“The potential for AI to massively transform our lives outside healthcare as well as in it cannot be overstated”

But the positive reviews of Ernie come with one important caveat. While it confidently discusses Western current affairs – albeit with a marked Chinese bias – when the topic turns to contemporary Chinese politics, Ernie’s previous certainty of tone pivots into marked diffidence, declaring itself unable to answer the question or, all too often, suggesting that “we talk about something else.”

Sourcing its answers primarily from officially-sanctioned Chinese websites, and with its operation closely overseen by the Chinese state, Ernie has learned to avoid saying anything that is contrary to the acceptable local line.

The potential for AI to massively transform our lives outside healthcare as well as in it cannot be understated

 

The potential contribution of the new artificial intelligence (AI) technology to improve healthcare cannot be overstated – there is, for example, some research indicating that not only is AI better at diagnosing disease than humans, it is also capable of communicating that information to patients more empathetically than the average clinician. 

Nevertheless, as Ernie shows, its potential limitations are becoming increasingly clear. AI has the capability of sifting the learning from huge data sets, but its functionality will always be limited by the data it is analysing and the rules by which it operates. If the data from which it learns is partial and the rules have inbuilt biases, the results – however impressive – will be flawed.

In medicine, this is not merely a theoretical problem but a fundamental one. There is a particular issue when it comes to AI models’ ability to deal with the medical needs of ethnically diverse populations. A panel of American medical experts assembled by Yale University reported in December that there are “numerous biased algorithms that require racial or ethnic minorities to be considerably more ill than their white counterparts to receive the same diagnosis, treatment or resources.”

As AI models increasingly start to drive eye care in the UK, this is a real issue we need to grapple with

 

As AI models increasingly start to drive eye care in the UK, this is a real issue we need to grapple with. Take, for example, the normative databases used with optical coherence tomography, where the size and breadth are often small and lacking in diversity. Given the fact that glaucoma prevalence varies significantly within parts of the population, it is essential that the databases on which such machinery draws should be as wide as possible.

The potential for AI to massively transform our lives outside healthcare as well as in it cannot be overstated. But both in healthcare and beyond, its use must be to support human activity rather than replace it. At the AOP, we are already exploring extended ways of using our automated chatbot functions to supplement – and again, the word is supplement not replace – the service our members get from our staff. But that is only scratching the surface of what it will be able to provide: in a few years, this column will be written by a machine far more capable than I am of basing an argument on safer foundations than those provided by a 50-year-old Benny Hill song. And for that, we can all be profoundly grateful.