I came across an observation this morning that stopped me in my tracks. (Wyn Hopkins post on LinkedIn)
Someone had apparently suggested using AI to make AI-generated content sound more human.
The reaction was that this was wrong on so many levels.
And I agree.
But the more I thought about it, the more interesting the question became.
Because perhaps the problem isn’t AI at all.
Perhaps AI has simply made an existing problem much easier to see.
Two Very Different Uses of AI
I think we need to distinguish between two fundamentally different things that are currently being bundled together under the heading AI-generated content.
Type One: Someone knows very little about a subject and uses AI to manufacture the appearance of expertise.
Type Two: Someone possesses knowledge, experience or ideas of their own and uses AI to communicate them more effectively.
Those aren’t remotely the same thing.
Imagine I know absolutely nothing about astrophysics.
I can nevertheless tell an AI:
You are an expert astrophysicist. Write an authoritative article explaining an advanced area of astrophysics to an educated general audience.
Within seconds, I have something impressive.
I could publish it under my name.
I could make a video from it.
I could perhaps even build an audience around it.
The problem isn’t that AI wrote the words.
The problem is that I am borrowing the appearance of authority that I don’t possess.
Now consider the opposite situation.
Suppose somebody has spent 30 years doing something unusual. They possess knowledge that isn’t easily found in textbooks. Perhaps they aren’t a particularly good writer. Perhaps English isn’t their first language. Perhaps they struggle to structure their thoughts.
They talk for half an hour.
AI organises those thoughts, removes the repetition, clarifies the argument and turns them into an article.
The words have been assisted by AI.
But the knowledge hasn’t.
That distinction matters enormously.
Stephen Hawking and the Communication Machine
There is an extreme example that helps illuminate the principle.
Stephen Hawking possessed extraordinary scientific knowledge, yet motor neurone disease progressively deprived him of the conventional physical mechanisms for communicating it.
Technology became part of the bridge between Hawking’s mind and the rest of us.
Nobody would seriously argue that because technology mediated the communication, the ideas somehow ceased to be Hawking’s.
Quite the opposite.
We should be grateful that technology enabled knowledge trapped inside one human mind to become accessible to millions of others.
AI potentially performs a much less dramatic version of the same function.
The important question isn’t:
Did technology help produce these words?
The important question is:
Where did the knowledge come from?
And that leads me somewhere rather uncomfortable.
Haven’t Humans Been Doing This Already?
If we’re worried about AI manufacturing the appearance of expertise, shouldn’t we ask whether social media has been doing exactly that for years?
AI can make mediocre knowledge sound authoritative.
But so can a professional studio.
So can excellent lighting.
So can editing.
So can graphics.
So can a beautifully rehearsed delivery.
So can confidence.
So can charisma.
So can a fantastic haircut.
None of those things is inherently wrong.
But neither is AI.
The question is what they are being used to amplify.
Are they amplifying expertise — or manufacturing its appearance?
That is a much more interesting question.
The Authority Problem
This becomes particularly important with professional education.
You will frequently hear expressions such as:
I always do this when I encounter this problem.
That sounds reassuring.
It implies experience.
But there is a question we rarely ask:
Where exactly did you encounter this problem?
Suppose someone demonstrates how to combine information from many spreadsheets.
The technique works.
The presentation is excellent.
Hundreds of thousands of people watch it.
Perhaps millions.
But that still doesn’t answer the professional questions.
Why were those spreadsheets being combined?
What business process produced them?
How many users were involved?
Who was responsible for the overall process?
What happened to the information afterwards?
Was the solution designed for one analyst — or an entire organisation?
What were the controls?
What were the audit requirements?
Who owned the data?
And perhaps most importantly:
If the person demonstrating the solution had responsibility for the entire business process rather than one spreadsheet within it, would they still have designed it that way?
That is where technical competence and professional authority begin to separate.
Local Expertise and Global Responsibility
This distinction is particularly important with Excel.
Almost everything demonstrated in popular Excel education starts from the perspective of the individual spreadsheet user.
Here are some files.
Combine them.
Here is some messy data.
Clean it.
Here is a complicated formula.
Simplify it.
Here is a repetitive task.
Automate it.
All perfectly reasonable.
At the local level.
But move several floors upstairs metaphorically — into the CFO’s office — and the question changes.
The CFO isn’t primarily interested in how cleverly one employee processes 47 spreadsheets.
The CFO should be asking:
Why do we have 47 disconnected spreadsheets containing fragments of the same organisational information in the first place?
That changes the problem completely.
Instead of improving the point-to-point movement of information between spreadsheets, perhaps we should be questioning the architecture that requires all that movement.
Instead of continually improving the spokes, perhaps we need a hub.
That is the difference between solving a local spreadsheet problem and solving a global information problem.
And millions of views don’t resolve that distinction.
Popularity Is Not Professional Validation
This is where social media introduces another difficulty.
We naturally treat popularity as evidence.
A video has two million views.
Therefore, it must be important.
Perhaps.
But two million views tell us something very specific:
Two million viewing events occurred.
They don’t tell us that the technique represents the best architecture for a multinational organisation.
They don’t tell us that a CFO should approve it.
They don’t tell us that the creator has ever been responsible for implementing such a process.
They don’t even tell us that the viewers themselves were in a position to judge those things.
Social media rewards content that is useful, entertaining, understandable and engaging.
Professional responsibility requires something else.
It requires context.
Judgement.
Consequences.
Experience.
And sometimes the ability to say:
Don’t solve that problem. Eliminate the reason the problem exists.
Rearranging the Deck Chairs — Beautifully
There is an uncomfortable analogy here.
Imagine the Titanic.
The deck chairs are badly organised.
Someone produces an absolutely superb tutorial showing the most efficient method ever devised for arranging them.
It’s beautifully filmed.
The lighting is magnificent.
The presenter is charismatic.
The before-and-after transformation is astonishing.
Ten million views.
Everyone loves it.
And the ship is still sinking.
That doesn’t mean the deck-chair tutorial is technically incorrect.
That’s precisely the point.
You can be completely correct about the solution while being completely wrong about the problem.
At an individual-user level, many spreadsheet techniques are ingenious.
At organisational level, they may merely make a fundamentally poor architecture more efficient.
The higher your level of responsibility, the more important that distinction becomes.
Which Brings Me Back to AI
And now we arrive at the rather delicious irony of this article.
I used AI to help write it.
AI has made this sound more like a polished article than the rambling monologue from which it originated.
But AI didn’t decide what I thought.
I did.
I didn’t ask:
Write me a provocative argument about AI, social media and Excel.
I spoke the argument.
I supplied the examples.
I made the connections.
I expressed the doubts.
I provided the professional perspective.
AI became part editor, part organiser and part writing assistant.
And that brings me back to the distinction with which I started.
The question isn’t whether AI was involved.
The question is:
What existed before the AI arrived?
Was there knowledge?
Was there experience?
Was there an argument?
Was there an original observation?
Was there something worth communicating?
Or was there simply a prompt asking a machine to manufacture the appearance that there was?
The New Test of Authenticity
Perhaps, therefore, we’re asking the wrong question when we encounter AI-assisted content.
Don’t ask:
Did AI write this?
Ask:
Could the person whose name is on it defend it without AI?
Could they explain why they believe it?
Could they answer questions about it?
Could they demonstrate it?
Could they describe where they learned it?
Could they explain where it fails?
Could they defend the decisions behind it when somebody with greater professional responsibility challenges them?
That test works remarkably well.
And interestingly, it doesn’t only expose artificial intelligence.
It exposes artificial authority.
Because AI didn’t invent the possibility of looking knowledgeable without possessing knowledge.
Social media perfected that long before generative AI arrived.
AI has simply made the trick available to everybody.
And perhaps that is why it makes us so uncomfortable.
It hasn’t created the problem.
It has held up a mirror to it.
Add comment