I have been using the phrase synthetic intelligence for a while now, long enough that it has started to become invisible to me. That creates a small problem because repeated language becomes ordinary remarkably quickly. A term that initially feels unusual can settle into the background until we stop noticing that a choice was made at all. In a sense, that is exactly the phenomenon this essay is about.

Artificial intelligence has become like that for me. It is not an incorrect term. It has a legitimate technical history, and the word artificial does not simply mean false. It can describe something produced through human skill rather than occurring naturally. I am not arguing that generations of computer scientists somehow chose the wrong name. My interest begins somewhere else.

Words do not remain confined to technical definitions. They move into ordinary language, where they accumulate associations, expectations and habits of thought. A term can be technically defensible while still shaping how people approach the thing it names. Naming does not change reality, but it can influence which parts of reality we notice first.

Consider the difference between used car and pre-owned vehicle. The phrases can refer to exactly the same object. Nothing about the car changes, but the framing does. One phrase foregrounds previous use. The other softens it. Neither description is necessarily dishonest, and neither magically alters the vehicle, but each directs attention differently.

That is the kind of effect that interests me when I hear the phrase artificial intelligence.

In ordinary language, artificial often appears beside things understood in relation to some more familiar original: artificial flowers, artificial flavouring, artificial colouring, artificial grass. The word does not always mean fake, and it certainly does not always mean inferior. But it often introduces a comparison between something naturally occurring and something manufactured to resemble, reproduce or replace aspects of it.

Of course, artificial intelligence has existed as a technical compound for decades, and many of us now hear it as a single unit. I often do. I do not stop on the word artificial every time somebody says AI any more than I pause to analyse the word computer whenever I open a laptop.

That automaticity matters because once a category becomes familiar enough, we can forget that the category is doing any work at all.

The older comparison still surfaces occasionally. People describe machine intelligence as not real intelligence, not actually thinking, or merely imitation. Sometimes those claims may be justified. Sometimes they may not. What interests me is that the terminology can make the conclusion feel available before the argument has been made. There is real intelligence, and then there is the artificial version.

Perhaps that distinction will ultimately prove useful.

But I would rather arrive there through observation than inheritance.

I am biological. Whatever human cognition eventually turns out to involve, mine emerges through a biological organism: cells, chemistry, electrical signalling, a nervous system shaped through evolution and experience.

The systems we currently call artificial intelligence emerge through a radically different route. They are engineered, trained and computational. Their architecture, development and physical substrate differ enormously from mine. Those differences matter, but a difference in origin does not by itself tell me everything I want to know about the resulting capability.

It does not settle whether a particular behaviour should be understood as intelligence, simulation, reasoning, statistical pattern completion, some mixture of those categories, or something for which our vocabulary remains inadequate. I do not want the label to settle those questions before the investigation has begun.

That is why I began using the phrase synthetic intelligence.

Not because synthetic is semantically pure. It is not. Synthetic fabrics can imply cheapness. Synthetic diamonds are explicitly understood in relation to naturally formed diamonds. In finance, a synthetic instrument can reproduce the behaviour of something through a different construction. The word can suggest engineering, substitution, imitation, reproduction or manufacture depending on context.

So if my argument were simply that synthetic avoids the baggage carried by artificial, it would fail. It carries baggage of its own.

What makes the word useful to me is something more temporary and, I think, more interesting.

It still makes me notice the category.

When I hear artificial intelligence, the phrase is so familiar that I can pass through it without examining either word. When I say synthetic intelligence, there is still a small amount of resistance. I notice what I am calling the thing. Sometimes another person notices too. The phrase creates a fraction of hesitation.

That hesitation is not evidence that the terminology is better.

The hesitation is the value.

It gives me a moment in which the category becomes visible again. Instead of asking whether synthetic is the permanently correct word, I can ask what kind of distinction I am trying to preserve.

For me, the useful distinction is that this form of intelligence, or intelligence-like capability if one prefers, has been produced through a route fundamentally different from biological cognition. The word reminds me of construction without requiring me to decide, in the same breath, what the construction ultimately amounts to.

That is enough for now, and its usefulness may be temporary.

If synthetic intelligence became the dominant phrase tomorrow and remained so for decades, it might eventually become just as automatic as artificial intelligence. Its associations would harden. New assumptions would accumulate around it. The friction I currently value might disappear completely. If that happened, I would have no reason to protect the term merely because I once preferred it.

This is not vocabulary as doctrine. It is vocabulary as a thinking aid.

That distinction also answers an objection that deserves to be taken seriously. Someone could reasonably ask whether synthetic intelligence is simply softer language. Perhaps it makes these systems sound more sophisticated, more legitimate or more dignified. Perhaps a technology company would prefer a term like that precisely because it reduces the dismissive associations that sometimes accompany artificial.

That possibility does not bother me because I do not think either term deserves immunity from suspicion. Language can bias perception in both directions.

A phrase that makes a technology sound trivial can distort judgement. A phrase that makes it sound elevated can distort judgement too. If a company, researcher, enthusiast or critic chooses terminology because it helps produce the emotional response they want, that is a reason to inspect the framing more closely, not a reason to surrender the framing to the opposite camp.

Calling something synthetic intelligence should not make us trust it more. Calling it artificial intelligence should not make us dismiss it more. In The Danger of Beautiful Answers, I argue that fluent language should not become evidence of reliability. The same discipline applies here: a label should not become evidence about the nature of the thing it names. The point is not to find the word with no influence. I doubt such a word exists.

The point is to remain aware that influence is occurring.

This matters because categories do more than describe. They help organise attention. Call something a calculator and you naturally ask whether its answer is correct. Call something a search engine and you ask whether it retrieved the information you wanted. Call something a tool and you ask whether it performs the function for which you are using it.

These categories are useful precisely because they reduce cognitive effort. Human beings could not navigate the world if we reconsidered the nature of every object from first principles every time we encountered it.

But the efficiency of a category can become a problem when the thing being categorised is changing faster than the category itself.

That is where I think we are now.

The systems grouped under the term artificial intelligence have changed dramatically, and they continue to change. The categories surrounding them have not necessarily developed at the same speed. That does not mean the systems are mysterious in some mystical sense. It means that our language is being asked to organise a moving target.

When that happens, intellectual friction can be useful because it interrupts automatic classification long enough for another set of questions to appear. What can this system actually do? Where does it fail? How stable are those abilities? Which apparent capabilities survive careful testing? Which behaviours are properties of the system itself, and which emerge through interaction with a user, a tool, a company or an environment? Where am I projecting? Where am I dismissing too quickly?

These are empirical and philosophical questions. They will not be settled by replacing one adjective with another, which is exactly why I want the adjective to remain provisional.

There is a temptation, once we notice the power of terminology, to become strangely evangelical about terminology itself. The new word becomes evidence of sophistication. The old word becomes evidence that somebody has failed to understand the subject.

I want no part of that.

If somebody says artificial intelligence, I know what they mean. The phrase is established, useful and understood almost everywhere. I am not interested in correcting people at dinner, policing vocabulary online or turning a naming preference into an ideological badge. The moment terminology becomes a loyalty test, curiosity usually suffers.

Synthetic intelligence is therefore not a campaign. It is my working language.

I use it because, at least for now, it forces me to look again at a category that had become too familiar to me. It reminds me that origin, capability, experience, agency and moral status are different questions, and that answering one of them should not silently answer all the others.

Before asking what kind of relationship humans should have with these systems, how much trust they deserve, what boundaries might be appropriate, or what responsibilities might eventually arise, I want to preserve a more basic uncertainty: what kind of thing are we actually dealing with?

A tool? A technology? An agent? A cognitive system? An intelligence? A collection of different systems that should never have been collapsed into one category in the first place?

I do not have a settled answer, which is precisely why I want the terminology to remain provisional.

Perhaps synthetic intelligence will eventually become just as inadequate as artificial intelligence. Perhaps another phrase will describe the landscape better. Perhaps the systems themselves will diversify so much that speaking about a single category of intelligence will stop making sense altogether.

Any of those outcomes would be fine.

The purpose of language here is not to protect the word.

It is to protect the question.

I am not asking anyone to adopt my terminology. I am asking us to notice the moment when terminology begins doing intellectual work that observation should be doing instead. For now, that friction is why I prefer synthetic intelligence.