By Antti Innanen, lawyer and entrepreneur
The anatomy of hype
The word “hype” originates as American slang from the early 20th century, a shortening of “hyperbole” (from the Greek hyperbolē, meaning exaggeration or overstatement).
The American journalist and critic Douglas Gilbert was the first to use the term actively, in 1925. He used it to describe marketing that deliberately exaggerates or distorts the facts in order to promote sales.
Its use originally spread in the context of the entertainment and advertising industries in the middle of the 20th century.
Hype involves the artificial generation of enthusiasm beyond an object’s real value or significance.
AI is everywhere. Information about it is published constantly. It fires the imagination and attracts investor interest. Although AI ought in principle to be a scientific field, many descriptions of and promises about AI systems are not based on empirical evidence or a clear conceptual foundation.
AI does not develop in a vacuum. The technological development, media attention, public perception, and regulation swirling around it together form a collective idea of what AI can and cannot do. When those capabilities are exaggerated, the consequences can be harmful technology, misleading policy, and distortions in research.
But how do you recognise hype? And how does hype differ from, say, a well-founded scenario about the future?
On the hype cycle
Perhaps the best-known example of research on hype is the hype cycle developed by the research firm Gartner, which describes the life cycle of technologies: initial enthusiasm, a phase of disillusionment, and finally realistic adoption.
It is a visual tool intended to map the life cycle of new technologies — and particularly the fluctuations in expectations around them. The hype cycle can help us understand where technological development sits relative to expectations, and when it is time to be critical rather than swept along.
The curve has its shortcomings, though. Gartner’s hype cycle is not a scientific model but rather a heuristic. It is based on the views of Gartner’s specialists, market observation, and a qualitative assessment of where technologies happen to sit “on the curve”.
Is the hype cycle also hype? It offers a simple and visually appealing way to explain complex technological phenomena. It gives an impression of order and predictability in situations where neither is often present. It is not derived from empirical research, and its phases have not been systematically validated with statistical methods.
And where on the hype cycle is the hype cycle itself? Is it in its own trough of disillusionment? Meta!
How do you recognise AI hype?
The philosopher of science Karl Popper can help here.
For Popper, science does not rest on proof but on falsification: a claim has to be in principle demonstrable as false in order to count as scientific at all.
For Popper, a scientific claim must expose itself to criticism. With hyped AI claims — such as “AI will replace all lawyers within two years” — the problem is not “exaggeration” but that the claim is often vague, imprecise, or not falsifiable.
If you cannot specify exactly what replacement means, in what context, and in what verifiable way, the claim is not falsifiable and so, for Popper, belongs not to science but to pseudoscience.
Not all talk about AI is meant to be scientific, of course. Sometimes the purpose is rhetorical persuasion, provoking discussion, or pure ranting. Sometimes you simply run out of space: it is hard to fit thorough argument and evidence into a limited character count.
But this basic principle is worth remembering, along with the question: what is this based on? Which studies point in this direction? Can this claim be shown to be false?
Negative hype
Not all hype is positive exaggeration. Although hype is usually associated with overblown promises and enthusiasm, its opposite — call it negative hype here — can be just as misleading.
Negative hype is often taken to be grown-up, expert, and sensible. But if negative claims lack empirical evidence, they are hype in exactly the same way the positive ones are.
Negative hype is closely tied to the precautionary principle, according to which new technologies should be avoided unless it can be proven with certainty that they cause no harm. The principle is appealing. Who would want to take an unnecessary risk?
But the precautionary principle easily leads to blind pessimism, in which everything new is interpreted as dangerous by default — as if we knew that nothing bad could follow from the current world, while everything new is a threat. In Popperian terms, negative hype is no more “grown-up” than the positive kind: it is a claim without falsifiability and without evidence.
For example:
“AI will never take jobs.” Sounds sensible, but would in reality require evidence about almost every future scenario. The mirror claim, “AI will take all the jobs,” would demand justification — so why not this one?
“Nobody will trust AI a year from now.” Exactly as much evidence as the claim “Everyone will be using AI next year.”
One of the central hallmarks of hype is that it presents something that will happen in the future as certain. Blindly optimistic and blindly pessimistic theories are, in that sense, exactly alike (thanks to David Deutsch): both predict things without sufficient justification for the claims.
Let’s return for a moment to Dario Amodei’s remark that within the next 3–6 months AI will write 90% of code, and that in 12 months nearly all code may be AI-produced.
Which studies point in that direction? Can the claim be shown to be false?
Does the sentence state something about the future as “certain”, without sufficient grounds?
What if we turn it around? “AI will produce no code at all in 12 months”? What is that based on?
Smells like AI hype, Dario.
Enjoyed this piece? Subscribe to get notified about new Tekoälyfoorumi articles straight to your inbox. We’d also hugely appreciate it if you left us a review on Google. Thank you. 💙
