What did we know? When did we know it? And what are we doing about it now? — Reflections on Covid
One topic left mainly unaddressed in our discussions around Covid-19 is data. What data did we have before this outbreak? What data are we collecting today? And how are we using both sets of data to help us end the current epidemic and prevent disease outbreaks in the future?
Today’s technology and surveillance systems allow us to gather incredible amounts of data, analyze it quickly, and apply our learnings to develop a clear course of action. If we were paying attention to the right subsets of data, we would have seen Covid-19 coming. But we didn’t.
I recently had a conversation with Alex Pentland, Professor of Media Arts and Sciences and Director of the Media Lab Entrepreneurship Program at the Massachusetts Institute of Technology. He provided invaluable insight into the type of data that is most useful for detecting early infections and emerging outbreaks. He also highlighted the gaps in the use of this data in the United States and our ongoing failures to properly plan for future outbreaks.
William Haseltine (WH): Thanks for taking the call. I am asking questions about what we knew, when we knew it and how can we predict COVID-19 and similar events. My brother Eric suggested we speak.
Alex Pentland (AP): We knew something like Covid-19 was possible, and we knew what was going to happen. The big problem with Covid-19 is that it is infectious while asymptomatic. That combination just does not work with human brains. We are fortunate that this virus is not as deadly as some.
WH: Yes, that is correct. But isn’t it common knowledge that by the time you have cold symptoms, you are not as infectious as a few days before?
AP: This is a longer asymptotic period and it is a more serious disease.
WH: I am not so sure of that. After all, this is a coronavirus, a member of a family that causes 20 to 30% of all colds. What is true is that most people don’t associate what they know about colds with Covid-19.
AP: We had some early warning out of China. However they were insufficiently forthcoming, and we were insufficiently paranoid.
WH: Could we with our technology — I am speaking of the type of open source intelligence we collect and analyze — know what was happening in China regardless of what they said?
AP: If they allowed us access to data resources.
WH: No, I am not talking about that. Can our intelligence community collect and properly analyze open source information?
AP: We would need access to data resources they have.
WH: If not our intelligence agencies, could commercial companies, using the type of data they collect and analyze, have understood what was happening in China early on?
AP: Yes. I sent you a little blurb about our Israeli spin-off Endor.com that is using telco data and doing a bang up good job of detecting changes in behavior that predict outbreaks. That you can do. You can also do that using financial transaction data. You may be able to do it using Twitter-like data, but that signal would be more complicated. Such information could raise suspicions for sure.
You could also backtrack even if you do not have full visibility into the whole population. It is possible to make inferences from a sampling of data points. I am sure the government can do that using data from Alibaba, Tencent, China Mobile Baidu and others. We do not generally have access to that data.
There are also some more sort of gray area companies that do app tracking. App tracking is a service to help your application run better. They get lots of data about what users are doing. But that information is fairly well controlled in China, and increasingly so. You might be able to do it with satellite data, but then there would be a huge number of false alarms. You could tell something was happening, but you would have to really watch closely. To understand what the data means you probably have to be suspicious already.
WH: I am interested not just in China. I am interested in the rest of the world. Remember that the last lethal coronavirus, MERS, came out of Saudi Arabia via an Egyptian tomb bat. It infected some camel. Camels were sent to Saudi. From camels it jumped into people.
AP: You mean Africa, South Asia, all those places?
WH: Even the United States. There is evidence now that the famous 1918 flu started in the American Southwest, what is called the Spanish flu. It should probably be called the cowboy flu!
AP: Really? I thought it came from Southeast Asia.
Telco data is, as far as I am aware, the best way to detect early infections because people change their behavior when they get sick. They also change their purchases behaviors and a variety of other things.
WH: You think the government is going to start issuing contracts with the private sectors to build better disease surveillance services? Will our government start contracting companies like yours and others to build better intelligence systems for specifically this purpose?
AP: The US track record is not great, let us just say, and there are severe privacy and security concerns. My experience is that they think they know what they are doing, and they are not looking to go much further than the normal sources they go to. That could change. There was this review that Eric Schmidt did of artificial intelligence (AI) techniques in the Department of Defense (DoD). Somebody seems to have listened to that, at least to a degree. The Israeli defense agencies and the Singapore agencies are already using this stuff. They are known for being agile, right? They are agile in using these technologies. I do not know who else is using it.
I think that for this particular thing there are probably ways to do it that will be acceptable to most democratic societies. Not mass surveillance. It is right on the edge. We are going to try to do it with aggregated (privacy preserving) commercial data.
WH: You can monitor purchases for various items. You could monitor attendance at various events. Eric suggests you can monitor parking lots. There is a whole series of things you can do.
AP: There is a company called SafeGraph that has footfall at every store. MasterCard has purchases across the whole country by SKU.
WH: Just America?
AP: In every country. However every country is slightly different. It is not a completely coherent system.
WH: In China and India, they use mobile apps to pay.
AP: That is the best type of data. Most of the purchase data is not very fine grained with respect to location or thing being purchased. Digital data is much better. The Indian systems are quite good as far as I understand them. I have not actually worked with that data.
WH: I bet the Chinese ones are good too. When I am in China, I see that nobody uses money anymore.
AP: That is right. If you have that sort of system, you can do quite amazing things. The United States will get there. Most countries are getting there because it is cheaper, it is safer. The real question is to what extent is it safer. Long discussion there. You can understand a lot from that high quality data. We just published a paper with the World Bank showing that you can predict elections pretty closely.
WH: I would like to see that.
AP: It does not go into exactly how you do it, but it tells you about it. I can try and send it to you. (Author’s note: here is the paper https://academic.oup.com/wber/article/34/Supplement_1/S9/5648240)
WH: That would be great.
AP: It is pretty amazing because what you see…
WH: Who uses it?
AP: Currently some commerce uses it. The politicians are not yet using it. Most defense infrastructure does not use it yet because they have a certain myopia of using strategic assets.
WH: And certainly, our health systems should be using it more. That a lesson COVID teaches us.
AP: I think you are going to see a lot more of that for sure.
WH: You may have some insight into what I call the Cassandra problem. That is: Cassandra tells Priam, “There are Greeks in that horse.” Priam says, “You tell me there’s Trojans in the horse, but I do not believe you. I want that horse.” That is the difference between knowledge and action. How do you bridge that gap?
AP: There is a good answer for it, but you will not be happy with it. Trust. You have to have data systems and predictions that people trust. Take weather predictions for example. The weatherman gets up and says, category five hurricane is going to land in 1.5 days. Everybody gets out and boards up their windows.
WH: Unless somebody with a Sharpie says it is going to go into Mississippi.
AP: Okay. But maybe it is in Florida. They get out the plywood and the hammers.
WH: That is true. Your first thing, is it credible? Is it trustworthy and credible? Suppose it is already thought to be trustworthy and credible, but people decide not to do it. That actually happens. Credibility is one point, but it is not the only determining
AP: It takes experience. You need to know what the tradeoffs are of action and inaction.
WH: You mean there needs to be an understanding of consequence.
AP: Some people board up the windows and leave; others shelter in place even defying the get out warning. Their logic is something like, I am high enough, the house is sturdy enough. I do not need to listen to the warning. However those generally are actually the lower risk people and they are probably halfway right.
WH: Or there is the other category that says, I have survived ten of these before, so I am going to survive this one.
AP: That is, I have gotten the signals and now I know the consequences and I can deal with the consequences. You need to have experience.
WH: But when they are perched on their chimney top hoping for a helicopter, they may have a different feeling.
AP: Okay. Darwin comes in then.
WH: But is it Darwin when it is our national leader making the decision? That is not all of us, it is one of us. I understand what you are saying. You need to trust it. You need experience. This is helpful because you need to have a feeling of consequence.
AP: Also, you may not need a uniform national response. What you need to have is a bunch of plausible responses, and some of them fail. And then people say, let us not do that again. But forty nine other states made it through okay. Louisiana, since you mentioned it, has gotten away with not educating people, not having health infrastructure and much more for a long time. That may change. New York, because it is so central in the world, probably should have had unusual guards and surveillance, something different than the rest of the country.
WH: But we did not.
AP: That will probably be a conclusion of this.
WH: We are going to see this hit everybody now unfortunately. Let me give you a different view. I have just written a piece on viruses as engines for machine learning. And what they are trying to do is crack our defensive code and find chinks in our armor. And among the chinks of our armor, our political philosophy, that is a behavior pattern that the virus is cracking, large and small. Depend on individual, community, country. Behavior is one of the codes the virus is trying to crack.
If you think about evolution as machine learning, you recognize how powerful virus evolution is. You throw up random mutations and you see what works. Viruses are doing that all the time. We are learning that our new ecological niche seems to be great for these little critters. We have had Ebola, we have had Nipah. We never had these things before. The appearance of deadly new viruses in the global population is increasing. We never dealt with a generally lethal coronavirus epidemic until recently. Now we have had three in the past 17 years. Coronaviruses have always been with us. The difference may be that now we have a much better ecological niche to fill. That is us!
AP: When SARS happened, the total amount of travel in the world was like a third, a quarter of what it is now. These things happen, but they just spread very quickly now.
WH: That is what happened with AIDS. HIV has been in Africa forever.
AP: One of the things I expect here is that you are going to see a more decentralized world. I am just writing a piece now outlining my thoughts on that.
WH: I would love to read that. If that does happen, it may happen slowly. I think we are going to continue to be traveling. There will be a supply chain rethink. Companies will hold bigger inventories. Manufacturing sources will be diversified. More manufacturing will be local.
AP: But take companies. This has been pressure for a long time. Not everybody goes to the central building every day. They have to work at home. Well, boy is that going to go up on steroids. People are going to learn that you can stay home three, four days a week and it is okay. And if you do that one day, you could do it in a local WeWork type place rather than downtown. That transforms the transportation infrastructure dramatically. That transforms the spread of viruses dramatically. I think it is likely to transform politics just because now cities are less critical as infrastructure.
WH: You mean reverse the current urbanization trend. Where is Mr. Moses now that we need him? The second Mr. Moses, not the first one.
AP: I expect that.
WH: Well, the second Moses was trying to lead people to the promised land, what we call suburbs.
AP: It was not inherently bad. Digital communications in the world are orders of magnitude better than they were only five years ago.
WH: No doubt about it.
AP: I mean, we could not do business the way we do now five years ago.
WH: Yes, I run a global foundation from my home office. I run it in China, Singapore, Philippines, and India.
AP: That is what I am thinking of. You are going to see a lot of this stuff we have been talking about for thirty years, digital management for example, happening much more rapidly. We are not going to snap back to the way it was.
Some of the logistics chain stuff is going to change too, financial infrastructure, for instance. Instead of having big banks that are too big to fail, we will have networks. That is sort of what the MasterCard and Visa model is, right? Except they have some really big nodes. You can now have alliances of smaller banks that are highly interoperable. They can put capital together in enormous ways. If one of them fails, it is not an issue. It is only when you get systematic stuff across the whole globe that it becomes an issue. A distributed network is far more resilient to catastrophe. When New York goes down, the world does not go down. 9/11 did some of some of that. Covid is going to do a lot more of it.
There will be tradeoffs. Innovation is highly correlated and predicted by our rates of diversity of interaction.
WH: I agree. Such changes mean that innovation is going to drop.
AP: That is right.
WH: Think of your own life and what gives you ideas and where you get them. Maybe that is the past. Maybe that is not the future because I will tell you, during this episode, I have become extremely productive through interactions like the one we are having right now. Before Covid, I was not motivated. It has been a very creative period for me.
AP: I think it is true for a large fraction of people. I do not know how sustainable it is.
WH: Consider how much more time we have to think now that we are not traveling all the time.
AP: For instance.
WH: I think it is a function of being in an emergency situation. WWII was a very creative period. Things were shaken up. We would not have talked before.
AP: All of a sudden it becomes interesting when you talk to many different sorts of people, which results in this burst of creativity. But I hope that we do not have another continuing crisis.
WH: This is going to go on for a lot longer than we think. The virus itself has turned out to be trickier than we thought, and our response has been inadequate.
Let me give you an example. About one third of people who were virus positive released from Chinese hospitals have no evidence of antibodies, and no evidence of neutralizing antibodies. There is a new report that just says a whole bunch of people in Korea who thought they were clear of infection, endured a second attack. The virus popped back up. Whether it was a lingering virus that reappeared, or whether it was a new infection, they cannot tell yet. Also, immunity to this virus is not very strong. That is true for coronaviruses in general. As people return to work it is not clear who was infected, who is immune and who might be a source of a new epidemic! Covid-19 could keep popping up.
AP: I hear you. On the other hand, it is interesting to have this dichotomy between severe cases and weak cases which seems to be a function of how you mobilize your cytokines. Some people get cytokine storms, other people do not.
WH: Individual response may vary due to many factors. Initial dose may be one. Underlying medical condition certainly is another. The response, positive and negative, may depend upon how recently you experienced a coronavirus induced cold. Most people have had one within the last 24 months.
AP: One good thing I see at MIT. All the bio guys are dropping whatever they were doing to work on this. It is really impressive how many different people are doing different things.
WH: It is true. It is amazing. When I first started working on HIV, we were lucky to have at our disposal tools that had been created only in the previous ten years. Without them, we would have been stuck. Ten years before that, we could not have even isolated the virus. Our tools now, almost forty years later, are much more powerful still.
AP: Yes. Clinical trials are bottlenecked. We need better, faster ways to do them.
WH: I agree. But we still need carefully controlled trials. If we go too fast, we can kill people. We are seeing how confusing, lousy and downright scandalous poos drug trials can be.
AP: Let me give you an example. There are all these people that are trying off-label treatments, right?
AP: There is no central database of the generic off-label trials. If you had that, then you are getting hundreds of thousands of samples, at least, of different treatments. There are techniques now for teasing apart the relevant statistics.
WH: I agree. You could begin to deconstruct the hydroxychloroquine or the Remdesivir trials. All we are seeing are the results of microscopic studies.
AP: A study where you run 100,000 people or a thousand people on one particular protocol is 1930s thinking.
WH: That may work for an approved drug, but not for a new drug. After observing attempts to repurpose new older drugs for Covid-19 treatment, I am far more pessimistic than I was earlier about the possibility that we already had drugs that would be effective against SARS-CoV-2. Now I am pretty sure we do not. I am certain that we will have them eventually, but I do not think we do now.
AP: I was surprised to hear, what is it, eighty percent of the people that go on ventilators die.
WH: That depends. There are hospitals in the United States where only twelve percent die. It is not patients in ventilators that die at such a high rate, it is people who are intubated. Care of those intubated can be done really well or badly. It is usually done badly. I will give you another fact. I just wrote a paper that looks at comparative in hospital death rates. Do you leave through the morgue or the front door? There is a fifty percent difference in the very best hospitals in New York City of your probability of expiring or not, once you get into the hospital, a fifty percent difference!
AP: From an information technology perspective, you would look for the key variables.
WH: That is exactly what led to lower death rates in some hospitals. Knowing the data and then what to do about it. The best results come from those hospitals that collect and analyze the data in real time and take systemic actions.
AP: My strategy is I am going to just stay alive until people figure out what works.
WH: Good luck to you. Please stay safe. Thank you for your time.