The potential for AI to cut waiting times is huge. Clearing the backlog will also take less time and the result for people will be less risk of mis diagnosis and also the risk of their illnesses getting worse before they are treated will be mitigated. This could save the NHS piles of cash, which they can then use to recruit more staff and pick itself up off of the canvas.
The NHS will still need consultants to apply their vast experience and knowledge whenever it is needed and to ensure that AI is doing its job properly. AI isn’t the “silver bullet” but it is another tool in the arsenal that will help transform our ailing health system.
I just hope they don’t train it on the WebMD data.
Query: “Splinter in finger, what do?”
WebMD AI: “Obviously cancer”
I’m no computer scientist, but I always feel like this is actually the most dangerous aspect of AI.
One of the key things of any safety critical system is determinism – you should be able to put in a set of data and reliably get the same results. AI (In the sense of things like LLM) is not usually deterministic. This means two people can present with the same symptoms and get different results. Or you give it 2 X-Rays and get 2 results, etc.
That said, more traditional machine learning, employed deterministicly, absolutely 100%.
Of course, part of the issue is that there isn’t a universal classification of these tech’s, and what people say when they mean AI can differ a lot. And a lot of the time it’s nothing more than a traditional algorithm.
Creating AI models that help doctors interpret test results is pretty promising.
Creating an AI chatbot to which you must say the magic keywords before you can speak to a doctor is not.
Yes, this is probably right but we must remember that trials using IBM’s Watson AI for healthcare, including cancer treatments, looked promising but eventually failed despite billions of dollars spent on it and tie-ups with some of America’s top hospitals.
The Commons Committee’s report in the story looks too vague, at least as summarised. Just some truisms and obvious cautions.
5 comments
The potential for AI to cut waiting times is huge. Clearing the backlog will also take less time and the result for people will be less risk of mis diagnosis and also the risk of their illnesses getting worse before they are treated will be mitigated. This could save the NHS piles of cash, which they can then use to recruit more staff and pick itself up off of the canvas.
The NHS will still need consultants to apply their vast experience and knowledge whenever it is needed and to ensure that AI is doing its job properly. AI isn’t the “silver bullet” but it is another tool in the arsenal that will help transform our ailing health system.
I just hope they don’t train it on the WebMD data.
Query: “Splinter in finger, what do?”
WebMD AI: “Obviously cancer”
I’m no computer scientist, but I always feel like this is actually the most dangerous aspect of AI.
One of the key things of any safety critical system is determinism – you should be able to put in a set of data and reliably get the same results. AI (In the sense of things like LLM) is not usually deterministic. This means two people can present with the same symptoms and get different results. Or you give it 2 X-Rays and get 2 results, etc.
That said, more traditional machine learning, employed deterministicly, absolutely 100%.
Of course, part of the issue is that there isn’t a universal classification of these tech’s, and what people say when they mean AI can differ a lot. And a lot of the time it’s nothing more than a traditional algorithm.
Creating AI models that help doctors interpret test results is pretty promising.
Creating an AI chatbot to which you must say the magic keywords before you can speak to a doctor is not.
Yes, this is probably right but we must remember that trials using IBM’s Watson AI for healthcare, including cancer treatments, looked promising but eventually failed despite billions of dollars spent on it and tie-ups with some of America’s top hospitals.
The Commons Committee’s report in the story looks too vague, at least as summarised. Just some truisms and obvious cautions.