracting the most benefit from AI such as LLMs requires specialised knowledge and skills, for instance to prompt it correctly. Plus, the whole AI value chain consumes massive amounts of energy, and so it is actually very costly for the world.”
De Aguas’s point highlights two hidden challenges. First, getting good results from AI takes expertise — understanding how to guide it, verify its answers, and use it responsibly. Second, the apparent “cheapness” of AI conceals significant environmental costs. Training large models consumes enormous amounts of electricity and water to power and cool data centers. Even running them at scale can leave a considerable carbon footprint.
Recent analyses suggest that training a single large language model can emit as much carbon as the lifetime of several cars, and that data centres supporting AI could soon account for a notable share of global electricity use. As AI applications expand, so too does the importance of making them more energy-efficient and environmentally responsible — something that lies at the heart of Integreat’s research.
Making AI truly affordable, then, isn’t about lowering its price tag. It’s about designing systems that use less energy, waste fewer resources, and deliver real value without hidden costs to the planet.
Myth #4: “AI Is a Single, All-Purpose System”
It’s easy to think of AI as one mysterious, all-knowing entity — a monolithic intelligence that can do anything we imagine. In reality, the field is far from unified.
“While many people might think of AI as one unified system, in reality AI is a broad field that covers many distinct subfields focused on different goals, including learning, reasoning, knowledge representation, natural language processing, perception,” explains Johan Pensar, Associate Professor at the University of Oslo. “These subfields again make use of a wide range of techniques from fields such as machine learning, statistics, mathematics, logic etc. And, importantly, each method used in AI is designed for its specific kind of problem.”
Understanding this diversity helps clarify why AI systems are specialised rather than universal. A model trained to predict protein folding isn’t automatically capable of writing essays or driving a car. Each application relies on carefully chosen methods designed for its particular problem, and no single system can do it all. Recognizing the modular, goal-specific nature of AI helps set realistic expectations and emphasizes the importance of human guidance in designing and applying these tools.
Few areas inspire more excitement about AI than healthcare. Stories of algorithms discovering new drugs or diagnosing diseases faster than doctors often make headlines. But what’s truly happening in labs and hospitals tells a more nuanced — and more interesting — story.
Take protein folding, one of biology’s great puzzles. Proteins are the tiny molecular machines that keep our bodies running, and their functions depend on how they fold into intricate 3D shapes. For decades, predicting those shapes from genetic sequences was an immense scientific challenge. Then, in 2020, DeepMind’s AI system AlphaFold made a major breakthrough: it could predict protein structures with remarkable accuracy, saving researchers months or even years of experimental work.
That achievement was genuine progress. It opened new possibilities for understanding diseases and accelerating early-stage drug discovery. But it didn’t mean that AI had suddenly learned how to cure diseases. Translating a protein prediction into a safe, effective treatment still requires long-term biomedical research, testing, and above all, human expertise.
AI assists scientists by narrowing the search space — pointing them toward promising hypotheses faster than before. Yet the responsibility for interpretation, validation, and clinical decision-making remains firmly with humans. As Integreat researchers often emphasize, AI is a powerful collaborator, not a replacement.
Used wisely, AI can amplify human intelligence and help tackle complex health challenges. Used carelessly, it risks generating misleading results or misplaced confidence. The difference, as always, lies in how we choose to use the tool.
Artificial intelligence is neither magic nor menace. It’s a tool — a remarkably powerful one — shaped by the people who build and use it. Understanding that helps us move beyond both hype and fear, toward a more informed and realistic relationship with technology.
AI will continue to evolve, but the principles guiding its development must remain steady: transparency, responsibility, and a clear understanding of its limits. By grounding AI in science rather than speculation, we can ensure it remains a force for human benefit — helping us solve real problems without creating new ones.
That, ultimately, is what Integreat stands for: making AI trustworthy, sustainable, and worthy of our confidence.
As Isaac Asimov once warned,
“The saddest aspect of life right now is that science gathers knowledge faster than society gathers wisdom.”
The challenge — and opportunity — lies in ensuring that our wisdom keeps pace with our machines.