AI for Healthcare: The Promise and Challenges (Part I)
A Conversation with Dr. Xavier Amatriain, Co-Founder and CTO of Curai
Dr. Amatriain and I discussed COVID-19, and more broadly, the promise and the challenges of AI for healthcare. We talked about wearable devices, digital twins, privacy and government regulation, etc. We diverged into Artificial General Intelligence, machine learning with experts in the loop, federated learning, out-of-band predictions, and many other topics. Dr. Amatriain offered advice for the younger generation who are interested in entering the field of AI.
This is Part I of a three-piece series. See Part II and Part III.
Are the best minds of the world working on the right problems?
If AI is as powerful or promising as all the hype suggests, why isn’t it solving the biggest problems we face, such as healthcare, climate change, sustainable development, etc.?
Instead, companies chasing profits are using AI and machine learning to get their customers addicted to social networks, online shopping, and games. In fact, Jeff Hammerbacher, software engineer and entrepreneur, was quoted saying “The best minds of my generation are thinking about how to make people click ads. That Sucks.”
It’s exciting and comforting to know that Dr. Xavier Amatriain, one of the best minds of our times, is solving healthcare problems using AI. It reminds me of a story Prof. Michael Jordan mentioned in “Artificial Intelligence — The Revolution Hasn’t Happened Yet.” Jordan and his wife had to make a difficult decision about their unborn child: either ignoring a diagnosis that predicts a 1 in 20 chance of Down syndrome, or going through a risky treatment that has a 1 in 300 chance of killing the baby. Fortunately, using his expertise in statistics and computer science, Jordan found that the diagnosis was due to a false alarm caused by outdated healthcare technologies.
Jordan calculated that every day thousands of people may have got a similar diagnosis and many babies may have died needlessly. Addressing the problem requires principles that help build planetary-scale inference-and-decision-making systems. Jordan said: “It occurred to me that the development of such principles … were at least as important as those of building AI systems that can dazzle us with their game-playing or sensorimotor skills.”
“I think we have a lot of really smart people working on algorithms and applications of AI and machine learning in domains that don’t really matter or matter that much.”
The AI society must reevaluate the impact AI can have in the present, rather than pursuing science-fiction machinery. So here was my question to Dr. Amatriain.
Q: Prof. Michael Jordan talked about how the AI revolution hasn’t happened yet and that we should be focusing more on AI in healthcare and medicine rather than game-playing AI such as AlphaGo. Would you agree that we should work more on AI in healthcare?
Yes, I would totally agree with that. I think we have a lot of really smart people working on algorithms and applications of AI and machine learning in domains that don’t really matter or matter that much. Game-playing might be one of them, or algorithms for trading stocks, and those are very advanced.
We haven’t had nearly as many people thinking about healthcare and about how to improve and how to use AI and machine learning to move medicine and healthcare forward. I will say that in recent times that’s changing and I’m glad that it is changing, but I would agree that it’s important to put as many lines and as much effort in those fields that matter. I would say healthcare, education, energy, probably those are the 3 top needs for humanity and those are ones that deserve all of our care and attention and there’s probably others.
It’s fine that we have some people working on them like Netflix, I was there before, and it’s okay to build a product like Netflix, but I think we really owe it to ourselves to focus on the really important problems humanity has.
Is AI powerful enough to tackle healthcare?
But, is AI ready to solve problems in healthcare?
I am not optimistic. For example, since the beginning of the pandemic, hundreds of papers have been published on using AI to combat COVID, but none of them produced tools that would be good enough to use in a clinical setting.
One of the reasons is, unlike domains such as engineering and business, it is not easy to apply AI in healthcare, where there are a lot of regulations. I recently had the opportunity to speak with Tim Cargol, founder and CEO of Spectrohm, a deep tech radio-frequency internal imaging company. While Spectrohm tackles a new method of inspection in security and medicine, their prototype has only been approved to be tested in the security setting. Mr. Cargol noted that he and his team had indeed spent a lot of time researching and working on their technology’s performance in the medical field, but eventually hit a roadblock when it came to the regulations for product testing.
In response to my question below, Dr. Amatriain noted that AI is not there yet, but we can start employing AI as advanced algorithms and using data to improve decision-making.
Q: Do you think algorithms, especially machine learning, are now powerful enough to crack the secrets of nature’s law and benefit human welfare?
I will say that I usually try to demystify algorithms and I think that cracking the secrets of nature is probably too much for what AI and machine learning can do nowadays. But, algorithms can be extremely useful in many fields of science, including medicine. If we try to think about the places where we can automate and we can scale and we can basically use data to improve our decision-making, I think that’s the key aspect of how we should think about algorithms in the case of medicine in particular.
When a doctor has to make a decision, they need to remember years of medical research and things they have studied during many years of their college studies and publications that have been published since then and they also need to understand everything that the patient is telling them and they need to make a decision with all that data. That is very hard to do. Honestly, if we didn’t have computers and we were doing all the computations for any kind of job by hand, imagine that we didn’t even have calculators and we were doing all the computations that we need to do for anything — an architect or engineer — by hand, with pen and paper, that would be really hard and that’s pretty much what we’re asking doctors to do. Algorithms can help tremendously and you can go from very simple algorithms, that are basically just checklists or decision trees, to much more complicated and advanced algorithms that use data to build more advanced ways to help decision making for medicine.
Now that again is very far from understanding and cracking the secrets of nature and biology and human beings as such. But it is still extremely helpful. So I’m very very bullish on the power of algorithms to help medicine and help us as humans in the conduct of medicine and even drug discovery and things like that. However I would say I don’t think we’re even close to cracking the nature and biology of such.
If you can’t measure it, you can’t manage it.
If you can’t measure it, you can’t manage it. This is especially true in the medical domain, where measurement is often expensive and painful. Thus, if AI and machine learning are going to make a difference here, they must make measurement easier. However, despite the advent of wearable devices such as Apple Watches and Oura Rings, we can collect very few statistics using these devices.
Curious about whether these devices are close to making a breakthrough in the medical domain beyond displaying heart rate and sleep activity, I asked Dr. Amatriain on his predictions for the future of wearable devices for healthcare.
“I think wearables are a very interesting way to get data. I’m not exactly sure that nowadays, or in the short term, if they are going to be the solution. And the reason for that is they’re still way too expensive.”
Amatriain, while acknowledging the potential benefits of wearable devices, put the emphasis on accessibility, which he considered far more important at the moment. Amatriain pointed out that only a small portion of the population own wearable devices such as Apple watches, and that the “last thing we want is for healthcare to only provide for the rich people.” In order to tackle healthcare accessibility, Amatriain believes we should focus on developing measurement tools on phones, or at local pharmacies, through which patients can access affordable and extensive testing.
Q: There is a saying, “If you can’t measure it, you can’t manage it.” This is especially true in the medical domain. However, despite the advent of wearable devices such as Apple Watches and Oura Rings, there’re still very few statistics we can measure directly through these devices. What’s your prediction of the future of wearable devices for healthcare?
I think as you were saying there’s a huge need for more data and data-driven decision-making in healthcare and then the question is, as you well formulated, how do we get that data and where do we get it from.
I think wearables are a very interesting way to get data. I’m not exactly sure that nowadays, or in the short term, if they are going to be the solution. And the reason for that is they’re still way too expensive. Maybe we don’t realize it ourselves because we live here in Silicon Valley and we think everyone has an Apple Watch or can have an Apple Watch or wearable but there’s a tremendous amount of the world population that cannot have an Apple Watch. I am interested in saying how we can help those people, and the last thing we want is for healthcare to only provide for the rich people. We want to get it to everyone, so I would say because of that, I’m less interested or less bullish on wearable and such, but I’m really really interested in saying how we can get, for example, at-home testing or people to do testing and to check their blood pressure and their temperature in a digital way, in a much more accessible and affordable way.
There are a couple startups that are doing really interesting work on how you test yourself only with the phone — you don’t need an Apple Watch or any other accessory. That to me seems much more interesting. That’s one, so the phone as a testing device, and then the other one is how we can get affordable and cheap testing to either the home of the patient or to places that are nearby, for example pharmacies. Pharmacies have become a place where people can go and do some testing, but I think we could invest in doing much more accurate and much more extensive testing in pharmacies so that the people that don’t have the ability to buy the device could go there and use that as a hub.
So I think wearables will, in the future, be probably affordable enough so that they could impact the majority of the population, but for now, I think we should focus on how do we make devices that everyone has, like a phone, that are useful for testing, or how can we get people to a place nearby where they can get extensive testing so that we can use that data to make better decisions.
(This is Part I of a three-piece series. See Part II and Part III..)
