In late 2016, leaders from Amazon, Facebook, Google and IBM announced the Partnership on AI , indicating the field may finally be reaching a breakthrough point in technical maturity. Many things have led to this point, including advances in analytics and the availability of massive data sets via a variety of sources including the internet of things. A.I. is poised to seismically shift every sector of society, from the automotive industry, which is reckoning with the future of self-driving cars, to the retail industry, where recommendation engines are displacing customer service roles.
Healthcare is no exception to this trend. AI in healthcare currently dominates all other industrial applications of AI in terms of equity deals, raising $1.8B across 270 deals since 2012. Last year the space saw a 5-year-high for deals and funding, according to CB Insights. Given this focus, how might health care be changing as a result of A.I.? We’ve rounded up 5 recent developments worth noting:
1) The FDA’s creation of a new digital health unit
The Food and Drug Administration is expecting funding in the fall for a new digital health unit. The unit will be charged with understanding how machine learning, A.I., big data, cyber security, cloud computing and more, will affect U.S. consumers. It will be staffed by people with technical expertise as well as industry experience with product development lifecycles.
The FDA had a relatively hands-off attitude toward the previous generation of digital health technologies, many of which were in the mobile and wellness applications, choosing to “focus only on the apps that present a greater risk to the patients if they don’t work as intended.” The creation of this unit indicates that this new wave of A.I.-based software applications have the potential to impact health care treatment decisions and warrant greater attention.
This concern for patient welfare is at the forefront of regulators minds: “When you start adding analytical AI for any image analysis—think of detecting cancer or some other serious disease—at that point people need to know when that detection means something and is real,” said Bakul Patel, the FDA’s Associate Director for Digital Health.
2) Alibaba’s entry into healthcare AI: ET Medical Brain
Alibaba Cloud, one of the fastest growing units of the Chinese retail and technology giant Alibaba, launched its ET Medical Brain this past March. ET Medical Brain is a suite of AI solutions designed to ease the workload of medical personnel. China has a severe physician shortage, with only 1.49 physicians available for each 1,000 people (compared to 2.55 per 1,000 people in the U.S.) The “Brain” is expected to serve as a virtual assistant for physicians, supporting imaging review and hospital system management, as well as playing a role in drug discovery and development.
Alibaba has strong prospects for success because in China the administrative state has a firm grip on available data. As a home-grown Chinese company, Alibaba is far more likely to be granted access to this data than foreign-based players like Google or Amazon. With populations of that scale, Alibaba could perform the kinds of massive deep learning exercises that could create solutions to many of the problems the healthcare industry faces.
Why does this matter so much? Think billions. Not the over $200 billion that Alibaba is worth, and the massive resources it can deploy, but perhaps more importantly, 1.37 billion, the population and wealth of potential training data available in China, the world’s most populous nation.
3) Google’s DeepMind’s partnership with U.K.’s National Health Service runs into regulatory trouble
DeepMind is a U.K.-based artificial intelligence company acquired by Google in 2014 for $500 million. It’s now at the vanguard of Google’s A.I., machine learning and deep learning activities. In 2015, DeepMind offered Royal Free NHS Foundation Trust, a series of U.K. hospitals, free use of a patient monitoring app. It was the first time a big private technology player had been given large-scale access to national health data.
However, in early June of this year, British regulators deemed the project, and its acquisition of 1.6 million patient’s data, illegal. The project had accessed sensitive information such as HIV diagnoses and mental health conditions that were outside the original scope of gathering training data of patients with liver conditions, all apparently with patient identifying information in tact. DeepMind’s troubles getting training data represent a challenge at large for the healthcare AI industry— how to structure health care training data access agreements at scale that don’t violate patient privacy.
4) Accenture releases promising report: AI to save industry $150 billion in the next 10 years
Released in June, this report projects the market for health-related AI to grow from around $600 million in 2014 to more than $6.6 Billion by 2021. Accenture predicts that AI could create $150 billion in health care savings by 2026. The report has an interesting breakdown of the ten AI applications that have the most potential to create value:
- Robot-assisted surgery – $40 billion
- Virtual nursing assistants – $20 billion
- Administrative workflow assistance – $18 billion
- Fraud detection – $17 billion
- Dosage error reduction – $16 billion
- Connected machines – $14 billion
- Clinical trial participant identifier – $13 billion
- Preliminary diagnosis – $5 billion
- Automated image diagnosis – $3 billion
- Cybersecurity – $2 billion
You can dive into the full document “Artificial Intelligence: Healthcare’s New Nervous System” here.
5) Rising Awareness of Data Inequities Potentially Underlying A.I.
Everything in health care A.I. isn’t coming up daisies however. An interesting challenge that has received some attention lately is the potential for inequalities in care delivery as it is practiced in partnership with artificial intelligence. A recent Quartz piece looks at the potential ‘tyranny of the majority’ that lurks in big data. Titled “If you’re not a white male, artificial intelligence’s use in healthcare could be dangerous”, it looks at the real challenges technologists and medical professionals will face gleaning insights relevant to minority groups from available training data sets. In the U.S., training data is skewed, not only because of the majority white population, but because in the first place, there are inequities in access to health care services that feed into training data. Problems are bound to arise for patients if predictions are made based on a population that isn’t relevant to them.
Elaborating on this inequity-creep, Quartz says: “Data coming from randomized control trials are often riddled with bias. The highly selective nature of trials systemically disfavor women, the elderly, and those with additional medical conditions to the ones being studied; pregnant women are often excluded entirely. AIs are trained to make decisions using this skewed data, and their results will therefore favor the biases contained within.”
Looking at the above, you can see that the challenges remain steep, but it is an exciting time to be in health care A.I.. Perhaps this time next year people in China will be scheduling appointments with Dr. ET Medical Brain, the FDA may be regulating every health AI tool’s development process, and the application of training data sets will have been made more fair and accountable— only time will tell.