Newsletter Monday
Hello Reader,
I’ve been thinking about an announcement that landed in my inbox this week.
It will undoubtedly be the source of a lot of headlines for slow news days.
The government has announced £85 million to fund twelve projects exploring new ways to tackle obesity, with artificial intelligence playing a central role in several of them.
One of the projects is here on my own patch in Leicester, Leicestershire and Rutland, looking at how AI might help people stay engaged with weight management programmes after starting medication.
On the face of it, I think that’s encouraging.
Obesity is one of the biggest health challenges we face and we desperately need to test new ideas rather than simply continuing to do what we’ve always done.
The headline figure sounds enormous.
In reality, it isn’t.
The NHS spends somewhere in the region of £11 billion every year on obesity-related illness.
Against that backdrop, £50 million of government funding is relatively modest. These are pilot projects designed to see what works, not programmes that are going to transform the nation’s health overnight.
The other headline you’ll inevitably see is that £35 million is coming from Eli Lilly.
Eli Lilly makes Mounjaro.
Cue the outrage.
Personally, I struggle to get terribly excited about that part.
Of course a pharmaceutical company has a commercial interest in obesity. We cannot run a modern health service without pharmaceutical companies, and whilst £35 million sounds like a huge amount to most of us, it is little more than loose change to a global business of Lilly’s size. The important question isn’t who helped fund the projects.
It’s whether the projects produce something genuinely useful.
What interested me far more was the AI.
AI is here to stay
I attended two separate events last week where conversations made me think of its strengths - and its limits - really lie.
I attended a parliamentary roundtable on health misinformation, and someone described approaching a technology company about making a relatively small change to the way their platform promoted health content.
As I understood it, the response was essentially, “Yes, we could build it differently… but why would we? It would cost more.”
I don’t tell that story because I think tech companies are evil.
They’re doing exactly what businesses are designed to do.
They’re optimising.
Algorithms are brilliant at recognising patterns.
If people pause on a certain type of content, they’ll be shown more of it.
That’s incredibly efficient.
The difficulty is that healthcare isn’t simply a pattern recognition exercise.
Then, just a few days later, I was sitting at a book launch listening to a GP who also works developing a women’s health app.
She described trying to persuade the developers that the app needed to account for cultural differences in how women describe menstrual symptoms.
Some South Asian languages don’t even have direct equivalents for words we routinely use in clinic, such as “flooding.”
If your app asks the wrong question, or assumes everyone uses the same language and concepts, you’ve already lost valuable clinical information before the conversation has even begun.
Again, the response she encountered wasn’t hostility.
It was indifference.
Because making software cope with all those nuances is expensive.
Optimisation suddenly becomes much harder.
Listening to both conversations, I realised they were describing exactly the same problem I see developing over my years as a GP.
Healthcare has become incredibly good at dealing with the straightforward.
We have pharmacists, physiotherapists, physician associates, online consultations, self-referral pathways, AI scribes and increasingly sophisticated digital tools.
None of those are bad ideas.
In fact, many of them are excellent.
But they all have the same effect.
They remove the simple work.
What’s left is the messy middle.
- The patient who doesn’t have the vocabulary to explain what’s wrong.
- The person whose blood tests are normal but whose life is falling apart.
- The patient who asks about weight loss but is actually grieving.
The more we automate, the more valuable that remaining human work becomes.
It's only as good as the question you ask
There’s another limitation that I suspect we’ll all become more aware of over the next few years.
AI can only respond to the question you ask it.
If you don’t know what you don’t know, you may never ask the right question in the first place.
Medicine has always involved recognising the thing that wasn’t obvious.
The diagnosis the patient hadn’t considered.
The question they didn’t know they needed to ask.
No app can reliably replace that.
None of this makes me anti-progress.
Quite the opposite.
I suspect projects like these may become the first time many older adults genuinely experience the benefits of AI in healthcare, and that excites me.
Used well, AI can make information more accessible, reduce admin, improve continuity and help people between appointments far better than just Googling something.
Those are real advantages.
But healthcare has always run on something that software struggles to create.
Trust.
Not because humans know everything.
But because good clinicians recognise when the pattern doesn’t quite fit.
That’s something I still wouldn’t want to outsource.
