Big Tech
AI takes center stage for Google Health
As Google aims to talk as much as possible about its progress on artificial intelligence in the face of stiff competition for the AI spotlight, it's become a focus of the company's health care efforts. So Google's annual health-focused Check Up event earlier this week was an hour-long showcase of its work on health AI. It wasn't until the very end of the event that presenters brought up health care efforts in search and YouTube, two of the broader company's most important products.
Of particular interest was Google's approach to developing an experimental generative AI product that can have appropriate medical conversations with patients. The company recently put out a pre-print in which a research-only bot, called AMIE, talked to actors playing patients following a standard interview technique in an effort to come to a diagnosis. At the Check Up, Google announced its would begin testing AMIE in the real-world with an unnamed health organization. But now it would only conduct early parts of clinical interviews under the supervision of doctors without producing diagnosis.
Google Health AI leader Greg Corrado told me that he didn't believe going after diagnosis as a final goal was the appropriate way to investigate the potential of generative AI. Instead, the company hopes to explore how the technology can employ diagnostic reasoning in service of assisting clinicians.
The cautious posture here was characteristic of all of Google's AI announcements where it emphasized that it hopes to develop AI for health care in ways that are safe, equitable, and responsible. It's also reflective of the constant insistence among AI developers that their products are intended to be used with "a human in the loop."
Also of note: The company's partnership with Apollo Radiology International to use AI to combat tuberculosis in India and a new open access data set of 10,000 images of dermatological conditions, developed in collaboration with Stanford Medicine.
Medical Devices
Why wearables aren't common in care yet
For years, tech and medical device companies have been promising that wearable devices that measure activity, sleep, heart rate and more will contribute to a world with more personalized medicine, but this possibility for the most part remains unrealized. In a new paper for the New England Journal of Medicine, Stephen Friend (formerly of Apple and Merck), Geoffrey Ginsburg (from NIH's All of Us initiative), and Rosalind Picard (MIT) explore some of the many obstacles that must be overcome for the tech to take hold as more than just experiments. In general, these can be grouped into challenges with the health care systems and with patients.
The problems with the health care system may be more evident. There simply aren't well-developed universal pathways for getting data from devices in front of doctors, nor are there established workflows whereby doctors look at the data. Reimbursement schemes for devices and the medical labor required to support them are nascent.
Even if those problems are solved, the devices will not be useful if patients do not embrace them. The authors say that there ought to be efforts to educate patients from all backgrounds and and skill levels on how to use technologies so that they are more readily adopted. And the authors argue that patients ought to be empowered with control and access over the data being collected from devices and an understanding for how they might use it themselves, for example to change lifestyle behaviors.
No comments