Artificial might be right. Intelligence remains contested.

I think part of our problem with artificial intelligence is the name itself.

Artificial is probably fair enough. These are human-made systems. Intelligence is a much more contested proposition, as Gignac and Szodorai discuss. The term came from people who were trying to build intelligent machines, and it reflects their view of the problem: that aspects of intelligence could be described and reproduced through computation.

That is a useful starting point for a research program. It is not a sufficient account of intelligence.

Yet we use the name as though that question has already been settled. It shapes what we expect these systems to do and how much authority we are prepared to give them.

A name from the people building the machines

John McCarthy is credited with coining the term artificial intelligence in the 1955 proposal for the Dartmouth Summer Research Project, written with Marvin Minsky, Nathaniel Rochester and Claude Shannon. The research gathering followed in 1956.

We usually describe this as a name given to us by computer scientists. More precisely, these were pioneers working across computing, mathematics and information theory. Their interest was in what machines might be able to do.

As Dartmouth recounts, the proposal began with the conjecture that every aspect of learning, or any other feature of intelligence, could in principle be described precisely enough for a machine to simulate it. A conjecture is something to investigate. It is not an established finding.

The name described an ambition. Over time, we have treated it as a description of the systems themselves. That shift matters because the label can imply something that still needs to be demonstrated.

Computer scientists were right to ask the question. The problem is treating their disciplinary framing as the final answer. Intelligence is not a question for computer science alone. Psychology, philosophy, neuroscience and the social sciences all have something to say about it.

The people who build a technology have expertise in its construction. That does not give their disciplinary perspective the final say on its meaning.

Artificial is the easier part

Artificial is the less contested part of the term, if we mean made by people rather than naturally occurring. It does not mean unreal or inconsequential.

But it should not encourage us to imagine these systems as independent of human society. People choose the data, build the infrastructure and decide where systems will be used. Organisations decide what they can access and what they are permitted to do.

There is a tendency to speak about AI as though it were arriving from outside society. It is being built within it. The choices and responsibilities are ours.

Intelligence remains contested

We have not arrived at a universally accepted definition of human intelligence, let alone one that settles the question for machines.

The American Psychological Association’s task-force report, Intelligence: Knowns and Unknowns, recognised that no conceptualisation achieved universal assent. It also found that intelligence tests measure useful abilities and predict some outcomes, while not easily assessing qualities such as creativity, wisdom, practical knowledge and social skill.

That does not make intelligence meaningless or impossible to measure. It means we need to be clear about what a particular definition or test includes, and what it leaves out.

For machine intelligence, Shane Legg and Marcus Hutter propose a definition centred on an agent’s ability to achieve goals across a wide range of environments. It gives researchers an explicit basis for their work. It does not resolve every other account of intelligence.

If intelligence means successful goal achievement, we can investigate that. If it means human-like understanding, we need to explain how we would recognise and assess it. Consciousness is another question again.

We cannot move between those meanings and assume we are still making the same claim.

As I have argued in We are not brains in jars, human thinking happens through our bodies and our relationships with other people. We learn by acting in the world, making mistakes and making sense of things together.

A computational account can tell us something important about intelligence without accounting for the whole of that experience. This is a transdisciplinary question, with implications well beyond engineering. Treating it as an engineering problem alone leaves too much out.

What has the system demonstrated?

A model that answers an exam question correctly has demonstrated a capability under particular conditions. That is useful evidence. It does not automatically tell us whether the capability will transfer to a different setting, or whether the system understands the question as a person does.

The same applies to fluent language. We can observe the output. The further claim that the system understands what it is saying needs its own definition and evidence.

I am not arguing that machines cannot be intelligent. I am arguing that calling them intelligent does not settle the matter.

We should be able to acknowledge substantial capabilities without treating every successful output as evidence for a much larger claim.

Why this matters for governance

The name matters because it can influence how we assign authority and responsibility.

Consider an automated hiring system. Calling it intelligent tells an applicant very little. They need to know how they were assessed, whether that assessment is valid and how they can challenge an error. Someone needs to be accountable for the decision.

Those are organisational and governance questions. The label does not answer them.

This is why I keep returning to the need to specify what we expect before we can govern a system. We need evidence about what it can do in its intended setting. We need to understand its limits and decide what authority it should have.

We do not have to settle the philosophy of intelligence before doing that work. But we should stop using the word as a substitute for it.

I doubt we will rename the field. We can, however, be more specific about the systems we are discussing. A language model, recommendation system or automated decision system gives us a clearer starting point than any general claim of intelligence.

Artificial might be right. Intelligence remains contested. The people who named the field gave us a research ambition, not a reason to stop asking what the word means.