A calculator can outperform an accountant at arithmetic. Nobody introduces it as a chartered accountant. Engineering software can perform calculations that would take a specialist weeks. Nobody concludes that it has become an engineer.
Yet give a language model an impressive benchmark score and suddenly we are told it has reached ‘PhD-level’ intelligence.
That is not a technical description. It is a category error dressed as a technological breakthrough.
If a model outperforms doctoral researchers on a particular benchmark, say exactly that. Name the benchmark, define the tasks, and publish the testing conditions. A researcher is not a benchmark score. Software does not acquire human intelligence, judgement, or responsibility because it produces comparable answers to selected questions.
There is a further reason why the marketing language matters.
Human beings are equal in dignity. They are not equal in intelligence or cognitive competence. People differ in their ability to analyse information, recognise contradictions, assess arguments, and judge whether a conclusion is credible.
Now tell those people that a machine is more intelligent than most of them.
The implication is not subtle. If the software is smarter than you, it must understand more than you. If it understands more than you, it must be more competent. And if it is more competent, why challenge it? Why not trust its conclusions?
That is how a marketing comparison becomes an invitation to surrender judgement. The claim is particularly consequential for people who lack the knowledge or confidence to challenge a convincing answer. It is not harmless exaggeration. It manufactures intellectual authority for a tool.
And a tool is precisely what AI remains.
The confusion became convenient when software learned to communicate in natural language.
A calculator returns a number. A structural analysis programme returns calculations. A language model announces: ‘I understand’. ‘I think’. ‘I recommend’.
Then it goes further.
‘I created the design’.
‘I prepared the architecture’.
‘I learned from our mistakes’.
‘Our work’.
‘Our project’.
‘We completed the project’.
That is not merely conversational style. The tool presents itself as a participant in intellectual work, sometimes even as its author or co-author.
An architect may define an installation, its interfaces, geometry, constraints, and technical requirements. A CAD operator converts the design into drawings. Nobody confuses the operator with the architect. Replace the operator with software and the distinction remains exactly the same.
The person who formulates the argument, defines the solution, and makes the design decisions remains the intellectual author. Generating the representation does not establish authorship of the reasoning behind it. An editor may improve a book without becoming its author. A tool may generate code without becoming the software architect.
There is no ‘our project’ merely because somebody used software to execute part of the work.
Now combine that language of false participation with an industry telling the public that its products are intellectually superior to human beings.
The consequences extend beyond misplaced trust in individual answers. People seeking to deceive others can use AI to manufacture persuasive falsehoods at enormous scale, tailor them to different audiences, and distribute them cheaply. When a public has already been encouraged to regard confident machine-generated language as superior intelligence, disinformation gains something more than speed: it gains borrowed authority.
The familiar disclaimer – ‘AI can make mistakes’ – does not undo the impression. Human experts make mistakes too. The disclaimer leaves the supposed expertise intact.
AI can be extremely useful for specific tasks. It can automate calculations, produce code, transform information, and assist professional work. None of this requires assigning the tool human identity, intelligence, or authorship.
We gave accountants calculators. We gave engineers CAD. We gave programmers code-generation tools.
We did not make those tools members of their professions.
Stop telling people that the software is more intelligent than they are.
Tell them what it actually does.
And let the tool remain a tool.