With all the debate going on it seemed right to tackle the question of AI and in particular Babylon’s grand reveal yesterday.
The stage show was to accompany the latest marketing, not peer-reviewed and published, but designed to look like a scientific paper. The point is that they have trained a computer to pass an exam, for MRCGP.
Exams are necessary but not sufficient to be a GP, as I’m sure they would agree with RCGP, but what have they really achieved with 100 made up vignettes and patients played by GPs? Others will answer much better than me on the safety of the process (follow this brilliant thread by @DrMurphy).
I’ve done my own trial of the AI chatbot based on two diseases I’ve personally had in recent months.
“Toenails brown and broken” I start. “Please rephrase…”
“Brown and broken toenails”, and so on. Absolutely no idea from Babylon, who end up asking, “Do you have any other symptoms?”
So I move on to my next trial:
“Wrist pain”. At least they recognise this, and there follows over four minutes a series of 39 questions, of which only 3 seem to me to be relevant, and the diagnosis comes out as:
“8/10 broken bone in the lower arm. Go to A&E”. I answered everything honestly. What I really had was tenusynovitis, tendonitis of the wrist, and I guess rather more common than a broken arm.
The way we do this with askmyGP is to let the patient type in on the very first screen whatever they want, then search for self-care advice. Try it yourself on
Try anything, medical, colloquial, badly spelled, phrases, anything. It’s not perfect but for example my trials above got me in two clicks to fungal nail infection, and the other to wrist pain where tendonitis was one possibility.
I could dress this up as AI and call it the answer to everything, but really it’s just our own algorithm to search NHS Choices better than its own search. It works, and crucially it’s very fast, much faster than having to register wtih all the details and answer 39 questions one after the other. Remember that most patients’ first concern is speed and convenience and this search costs nothing.
We had a clever history taking algorithm in our version 1 software, and very good it was too, but not good enough. Patients got bored and GPs got fed up with too much irrelevant information.
The much simpler interface, respectful of patients’ ability to express themselves, has proven hugely more popular and that has enabled us to move over half the demand online with practices that really understand the benefits.
You have no doubt heard of Elon Musk, technology billionaire and founder of SpaceX and Tesla motors. On that company’s calamitous production problems, flowing from their overambitious automation, he said last month, “My mistake. Humans are underrated.”
You probably haven’t heard of Dr Ellen Stofan, former NASA chief scientist, who said last year they are sending a human to Mars because they’ll get more information, sooner and cheaper than by sending a robot.
I’m an engineer, as keen as anyone on the benefits of technology for humankind, but to get benefits we have to understand how computers can help, not pretend with smoke and mirrors. They are good at searches, communications and analytics, simple and repetitive tasks. For the tricky stuff, we need HI, human intelligence.
Who knows what they might do in some unknown future, but we have a problem right now in general practice, and it’s not a lack of intelligence.
There’s a fundamental difference between Babylon’s start point and ours. They work from DISEASE and have put together an algorithm to try to convert Q&A into diagnoses.
We work from DEMAND and very simply get it to the right clinician to triage in seconds and decide how to care.
That is only the start and tomorrow I’ll discuss our Systems Thinking approach to intervention, in which we’ll see that technology is but a small part.
Time will tell which gives the greatest benefit. but for a taster of how much can change in only a couple of weeks just listen to this interview with Dr Dave Triska .
Harry Longman
PS If you’re on Twitter, click to follow Dave @dave_dlt for some moving reports of the change they have undergone.
PPS Did you hear about the latest AC – Artificial Caring?
Babylon/GP at Hand – AI vs HI
With all the debate going on it seemed right to tackle the question of AI and in particular Babylon’s grand reveal yesterday.
The stage show was to accompany the latest marketing, not peer-reviewed and published, but designed to look like a scientific paper. The point is that they have trained a computer to pass an exam, for MRCGP.
Exams are necessary but not sufficient to be a GP, as I’m sure they would agree with RCGP, but what have they really achieved with 100 made up vignettes and patients played by GPs? Others will answer much better than me on the safety of the process (follow this brilliant thread by @DrMurphy).
I’ve done my own trial of the AI chatbot based on two diseases I’ve personally had in recent months.
“Toenails brown and broken” I start. “Please rephrase…”
“Brown and broken toenails”, and so on. Absolutely no idea from Babylon, who end up asking, “Do you have any other symptoms?”
So I move on to my next trial:
“Wrist pain”. At least they recognise this, and there follows over four minutes a series of 39 questions, of which only 3 seem to me to be relevant, and the diagnosis comes out as:
“8/10 broken bone in the lower arm. Go to A&E”. I answered everything honestly. What I really had was tenusynovitis, tendonitis of the wrist, and I guess rather more common than a broken arm.
The way we do this with askmyGP is to let the patient type in on the very first screen whatever they want, then search for self-care advice. Try it yourself on
Bramley Demo Surgery
Try anything, medical, colloquial, badly spelled, phrases, anything. It’s not perfect but for example my trials above got me in two clicks to fungal nail infection, and the other to wrist pain where tendonitis was one possibility.
I could dress this up as AI and call it the answer to everything, but really it’s just our own algorithm to search NHS Choices better than its own search. It works, and crucially it’s very fast, much faster than having to register wtih all the details and answer 39 questions one after the other. Remember that most patients’ first concern is speed and convenience and this search costs nothing.
We had a clever history taking algorithm in our version 1 software, and very good it was too, but not good enough. Patients got bored and GPs got fed up with too much irrelevant information.
The much simpler interface, respectful of patients’ ability to express themselves, has proven hugely more popular and that has enabled us to move over half the demand online with practices that really understand the benefits.
You have no doubt heard of Elon Musk, technology billionaire and founder of SpaceX and Tesla motors. On that company’s calamitous production problems, flowing from their overambitious automation, he said last month, “My mistake. Humans are underrated.”
You probably haven’t heard of Dr Ellen Stofan, former NASA chief scientist, who said last year they are sending a human to Mars because they’ll get more information, sooner and cheaper than by sending a robot.
I’m an engineer, as keen as anyone on the benefits of technology for humankind, but to get benefits we have to understand how computers can help, not pretend with smoke and mirrors. They are good at searches, communications and analytics, simple and repetitive tasks. For the tricky stuff, we need HI, human intelligence.
Who knows what they might do in some unknown future, but we have a problem right now in general practice, and it’s not a lack of intelligence.
There’s a fundamental difference between Babylon’s start point and ours. They work from DISEASE and have put together an algorithm to try to convert Q&A into diagnoses.
We work from DEMAND and very simply get it to the right clinician to triage in seconds and decide how to care.
That is only the start and tomorrow I’ll discuss our Systems Thinking approach to intervention, in which we’ll see that technology is but a small part.
Time will tell which gives the greatest benefit. but for a taster of how much can change in only a couple of weeks just listen to this interview with Dr Dave Triska .
Harry Longman
PS If you’re on Twitter, click to follow Dave @dave_dlt for some moving reports of the change they have undergone.
PPS Did you hear about the latest AC – Artificial Caring?
What you can read next
Dear Mr Hunt
Christmas Peace Dividend
How do patients want help from their GP? The data.