We Call AI an Intelligence; That Is Exactly the Challenge
The parallel between artificial intelligence and human intelligence is more than a terminology problem. It is a challenge that can turn into a risk.
Two reactions meet me very often as soon as a conversation turns to artificial intelligence. Some hold almost impossible expectations and delegate tasks and processes uncritically. Others fear AI developments too much, feel overwhelmed, and end up blocking progress as a result. Both attitudes are problematic, and both probably share the same underlying cause. It is the word we use.
A term chosen for distinction, not for description
The term artificial intelligence originates from the proposal for the Dartmouth Conference, drafted in 1955 by McCarthy, Minsky, Rochester, and Shannon, and established at the conference in the summer of 1956. The motive behind it is revealing. McCarthy did not choose the name because it described the matter most accurately. He wanted to distinguish the young field from cybernetics, as it then existed. Various other terms were discussed beforehand, such as ‘automata studies’, ‘complex information processing’, ‘neuraldynamics’, or ‘advanced automatic programming’. Artificial Intelligence (AI) then prevailed.
By his own later account, no one on the team really liked the name, because the actual goal was ‘genuine’ intelligence, not ‘artificial’. From the outset, the term carried more positioning than scientific precision. Anyone who leads with this word today is working with a brand from 1956, not with a definition.
What the machine actually does
The language we have built around these systems is threaded through with human terms. Machines ‘understand’, ‘learn’, ‘decide’. We seem to have no other words for these mechanical processes. The philosopher John Searle showed as early as 1980 that, in the ordinary sense of the words, it is inaccurate to attribute understanding, belief, or perception to a program. Floridi and Chiriatti carried this line forward for modern language models. These systems generate text that appears human, without human understanding behind it. What actually happens is statistical processing of patterns at very large scale, over vast volumes of data. That is impressive and operationally useful, but something different from what the word intelligence suggests.
These terms, and the understanding tied to them, are not harmless. When a report states that the system ‘assessed’ a candidate or ‘identified’ a risk, the word choice imperceptibly shifts attribution. A statistical output that a human ought to review turns into a verdict that gets followed. The language pre-empts the decision that still needs to be made.
The situation is sharpened by a recent finding. Jones and Bergen reported in 2025 that large language models pass the Turing test, which tests whether a machine possesses human-like reasoning ability. That does not prove that machines think. It shows that this test measures imitation, and imitation is exactly what the misleading label reinforces. The more human the output appears, the harder the false expectation grips.
Is the term artificial intelligence wrong?
One might be tempted to simply call the term wrong. It is not that easy to answer, because no generally accepted scientific definition of intelligence exists (cf. Neisser 1996). The term is not wrong. It is more that it is imprecise, emotionally charged today, and creates false expectations. It invokes a comparison with humans that the technology does not deliver on, and it is exactly this comparison that steers how executives behave.
Both over-expectation and refusal cause problems
Anthropomorphising language activates human intuition, and that has measurable consequences. Experimental research on overreliance shows that people follow a machine’s recommendations even when the context argues against it (Klingbeil et al., 2024). The OECD describes the same thing as blind trust in technology, and Ruschemeier and Hondrich (2024) show, both legally and psychologically, that in this state one’s own judgement is no longer actually exercised. Anyone who believes they are facing an intelligent machine checks less (cf. Automation Bias). That is the first misjudgement, over-expectation, and it leads directly to blind delegation.
The second misjudgement is its mirror image. Anyone who reads the word intelligence as a threat, as a thinking and possibly superior entity, falls into defensiveness. Dystopian films such as 2001: A Space Odyssey, Terminator, or The Matrix do not help. Even though it is only science fiction, such images have shaped fears that stand in the way of a sober assessment.
In my personal view, at board and executive level we often start from a false premise, and that premise then becomes the reference point for the entire discussion. One example is the term ‘avatar’ of a board member, meaning the digital representation of a person in a virtual world. Exactly two problems hide in that idea. On one hand, a human being cannot be represented by a machine that works probabilistically rather than deterministically, has no free will, and cannot be held culpable. On the other hand, the effect reverses. A virtual system influences our real world, the one in which we carry ethical and legal responsibility. Both points are more than linguistic pedantry. Treat a system like an acting agent, and it creates a false impression of human control, and responsibility gets misattributed (cf. Lima et al., 2022).
What this means for the board
A system that rests on probabilities rather than certainties and facts, carries biases from training data, reacts partly unpredictably to changes in data and models, and is barely traceable, must be treated accordingly. As a machine, not as a representative of a human being. This is exactly where the board’s responsibility lies. On one hand, it must create the conditions to monitor and control the system; on the other, it must operationally demand that this actually happens in practice. Responsibility stays with the people and organisations that deploy it. We must not anthropomorphise machines, and we must not treat them as representatives of humans, because our human intelligence and our value system cannot be transferred onto an artificial intelligence.
We will not shake off the term artificial intelligence; it is too firmly established for that. But we can use it more consciously, and we should neither anthropomorphise it nor attribute human qualities to it. Anyone who carries responsibility does best to define the word as what it is: a label from 1956, for a machine that did not yet exist at the time.
A board does not need to understand artificial intelligence in technical detail, but it does need to build technology competence. Part of that is the right mental model. Anyone who thinks of a machine and a system, rather than picturing a superior colleague with a magic wand, has already taken the decisive step.
Language shapes our reality. Loosely after Wittgenstein: the limits of our language mean the limits of our world. It is up to us to shape those limits, and it is up to us, as human beings, whether that becomes a risk or not.