In this week’s episode, we speak with Tiernan Ray, a freelance journalist who writes for ZDnet, the Street.com and other publications, about healthcare technology innovation including the power of artificial intelligence – its pace of development and where it may lead us in the 21st century and beyond.
Jeffrey Friedman: Hello, and welcome to the RP health cast by Rooney Partners. I am your host Jeffrey Friedman. Our guest this week is Tiernan Ray. Tiernan’s been covering emerging technology and business for almost twenty-five years. He was most recently the technology editor for Barrens where he wrote The Daily Market coverage for the Tech Trader blog as well as feature stories, cover stories, and a weekly print column. Prior to that, Tiernan was a reporter for both Bloomberg and for Smart Money today. We will be talking to Tiernan about some of the recent stories he has written on healthcare technology both in medical technology and in artificial intelligence.
Jeffrey: Tiernan, welcome and thank you for joining us today.
Tiernan Ray: Thanks for having me Jeffrey.
Jeffrey: Right. Now, you are a veteran technology reporter. And in Baron’s you wrote daily market coverage for the Tech Trader blog for nearly a decade, while at the same time, you were writing their feature stories, their cover stories, in a weekly column all on Tech for the print magazine. Now, will you continue to write about tech your increasingly — we noticed, more focused on healthcare technology. So, what is that transition been like?
Tiernan: I had to really go back and try and reinvent my skills Jeffrey, because I spent so much time focusing on semiconductors in networking and I was talking to public companies and I noticed that when I left Barons in twenty-eighteen, I really needed to go back into areas of research and challenge myself to figure out what are things that are going on and say artificial intelligence in life sciences and to dig into basic research going on and to school myself, really. So, it was kind of job retraining of a sort, for me.
Jeffrey: Now, it is interesting because as you started with semiconductors and they were doubling, you know, capacity every couple of days, right. Technologies always move fast. Developments now in healthcare technologies is kind of similar. It is related right now, especially COVID related once.
Jeffrey: So, they are moving even faster, you know, every moment. Can you lend a little insight into what it is like to keep up with everything as you are talking about retraining? How do you identify what you are going to cover in such a crowded and fast-moving health tech news environment?
Tiernan: Yes. I felt I was in a position Jeffrey at Barons for years where public companies came to me and wanted to tell me stuff. And when I decided I would have to broaden my knowledge and find out about areas of life sciences. For example, that I did not know about I started to dig in places where I had previously looked and one of the things that astounded me was this whole field of preprint publications. So, the process is, you know, you go on Twitter and you see somebody says, hey, we just released our paper on archive, which is the Cornell-operated preprint server. Anyone can go and download the PDF and there is just tons as hundreds of papers on a daily basis that show up from researchers. These are things that have not been peer-reviewed. They are the kind of thing that will eventually show up in nature magazine or science magazine.
But some of them are amazing kinds of discoveries in all fields in artificial intelligence, in physics, in life sciences. And so I discovered this, I had no idea while I was at Barons with this whole phenomenon had popped up in the preceding decade or so where you just go to this server and there are tons of stuff and it is just like kid in a candy store, if you like to research. And so, it is also to use another metaphor drinking from the fire hose. So, I had to have — kind of a learning curve, where I just started reading stuff I found on artificial intelligence, in genetics, on the archive preprint server and got into this rhythm of constantly checking. And it is the kind of thing that I continue to do for the last year and a half since Baron’s — just every single day. I will go and check out what new research has been posted and it is not just me Jeffrey, you see stuff in the New York Times, from the post in the Boston Globe and Financial Times that is pulled from a preprint server where basically every journalist is trying to get ahead of the curve in what is the latest COVID research.
And so they are going on archive or versions of it called bio-archive or meta archive and they are pulling down the latest report from Harvard that is not yet in publication, that just been thrown up there and it is kind of amazing, front kind of moving wave, you know, leading edge of research from these labs.
Jeffrey: That is really interesting. So, is it like one of those old school bulletin board servers that people are just posting?
Tiernan: It does not have the social interaction yet, Jeffrey. The social interaction component appears to happen to an extent on Twitter where a scientist will post something, there will be a thread of discussion about it. These are really bare bones. I compare it more to FTP, file transfer protocol, in the early days where you just cannot believe that there is a resource. You have broken into some folder, and there is just stuff there and it is like, oh my God, this is amazing.
Jeffrey: Wow. So, as an example, I think when one of your stores, last month in New York, right. Now, New York last month — we were like in Dire Straits, right. Our healthcare leaders were looking to source as many ventilators from around the world as we could.
Jeffrey: And you wrote a story about a team of a hundred astrophysicists who were working together from quarantine to develop a super simple cheap ventilator that they hope to make in order to help patients with COVID-19. You know, is this the type of things that you were finding?
Tiernan: Yes. Amazing. Just amazing. I stumbled upon this report and I just saw, they have a name, Mechanical Ventilator Milano. And so that is the kind of thing where you see that in an article on a preprint server and you say that is fascinating, wonder what that is because they already have some kind of branding. I do not know if it means anything. And I dug into it and as I read it, I saw, “Oh, my God, this is kind of incredible”. These are people who are — you know, start to follow the author citation. So, the lead author is a gentleman named Christian Galbiati who is a physicist at Princeton University. And full disclosure, I went to Princeton. So, I am kind of intrigued as soon as I see a Princeton reference. So, I started tracking it down and I said, “Okay. He has been working on something called Dark Side, which is this detector to try and pick up traces of dark matter in the universe that is built in a tunnel several miles underground under the Gran Sasso mountain range”, which is a kind of a belt across the middle of Italy. He has been working on this for years with a huge team of physicists.
And so, suddenly he is popping up here with over a hundred authors, on a paper about a ventilator and second reading it started digging into it, and I reached out to him and they were kind enough to respond to me. And this is just sort of one of those gems that shows up where you are just sort of kind of wandering through the preprint server and you stumble upon, “Oh, my God, there is like just an incredible piece of fully-developed research an incredible team of people and it’s right there”. And you feel that, “Wow! I am the first person”. I know Jeff, seamless.
Jeffrey: And I mean, it is such rabbit hole that you may have been, absolutely. But — so, like a story like that, the news is moving so fast and in New York, the infection rate curve has peaked. It is declining. Our need for additional ventilators has gone down. Now, looking back. How did it all turn out? Did they make the ventilators? Is the story still relevant?
Tiernan: Bill very relevant. I spoke yesterday with Cristiano Galbiati and he said this, “We, all in the physicist at Princeton”. And it was very moving Jeffrey because he told me in sweeping detail about the technical aspects, but he also gave me a perspective on how it came together. You know, they went into lockdown in Italy, in early March. And he said to me we were really kind of feeling at a low point, because in Italy we felt it was just Italy. For a while, you may recall in an early — late February, most of the say, “Oh, it will leave the hot spot”. And it was before things have been implemented even how. And so, they felt alone and he was talking with a friend who runs a gas company that he partners with to develop these dark metal detectors and they said, you know, we since we have no labs to go to at the moment, because we are in lockdown, couldn’t we do something about what is going on?
And so, he and this friend brainstormed about, you know, building these large detectors for antimatter is — you know, kind of a really complex version of moving air in and out of a chamber. Kind of like what you would do with a ventilator. And so, they just got underway and he and Cristiano sent messages to collaborators at Fermilab in the US and Illinois, and other facilities around the world, and got some of his top people to kind of quickly throw together the diagrams, the technical specifications, and within days they had gotten what was basically shown in this paper that I found on archive just by brainstorming. And he said, you know, this was a way to move beyond feeling frozen. He said I was just four days when the lockdown happened in Italy. I was just frozen. I did not know what I was doing. I could not function. And so, gradually by going back to what he knew how to do and bringing together a team of people and seeing that there were connections around the world, that they were not completely isolated. That there was partnership and friendship. This was something that kind of lifted everyone out of this.
And so, it was an incredible, been an incredible effort. They are now under, starting volume production, and they hope to get to thousands of units per day. They have a sort of initial run of fifty units if it tests them. They got FDA emergency approval for this ventilator. And so they are kind of on their way now and he does the basic science. He was not in charge of manufacturing runs, but he did indicate that this is still a current need. And one of the things he is concerned about is, there could be as many people say a second wave and the fall. And so, as far as he is concerned, building ventilators is still a critical thing to do.
Jeffrey: Well, what started out as a cathartic exercise is lifesaving.
Tiernan: And could be — yes. And could be something you really still want again depending on the shapes of these curves in many nations of the world.
Jeffrey: Yes. That is great. One of the topics — another topic you have been reporting on as a contributor for ZDNet has been the significance of healthcare modeling. And you recently wrote about the concept of super spreaders, and that’s really been in the news lately, especially with that nightclub guy. Maybe got to talk to a paddock, but in terms of healthcare modeling, do these one-off super spreaders, you know, can you talk about how that may affect the modeling and what is modeling being used for to track the virus?
Tiernan: Yes. Everyone is trying to figure out how can we do better than what has been the — known as the susceptible infectious recovered model, which has been around for over a century. That is the main model that you hear about from Oxford and from up in Seattle, from Bill Gates’ group. All of these models that are used by public policy experts and by governors to decide what do we think is the shape of the curve and how can we flatten the curve? The problem with them is a very generic and they are very abstract. And so they do not capture a lot of real-world data. There are people doing what is called Curve-fitting, where they take some parameters to try and guess what they think will happen with the spread of the infection, with the doubling of cases.
And so, everyone has been trying to get — do better than this and one of questions now becomes — that is tied to these models is, how do you do testing to fill in real-world data as opposed to simply mathematical exercises? And so, one of the things again that showed up in just amazingly preprint, is a bunch of authors from Google. One of whom is a off your right, who is a data scientist at Google. He had a couple of colleagues with help from Tel Aviv University. They put out this paper and they said, “You know, everyone’s talking about testing now. You do not want to go out and mass test everyone because it is just not practical”. What you should do is you should follow the super spreaders, and the super spreaders are maybe index patients, first patients, who appear to have been able to, for whatever reason, spread the disease to more people than what is the average transmission rate of a disease. And these are again statistical terms because we do not know the mechanism.
But it does seem to be that in any kind of epidemic or pandemic, you can actually find individuals who, if you trace all their contacts, have led to more infections than what you mathematically model. And so, as being the average — and so, they are saying, “Go and do contact tracing”. And this could be something you can imagine as the Google or Apple kind of Bluetooth tracing system on a smartphone. It could be using GPS signals. There are no people talking about using GPS signals. And it could be plain old-fashioned contact tracing where there is a kind of like the intake interviews on a hospital, someone in hospital, who after someone test positive, you say, “Okay. Who have you been in contact with?”.
So, there’s all kinds of forms of this, but it shows you that we are still trying to fill in the blanks with these very abstract kind of rigid models that have been around for a hundred years that are statistical that do not reflect the details of spread of disease. And two, we are trying to decide how we are going to use testing when we are in a testing constrained environment? We do not have the number of tests. We should have, we have stumbled in the US in getting testing rolling. It is because we have to make these choices, decisions about how to use resources and what is going to be most effective.
Jeffrey: Yes. I guess it is fascinating and so important and that brings us to another topic of artificial intelligence, which I know you have been covering for more than a decade. So, what are you seeing now that you find particularly interesting as it relates to AI in healthcare?
Tiernan: I think that combining AI right now with the work of human experts is pretty interesting. I took a look at — there is a lot of attempts, Jeffrey, that you may have seen to try to use AI to look at chest x-rays or CT scans as a basic way to look for COVID and there has been a lot of struggle in this area. I spoke to scientists who worked on this and the problem is that you do not have enough data. It takes time to train AI. And so, in this rush to try and implement AI, I think we are seeing people saying we need — we are realizing the gap — one of the gaps — the big gaps is implementing AI and machine learning and deep learning with human processes, with human systems, where experts are working. The model, when things were, you know, kind of casual, before the world change was to say, you know, we are really impressed with what the machine does.
And we are just writing everyday about what the machine does. And I think, when systems suddenly are stressed and you see human beings rushing into the breach with their skills, be they radiologists, who can look at an x-ray or first responders. You say, “Oh, okay. This is what it means to really be able to handle situations”. And maybe you need to take these machine learning models if you have been building and have them actually integrate with human expertise because human expertise makes quantum leaps. That is what happens in times of stress and you cannot spend the next decade, sort of leisurely training a system. You might want to add in some of this expertise.
Jeffrey: Yes. I mean, I think last year or two years ago, IBM’s Big Blue did that experiment on breast cancer?
Jeffrey: Right. And they — you want to talk about that?
Tiernan: Yes. So, the IBM Big Blue experiment with breast cancer, and similar experiments with radiology, Jeffrey, always come down to a kind of obsessed highly abstracted model of diagnosis where you know a probability after a fact, just by virtue of how things you train that this diagnosis would have been something that would have led to — you know, the correct procedure for the patient, it is all after the fact. And so, the consistent flaw and these kinds of you know, AI does better than a human expert kind of thing is, they are always looking retrospectively at what has been gleaned and they are not true to the actual scenario, which is you are a female patient, you come to your doctor and you get a certain diagnosis and there is a level of uncertainty. And this human being in front of the doctor has to actually be given advice and it has to be a procedure pursued. And this is the moment of decision and you can be helped by statistical tools, which is what machine learning is generally. But there is still a choice about a person and what is the proper standard of care for that person?
And so, I think this is actually another instance where all of these models, despite being kind of remarkable science, have to be — they have to be integrated into the work of human experts and people in fields. And we have seen this across the healthcare landscape. We saw it with DeepMind working in British hospital system in diagnosis as well. That there was only — there was a limit about what could be achieved with these AI models because at some point, you sort of have a connect with actual clinical practice. And so there is this big challenge of how do you get these systems to work with the reality of a clinician and what they have to do with uncertainty, rather than simply statistically modeling what you know long after the fact by looking at case series.
Jeffrey: All right. So clearly, you feel we are not there. There needs to be human interaction. All right. So, for the tough question then, if you were to pull out your crystal ball, where do you see AI taking us in five to ten years from now? You know — also do you think it is a good place for mankind or will it be a scary one?
Tiernan: It is going to focus; I think it is going to focus on language principally because it is the area we have seen the most progress and that progress is still unfolding Jeffrey. The Transformer, which is a seminal breakthrough in two thousand seventeen from Google that has informed all of natural language processing and language translation by machine learning has continued to pay dividends and some of the things that are happening with just even chat Bots and with natural language translation between French and Spanish and Hindi and Yiddish and Chinese and being whose giant capabilities are incredible and there is a kind of, you can see there is an industry within AI of just the language stuff.
And so, I think that you are going to see, for a while, a concentration of achievement and breakthroughs in the area of language and that is no small thing, because if you have increasingly sophisticated models of using language, they can flow through to lots of other tasks that have to do with describing things seen in pictures. Tests that have to do with how you respond to speech online, writing social. And there is a qualifier, which is that more and more these large companies that dominate AI like Facebook and Google, will combine language capabilities with other kinds of signals, be the images with a sound, to do what is called multimodal learning and that will enhance, I think, the language part. But I feel like the language part is going to be pretty profound revolution as it flows into the world because text writing can travel so quickly and we have yet to see the full impact of that because we have yet to see all of the technological breakthroughs that will happen in that area of AI.
Jeffrey: Well, as the speed of technology changes, I am sure we will see it pretty quickly. Tiernan, thank you so much for your time today, especially with everything out there.
We hope you enjoyed this week’s podcast. If you have any questions, comments, or future story suggestions, please reach out to us on social media or email us at firstname.lastname@example.org or visit us on our website at rphealthcast.com. Additionally, if you like what you hear, please follow us, review us, and share us with your friends and colleagues. Thank you, and we hope you enjoyed the RP health cast.