Clinically Intelligent is written by Juwon Akinyande in a personal capacity and is not affiliated with, endorsed by, or representative of Barts Health NHS Trust or any other NHS organisation. Content is for general educational purposes only. It does not constitute clinical, legal, or information governance advice. Before applying any guidance to your own practice, consult your Trust information governance lead, your Caldicott Guardian, your line manager, and your professional body.
Good to be back. This one took some time to build properly. But here we are. Let’s get into it.
A few people saw a poster I made and said "that's AI."
They were not completely wrong. AI was involved.
But they were not completely right either.
I knew what I wanted the poster to do. I did not want another standard poster that people looked at once and forgot. It needed to catch the eye, but still contain all the information somebody needed without them having to find the email that came with it.
I looked for inspiration, picked out elements I liked, wrote the text and decided what I wanted the poster to look like.
Then I reached the limit of what I could make myself in Canva.
Some of the visual ideas in my head were difficult to create with the tools I had, so I used AI to generate some of the imagery, create silhouettes and polish parts of the design
The idea was mine. The message was mine. Most of the design decisions were mine. AI helped me execute parts I could not easily create myself.
It helped put what was in my head onto the page.
So when somebody looked at the finished poster and said "that's AI", I understood what they meant.
But it also made me wonder:
When does "I used AI" become "AI made this?"
I think the problem is we are trying to describe a new kind of creative process using either-or language. Human or AI. Real or generated. Mine or the machine's.
Creation was never that simple.
Creation is a chain
We talk about creation like it happens at the moment someone puts pen to paper. It does not.
You imagine something. You decide what you are trying to say. You make it. You look at what you have made and decide it is wrong. You change it. Eventually you decide: that is what I meant.
The making is only one part of it. So is conception. So is judgement. So is selection. So is revision.
AI does not simply enter a piece of work. It enters somewhere along that chain.
And where it enters matters.
Legal scholar Ezieddin Elmahjub describes this as a "broken causal chain." His argument is that with traditional tools, human intent is translated relatively directly into expression. Generative AI inserts a probabilistic system between the person's intention and the exact words, pixels or sounds that appear. A prompt can strongly influence a work without determining its final expressive form.
That gap matters when we are trying to work out who made what.
If I use a paintbrush, the brush does not decide what the face should look like.
If I ask an image model for "a bird flying over London at sunset," I have specified something important. I have not chosen the exact angle of its wings, the shape of the clouds, the lighting or the colours. The model supplies much of that.
But there is another part to this.
The fact that AI performed part of the making does not mean the human performed none of the creating.
Where did the judgement come from?
Imagine two people both produce a 1,000 word article using AI.
The first spends an hour thinking. They know the argument they want to make. They write notes. They decide what they believe. They work out the structure. They identify the examples. They know the conclusion.
Then they use AI to help turn some of that thinking into cleaner prose.
The second types:
"Write me an interesting 1,000 word article about AI."
Both used AI.
That statement tells you almost nothing.
The relationship between the person and the machine was completely different.
In the first case, AI entered the chain after most of the thinking had already happened. In the second, AI entered before the person had worked out what they wanted to think or say.
That matters.
There is a difference between using AI to express a thought and using AI to supply the thought.
So the question is not really "how much AI?"
It is: where did the important creative decisions come from?
Who conceived the central idea? Who decided what the work was trying to achieve? Who determined its structure? Who produced its expressive details? Who evaluated the alternatives? Who made the final judgement?
That gives you something more useful than arguing over whether something is "AI."
It gives you something closer to the work's provenance: where it came from and how it was made.
We need better words
We already have phrases like AI-generated and AI-assisted. What we do not have is shared everyday language that tells another person where AI sat in the process.
Three distinctions would get us surprisingly far.
Made with AI
The human controls the substantive creative process and uses AI for specific parts of it.
You write the article. AI checks the grammar.
You design the presentation. AI removes the background from an image.
You create the poster. AI generates an element that you place within it.
The defining feature is not that AI did very little work. It is that the important conception, direction and judgement stayed with the human.
That is how I would describe my poster.
Made alongside AI
The machine contributes substantive material.
You provide the argument, evidence, notes and direction. AI drafts. You rewrite. It suggests alternatives. You reject some, keep others, reorganise the structure and continue until the final work says what you actually believe.
The final output emerged through repeated interaction between human judgement and machine generation.
Made by AI
AI supplies most of the substantive expression from limited human direction.
"Write an article about burnout."
"Make me a poster for a hospital conference."
"Generate a photograph of London in 2050."
The person may choose between outputs or edit lightly afterwards, so even here human judgement has not fully disappeared.
An accurate description might sometimes be: AI-generated, human-curated.
That is more informative than either "I made it" or "the AI made it."
None of these categories is automatically good or bad.
They describe where human judgement and machine generation sat in relation to one another.
Why our current language damages honesty
If saying "I used AI" makes people treat your work as fraudulent, lazy or less legitimate, people will eventually stop saying it.
Some already do.
AI use gets hidden because "that's AI" has become an accusation rather than a description.
That is the opposite of what professional environments should want.
We should want people to be able to explain what actually happened.
"I wrote it and used AI to edit the language."
"I supplied the analysis and evidence. AI drafted an initial version that I substantially rewrote."
"AI generated the first draft and I reviewed it."
"I designed the piece and used AI to create some of the imagery."
Those statements tell you something.
A simple yes or no declaration does not.
Transparency becomes easier when honesty does not automatically trigger judgement.
This matters more in healthcare
Healthcare already runs on provenance.
Who assessed the patient. Who interpreted the result. Who made the decision. Who documented it. Who checked it. Who is accountable for what happens next.
AI does not remove those questions.
It makes them more important.
Consider two discharge summaries.
In the first, a clinician writes the summary and software corrects grammar.
In the second, an AI system reads the record, generates the clinical narrative, selects what it considers important and proposes the final wording. A clinician then reviews it.
The finished documents might look almost identical.
The production processes are not.
Those workflows create different risks and require different forms of oversight.
An AI-generated clinical summary might omit something important, introduce an unsupported detail, or frame information in a way the clinician would not have chosen.
And we are not immune to automation bias. Once a polished paragraph is sitting in front of us, it can be easier to approve than to reconstruct the reasoning independently.
So "human reviewed" is not enough information either.
How was it reviewed? Against what? Who reviewed it? Who remains responsible?
Accountability is not determined only by whose name appears at the bottom of the page. It is shaped by how the page came into existence.
That is why language about AI involvement is not just paperwork.
It can be part of governance.
Not surveillance. Not policing whether someone used a tool.
Simply making the chain of responsibility visible.
Sometimes the friction is the thinking
There is another reason where AI enters matters.
It affects us.
If every time we face an empty page we immediately ask AI what should go on it, something changes.
We stop struggling with the blank page.
And that struggle has value. It is often where we discover that we do not yet know what we think.
Writing makes you choose. Design makes you decide what matters. Analysis makes you confront what does not add up.
Sometimes AI removes friction that should be removed.
Sometimes the friction is the thinking.
This has been one of the hardest issues I've written.
Partly because I could not find the right words for the difference between something made with AI and something made by AI.
It only started making sense when I wrote it out.
Before that, it would have been much easier to open an AI tool and ask it for the answer.
Working it out on paper
It is becoming very easy to do that the moment an idea is not fully formed. I do it too.
But sometimes you need to give yourself a chance to work something out first.
Write badly. Draw circles. Cross things out. Change your mind.
Somewhere in that mess, the idea starts becoming yours.
AI can help afterwards.
I just do not want us to lose confidence in our own ability to begin.
That does not mean avoid AI. It means be deliberate about where you invite it in.
If I have already wrestled with the argument, AI might help me express it faster. If I ask AI for the argument before I have wrestled with anything, I may gain speed while losing something else.
The easiest part of thinking to outsource may be the part we most needed to practise.
That matters in education, in clinical reasoning, in any profession where the output is only part of the purpose.
Sometimes we need the answer.
Sometimes we need to become capable of producing it ourselves.
Those are different objectives.
So was my poster "AI"?
Not in the way that question usually implies.
AI generated some of the visual material. I conceived the purpose of the poster, developed its message, designed the composition, selected what belonged, rejected what did not and made the final decisions.
That is why asking whether the poster was simply "AI" does not tell us enough.
The more accurate description is:
I made the poster with AI assistance.
That sentence does not diminish what AI did.
It does not exaggerate what I did.
It describes the process.
That is what I want us to get better at.
One thing worth naming
In Issue 08 I wrote that AI can help you think faster but it cannot think for you.
I still believe that.
But I would add something now.
The important question is not only what you did with AI.
It is where you allowed AI to enter your thinking.
Did it polish something you had already created?
Did it help you develop something you were creating?
Or did it decide what the thing should be before you had decided yourself?
Those are different relationships with the technology.
We should have language capable of saying so.
The future is unlikely to divide neatly into things made by humans and things made by machines.
Most of us will increasingly make things with machines.
The important difference will be whether we can still see where the human judgement lives.
And whether the person using the system can explain, clearly and honestly:
This is what I thought. This is what the machine contributed. This is what I decided. And this is what I am willing to take responsibility for.
That, to me, is authorship in the age of AI.
That is all for Issue 16. Next week, back to a tool.
The AI tools discussed in Clinically Intelligent are consumer products. They have not been independently assessed by the author against DCB0129 or DCB0160 clinical risk management standards, and they may not be approved for clinical use by your employer. Before using any tool described in this newsletter in connection with your clinical practice, you must satisfy yourself that its use is permitted under your Trust information governance policy, your DSP Toolkit obligations, your professional registration requirements, and any applicable contractual terms with your employer. The author accepts no liability for use of any tool or workflow described in this publication. Patient identifiable information must not be entered into any consumer AI tool under any circumstances, irrespective of any guidance contained in this newslet