I can’t wait to make an appointment with Apple’s AI doctor

Macworld

With Apple having delayed its most ambitious Apple Intelligence features, you might find yourself thinking: it’s probably best if the company just focuses on other endeavors for the time being, like a rumored iOS redesign or even pushing forward its Vision Pro platform. Nope, not so much. According to a report from Bloomberg’s Mark Gurman, Apple is embarking on future AI-based features, including most prominently an “AI-based doctor” dubbed Project Mulberry.

Yep: Just when you think you’re done talking about AI, they pull you back in.

It ain’t called ‘hardware’ because it’s easy

On the face of it, this venture is no surprise. Apple’s been heavily invested in health technology for years—I’ve argued in the past that it’s a signature endeavor of Tim Cook’s tenure as Apple CEO. From the Apple Watch to the Health app, Apple has made it clear that it considers health features part of the core functionality of its platform.




Apple seems to have hit a wall when it comes to developing more health features in its hardware. So software and AI seem like a natural transition.Foundry

But the company has found itself stymied on the hardware front in recent years. Reported attempts to develop both non-invasive glucose tracking and blood pressure monitoring have proved to be difficult nuts to crack, and will probably take a few more years before they are practical. Fundamentally, those are just very difficult problems to solve when you’re limited to products like a watch on the wrist or headphones in people’s ears.

Given those limitations, it makes sense that Apple would turn to software solutions to fill in the gaps in the health arena. And on the face of it, some of what AI technology offers provides a reasonable option: the ability to synthesize and summarize disparate data and turn that into helpful advice.

But, surprise, surprise, it’s not without its potential pitfalls.

Doctor, doctor, Mr. MD

According to Gurman’s report, Apple is working on training this AI doctor on information from in-house physicians. Which, while positive, doesn’t necessarily eliminate the risks of relying on AI technology—specifically, its tendency to make up information. Even in cases where large language models are trained on specific data, it still demonstrates those kinds of problems. Obviously, health is an extremely sensitive area: one would have to be extremely careful not to give incorrect advice that might inadvertently lead to harm.




Image Playground’s results aren’t very impressive. Can an Apple Intelligence health advisor do better?Foundry

One option would be to carefully control and monitor inputs and outputs as Apple does with some of its other AI-based products, such as Image Playground. But as Image Playground’s mediocre output demonstrates, that’s not without its own problems—a health agent that is carefully controlled could result in offering anodyne, common-sensical advice: sleep more, eat better, exercise more. One hardly needs any technology feature, much less an AI-based one, to provide those kinds of “insights.”

But the risk is that no matter how well Apple trains its model, it can’t avoid the possibility of bad advice. The non-deterministic nature of an AI language model means you can’t predict exactly what output it will give. And if it’s untrustworthy, just like in the case of a feature that tells you when to leave to pick your mom up at the airport, people simply won’t use it. Health is a place where people don’t want to mess around, and even a one-percent chance of bad advice will likely be too much for a lot of people.

Health and human services

All of that said, it is worth noting Apple’s excellent track record with health and fitness. Features like AirPods Pro’s Hearing Aid and the Apple Watch’s A-Fib detection have had meaningful impacts on people’s lives, and many of those features themselves rely on machine learning in one form or another.

The devil is always in the details, and reports like these only give us part of the picture. People—myself included—are, I think, rightly skeptical whenever a new feature purports to be based on AI, given the contentious nature of the technology. It’s entirely possible that what we envision this feature to be is not exactly what Apple will eventually roll out—if it even ends up bringing it all the way to fruition, hardly a guarantee given its recent track record.

Still, a move like this is classic Apple, not necessarily just about selling more products—even services, for which the company is notably hungry for revenue—but about enhancing its platform and making it more attractive to consumers. And I even believe, if I may be permitted a non-cynical moment, because the company is truly passionate about helping its customers live happy and healthy lives.