Artificial Intelligence says “calm down dear”
I attended a fascinating lecture this week about ethics in artificial intelligence (AI) and robotics. Prof John Murray from the University of Hull shared with us the story of the Babylon Health app which uses AI to power a symptom checking chatbot.
So far, so efficient but the Sunday Times story “It’s hysteria not a heart attack GP app tells women” revealed that that the chatbot gave different diagnoses for men and women entering the same potential heart attack symptoms. The men were told to go to A&E whereas the women were told they were probably having a panic attack — shades of “calm down dear”! So can AI be patronising and dangerous or is it just in need of some fine tuning?
It’s well known that heart attack symptoms are different in women and men but that doesn’t fully explain the discrepancy. Could it be that the data the AI is using to drive its Q&As has bias built in? As Babylon Health themselves said in response to the Sunday Times article: “The medical research community is well aware that a lot of evidence has bias, partly because so much work has focused on white males.”
The thing is, we expect machines to operate to a higher level of efficiency than people; we expect a self-driving car, a self-flying plane or an automated doctor to be perfect whilst we accept that their human equivalents are fallible and will make mistakes. Ethics and governance needs to be developed hand in hand with the new technology so the debate is had openly about what standards AI needs to reach. And we certainly don’t expect — or deserve — to be patronised by machines.
A highly impactful but woefully neglected area of women’s health is the menopause. It’s such a huge change in a woman’s body, often with profound effects on mental health, self-confidence and self-perception, that we (50% of the population) need to know that it is starting, how far along the pathway we are and what we can do (beyond herbal remedies) to alleviate the symptoms and crack on with our lives. In recent years it has become more acceptable to talk about the menopause, but there’s still a cloud of Victorian “don’t-talk-about-the-change” embarrassment around the subject. Not talking about it doesn’t mean nothing is happening but it may be one of the reasons that so little pharmacology is focused on this area.
There’s a fantastic opportunity for AI to help detect the subtle physiological changes that come with the menopause.
The standard GP tests are hopeless at detecting hormonal changes until well into the menopause, by which time too many women are lost in a fog of plunging self-confidence, anxiety and years without a good night’s sleep for which they have no obvious cause so they are left without diagnosis or remedies.
Looking for scholarly articles on AI and the menopause just now produced very little. One article (which had not been peer reviewed) suggested explainable AI gets as good results in detecting the menopause from samples taken from people’s skin as is achieved by more invasive sampling.
If we had better detection of the signs of the early-to-middle stages of the menopause, we could be on the path to better balance in our lives and to our middle years being less of a torment.
So come on all you researchers and healthcare businesses — where else in the 21st Century are you going to find a ready-made but totally under-served market of 50% of the population?
