Robert Fink on Lessons Learned From Using Data In Crisis Response To Reduce Risk and Support Mission

Informing your decisions with intelligence analysis and insight brought to you by the team@ooda loop.com.

Hello Robert Fink. Welcome to the Oodacast.

Hi Bob, good to see you.

Yeah, good seeing you. And Robert, before we get into this, I wanted to give our viewers and listeners a little context and say that I've had the pleasure of interviewing you before. I interviewed you for a series of three articles that we did at CTO Vision. And that's back when you were the architect of Palantir's Foundry product, correct?

Yeah.

I interviewed you on topics like data architecture and best practices there use of open source, including how you guys use many common open source development tools, development environments and also open approaches to data in your development and open approaches to data in your delivery to customers. And we also did a post that talked about your overall open platform, correct? Yeah, yeah.

This was what, two years ago, maybe 2018.

I think it was published in say September 2018 and it seems like yesterday. The world has really changed since then in big ways, including your position at Palantir. You at the time I think your title was Architect of Foundry or something like that. But now you are head of R and D and I noticed from your bio you are continuing to work R and D. You're also working in your commercial business in the APAC region and you're still working a lot of customer facing things and you're still leading and shaping a lot of the data integration activities. But it seems like you're spending more time with your strategic partners and even some strategic customers, is that right?

I think I got really interested in, and in a way this is back to the roots because I spent way too long in academia before my time at Palantir. So I was interested in sort of, you know, the strategic technology development outside of Palantir and at Palantir. So what should Palantir look like in three to five years? Where should we invest? What sort of areas should we focus on? And so basically what I try to do is sort of play around with a lot of new developments, try them out, see which of them we should take on as technology at Palantir. And as part of this I work with academic institutions. So I'm trying to get off the ground a sort of collaboration program between academic institutions and Palantir. Things like sort of research interns that would come and do their research with us. So that's sort of half of my work and then the other half of my work is actually talking to our customers, mostly at the sort of CTO level. CIO level. So sort of senior management in the, in the data realm, let's call it, or the IT realm and explain to them what we do. Explain to them what Palantir is. Explain to them what Foundry is. Obviously having started the development of Foundry now many years ago, I know it quite well. And so they have a lot of sort of more. I guess, you know, they're not interested in the minutiae of. Can I right click here and turn the button blue? But they're more interested in the sort of grander questions around our development philosophy and how we think about trade offs between sort of security and velocity and how we organize our development teams and all that sort of stuff. And that's something I talk to them too.

And Robert, I wanted to mention before we get into this also your academic background because I think not only do you. Are you staying busy doing that, but you're just a very interesting person with a doctor's degree in physics.

No, no, no, not that interesting. Unfortunately. I started my academic career in physics. I did five years of physics, so roughly this was called diploma in Germany, which would now be bachelor's and masters. And then I decided I wasn't actually good enough at physics, so I sort of switched to computer science and did that in parallel. And then my PhD is actually in computer science. Oh, great.

Okay. Database theory.

Correct.

Yeah. Well, I find all that extremely interesting and I would love to have discussions about that. In fact.

Well, we should do a separate interview just on my thesis and then I'll explain all of that to you. It'll be very boring.

No, no, no. I think me and people like me would find that extremely exciting. It may be. It might be better if we wait until all these social distancing rules are over. So we could do that over an adult beverage. I think it would be.

That sounds great. Just read.

But you know, part of the reason I wanted to do this interview is I watched. Another thing you've been doing is a lot of writing. Yeah, I saw one of your recent Medium posts that really resonated with me. Not just because of the current pandemic, but because of every crisis I've ever been involved in in my entire career. I think what you wrote there was so applicable and the title was something like I have it here. The In Crisis Response Answer the Simple Questions First. And as I read that Medium post, you made a lot of sense in there and talked about a lot of things that I want to hit on, in fact.

Well, most of them are very obvious. You'll see but, you know, there's still something you might forget every now and then and then try and chase something really complicated. But I think, I mean, we'll get into the details, but I think when you're in these sort of unknown times, and this could be natural disasters, or it could be sort of personal issues like you're having a car accident, when you're running software systems, whatever you're doing, when you have a crisis mode, you sort of wipe all prior knowledge about the world and the state that you're in and you start from scratch. And I think it really pays to just focus on simple questions first and then try to do the more elaborate thing.

Yeah, and I think when you wrote that, it resonated with me so much. Not just in this world of crisis response and computer science, but through human history. You have these great leaders who are able to succinctly put what they're all about in one sentence. Whether it's Hannibal, I will cross the Alps, I'll find a way or make a way. That was simply put. And now it's a very different world now. It's very chaotic. It is a crisis and it's a modern world with lots of technology. That article really resonated with me. We'll have a link to that so people can read the details through this conversation. I would love to ask you more thoughts based on that on, like, what Palantir is doing in support of healthcare, including research and current outcomes, but also what Palantir does in crisis response. Maybe some of your personal experience there and then also I want to talk to you about privacy and security.

Okay.

In the interest of time, I would love to just jump in and ask.

Let's do it.

Maybe starting with your personal experience at Palantir in a crisis response.

So thinking back, I joined Palantir in 2013 and due to some mishaps with visas, I actually didn't start in the us I came from Germany, didn't start in the us, But I started in Australia because I was able to get a work visa for Australia more easily than for the us. And so basically just the week I arrived in Australia, or maybe the week prior, I don't remember exactly, Typhoon Haiyan hit the Philippines, and so I basically landed in Australia in a moment of crisis. And we as a company had decided at the time that we would go there and help. And so basically, you know, half of the Canberra office work was. Was in the process of being equipped with laptops and servers and drones and cameras and all sorts of stuff, tents and sleeping bags. And some of those folks were, were about to actually board a plane and go to Philippines, to the Philippines to, to basically help. And so this was sort of my first starting at Palantir and already in a crisis mode. I joined a little too late to join that mission by actually going, but instead I stayed in the office and helped them with some data integration work and some programming work and basically sort of supported what they were doing for the Typhoon Haiyan response that we collaborated with there. So that's sort of my very first point. And then I think you asked about personal experiences with crisis. When I think about my career at Palantir, the mean I worked in the foundry development organization and obviously this is a piece of software and things go wrong. You have bugs, you have outages, customers call you and say that certain things don't work anymore. And so basically every day is like a mini crisis when you run these, these complicated infrastructure systems. And so basically my closest experience here is actually maintaining the software going through outages. Some can be just a little bit annoying, some can be fatal. You know, if you're, if you're. Fortunately we're not running Bitcoin. If you're running Bitcoin and you see that a hacker is stealing millions of dollars of you, that's a very fundamental crisis to your software system. So, you know, like any other software, we have bugs and outages and just working with our engineers through these points of crisis, these little mini crisis, if you want, is basically what, what sort of educated me about the mindset of how you actually operate when you are in a crisis mode.

I see. Yeah, I never would have thought of that ever. And I want to talk more about crisis and privacy and security especially. But first, since we're in this pandemic, everybody's thinking about healthcare, health issues. Personally, I believe that this decade is going to be the decade of biotech. There's going to be so much money dumped into not just this pandemic, but all aspects of biotech and health care. And when it comes to health care, of course there's patient outcomes and there's long term research. And I think you guys are playing across that spectrum.

Right?

Even like say cancer research.

Yeah, we have a very long standing collaboration with the NIH in the United States on specifically medical research. So I mean, if you look at what that space is about, obviously there are subject matter experts that know all about cancer, but then there is also sort of computer scientists that know how to work with data. And one of the constraints that exist there is basically bridging these two walls. So how can a computer scientist actually work with the data and a medical expert work with the same data so they can drive medical research forward? And if you think about the data sources, there are these things like high throughput screening, there's, there's genomic sequences, mass spectrometry, all that stuff generates a huge amount of data. But then there are these challenges that the tools to actually work with that data are built by computer scientists, mostly for computer scientists. They're not built for medical experts or researchers. And the data is extremely high scale and is very heterogeneous. So it might come from, you know, a dozen different sequencing systems that all have slightly different vocabulary and nuances in how they store data. So it actually takes a lot of upfront work to get through that data before you can ask your first questions. And then there are, and this is often overlooked, there are regulatory requirements that often prevent collaboration on these data assets. You know, this is medical data, so you have to be extremely sensitive on how it gets collected, where it gets stored, how it gets processed, when it gets deleted. So all of these sort of regulatory or data governance questions make it really hard for these researchers to collaborate. And so one of the places where we started with medical work was actually to provide data systems to these researchers so they can focus on research. If you're interested, there's a nature paper that we just wrote on quite an interesting subject. This was on lung cancer in black populations. And it turns out this, this was quite a challenge for medical researchers because black populations are underrepresented in these research studies. And so they just didn't have enough data. Each, each of these research institutions individually didn't have enough data to actually come up with meaningful results. And so what they wanted to do was collaborate and basically pool their data sets so they can collaborate and have a sufficiently large data set to actually conduct their studies. And basically we gave them our software platform so they can do this collaborative data analysis there.

Yeah, that's cool. The collaborative data analysis. And that's one thing that, to me, is different as I've looked at the history of this kind of research. In fact, maybe oversimplify, let's say about 15 years ago, as the IT revolution was really kicking off, people started asking great questions about, hey, what if I look at every kind of cancer there's ever been and everyone who's ever been successfully treated and every successful treatment, then I can correlate that and say, when I see this kind of cancer, here's the right kind of treatment. Well, they found out quickly that was not that simple because of all the genetic differences and then there's the epigenetic differences and the environment and it gets so complicated. So it's not just a simple data analysis problem. And now you're adding on top of that, not just the ability to do analysis, but the collaboration between analysts and researchers.

Yep. The governance piece, I think is a major component. Like you, you can't ask these researchers to share their data assets with other researchers for, as I mentioned, regulatory reasons, but also for just maybe you could call them egoistic reasons. Right. So for these researchers, their data collections, they make or break research careers. It's extremely expensive to build up these research data pools. You have to run year long studies. And so they wouldn't just go and like ship a DVD of their data to someone else because that's their career. Right. So if you want them to collaborate, you give, you have to give them confidence that the data is secure, that they have auditing capabilities so they can actually see who's using their data. There's a really interesting, that's maybe another interview, but this really interesting thing around sort of attribution. So if these people start collaborating, collaborate, collaborating on data, maybe the data platform can actually sort of automate the attribution problem so that you want to write a paper on this analysis. Well, you also have to give credit to that person, that person and that person because it was their data initially. And so all of like this entire space I think is really interesting for us.

Yeah, it is fascinating. And then something more personal to everybody that's personal and long term and can impact all of humanity. But all of us also have more short term interest in healthcare like patient outcomes. And are you guys involved in that side of the healthcare business as well?

Yeah, I mean, maybe the most proximate example here would actually be from our work during the COVID crisis. So we work with a couple of companies in the medical supply space. So if think of ppe, personal protective equipment, it's a bunch of companies, we work with some of them and basically, I mean, if you look at the COVID outbreak, this basically dramatically changed their business from one day to the next. So demand for, let's say masks, face masks would spike by probably orders of magnitude at the same time your supply chain breaks down. And so they basically start from scratch. Right. So what they did is they paused all orders and they said, well, let's try to prioritize which countries have what type of demand, what can we produce in what facility and so on. And so what we basically help them with our software is to sort of integrate the data that they have and that they need to make these decisions and just come to these decisions more quickly and more accurately. So I think this is like a very proximate medical outcome because you know, it gets protective gear to the, to the practitioners, to our doctors, but also to patients and so on.

Yeah. And you just reminded me of something else I saw. A recent, wasn't a press release, it was a press conference in the Pentagon where the four star general in charge of US Northern Command was talking about their actions in support of the crisis response. And here in the US of course the military is very constrained on what they can do in support of civil authorities. And when we do it, it's, it's via an organization like U.S. northern Command. And in this press conference he called out like four companies that he wanted to thank for helping in their planning effort. One was Apple and forget the other two. One was like building some mobile security software and one was Coke or a.

Chocolate company in there somewhere.

Right.

They have to eat and drink.

But I think in the context he was thanking you guys was in helping his operational planners be able to forecast and say we're going to need X over here or we need to respond to something that seems to be happening over here.

Exactly, yeah. So I mean this is, this is a really good use case. The, you know, they use our software to obviously they use the data they already have. I mean in theory they have databases that tell them what supplies they have in what place and what soldiers they have with what capabilities and skills and equipment and so on. But there could be dozens or hundreds of these databases. And so basically they use our software to integrate that data and then just come come up with better global decisions.

Yeah. So I think first of all that's pretty cool. Congratulations, you have a four star. Talking about you guys in public, you know, they don't do that. They really don't.

It's been a long road. So.

And also to get back to this thesis of your medium article that I liked so much, you were saying ask simple questions first. I bet you that Four Star is asking simple questions of his database integration with you guys.

Yeah.

But I did want to talk a little bit more about the medium post.

If that's okay with you. Yeah, sure, let's do it.

What did you mean by. Maybe I shouldn't say what did you mean by ask simple questions first but give me some examples of in a crisis response, what should policymakers and designers and data architects build their solution to do when it comes to asking simple questions first.

Okay, so I have a number of examples. We can go through them. Interrupt me when it's too lengthy. So let's start with sort of what happens in a crisis. So basically, the status quo is annihilated. All of the things that you believe were true about the world and sort of your context that you're operating in, they're no longer true. And there's a very, very abrupt reversal of priorities. And this is, I would claim, true in all types of crises. So I mentioned things like you're in an accident. What happens when you're in an accident? Well, the conversation you just had with the person on the driver's seat is no longer relevant, right? All that context is gone. Now you care about simple things like who is alive, who's breathing, who is bleeding, what type of injuries do they have, Is everybody accounted for? So, like, very simple questions. All the priorities you had before getting to your meeting in time, wherever you were driving, they don't matter. If you look at other things, like I mentioned, software outages in. In software companies, we actually use the same analogy. So when we treat these software outages, we call the first step to stop the bleeding. So the first thing we want to do is we want to gather some data. What are the symptoms of the outage, which systems are affected, what customers are affected, how severe is it? Maybe we can even figure out what changed in the software system. So maybe there was an upgrade. So you start with just some simple data collection and some very simple questions. And so then the first thing you do is what we call, again, stop the bleeding. So maybe you send an outage report to your users. Maybe you take down the website and replace it by, like, a little banner that says, sorry, we're having a problem. But maybe again, if you're Bitcoin, you actually pull the plug, right? That's what it takes to stop the bleeding. Otherwise, you lose millions of dollars a second if you're being hacked. So you do the very simple things. And in these software crisis scenarios, what newcomers to that field often do is they try to jump directly to the complex questions. So they wonder, what. What's the root cause of this? Right? Could I have prevented this in some way? In theory. And those things are interesting, but they don't matter in a second. Like, if you're Bitcoin and you're losing money, you don't care what the root cause is. You just want to stop losing money once you stop the bleeding. And then once you're in a steady state, then you can actually move on, and you can wonder if your testing coverage was good, if you can root cause the problem and trace it back to some particular action. Maybe you need to update your developer documentation and work on your engineering culture. Maybe you have to have code reviews that you didn't have before. So all of these secondary questions are something that really don't matter in the first instance. I think I spoke about Typhoon Haiyan before. When you look at our work in this kind of disaster response, again, you start with really simple questions. They seem almost stupidly simple. But you're wondering which region, regions are affected. It's incredibly hard to get that data because all of the communication systems break down typically right in a hurricane or in a mass flood. So how do you actually gather the data to see what regions are affected? How many people are affected? Do they have drinking water? How much food do they have? How long can they survive? How many doctors do they have? So you need these very. This basic vocabulary. And when you have that vocabulary and you give it to human decision makers, they are actually pretty good at making intuitive decisions and prioritize the work. So they can say, okay, I have this much water these regions, so I'll send this much water there and this much water there. The doctors should go, you know, three people go there and two people go there, because I know how many people are injured. And you don't really need sophisticated machinery to do that. You don't need a machine learning algorithm. In fact, machine learning algorithms are probably pretty bad at this because this is an unseen situation. And, you know, you know a lot about machine learning. So, you know, obviously machine learning is very good at repetitive sort of pattern recognition tasks. But there's no pattern here. Everything's new. Every. Everything's different. Everything is chaotic. And humans are actually really good at this. You just have to give them the data so they can. They can make their assessments.

Yeah, it seems to me like the kind of machine learning that would be at play in a crisis like this is the machine learning that's going to be on the sensors that are deployed to, like, look for people or, you know, check damages or see these, evaluate the damage.

Pattern recognition. Right. There's some. Some really cool research papers on things like damage assessment. So if you have satellite or drone footage of particular regions, which are basically the prior state before the damage, and then you send up a drone and take a photo, you can run algorithms that tell you how many building structures are destroyed, what how much, you know, maybe you can even see that electricity lines are destroyed, that maybe water lines are destroyed, and so on. So those are perfect tasks for machine learning. But you would never, at least not today, use a machine learning algorithm to tell you where to send your doctors or how to prioritize certain things. That's just not something that these algorithms are good at.

Yeah, yeah. And then of course, we're deep into this global pandemic, and early on, lots of chaos and people wondering what to do. And your article sort of, to me, kind of calmed that chaotic thinking for me and said, you know, as you implement solutions, ask simple questions first. So would you put the COVID pandemic in context of this, put simple questions first kind of design methodology?

I think so. I think absolutely. I mean, if you look at some examples and caveat, I'm not a medical expert, so I just, you know, repeat what I read in the newspaper. But if you look at the first couple of weeks, you know, what is the simple, simplest question in this pandemic? It's things like how many people are infected? How many people died? These kind of, you know, very basic questions that you're wondering about the outbreak. These questions had a huge variance in answers because they're extremely simple but extremely hard to answer. And in part, they're hard to answer because the testing methodology is just not mature. And this was a new virus, so people just didn't know how to exactly detect it. But in part, they're also hard because the data and the information you need to answer them is so dispersed and is so decentralized. Right. So you have maybe different doctors that do these tests. Maybe you have healthcare workers that go and do home visits. Some patients go to doctors multiple times. Do they report the same test results or do they get reported multiple times? So basically, the data that you have that could answer these simple questions is extremely messy, extremely hard to come by. And it just took a while for the systems to get to basically become established so that you can have standardized reporting that people report with the same maybe data formats, and that decision makers can actually make sense of the data and can report coherent numbers one day after the next. You can make sense of trends all of a sudden. But again, this is like the most basic questions I'm not trying to answer. What's the right strategy for opening up the economy after the crisis? Right. That's like the third order above and beyond this very simple, basic, let's assess the status quo, let's try and stop the bleeding. What type of should we like another example here is, should we close restaurants? Should we close schools? Should we close all of them? Well, we decided to be defensive, probably for a good reason, then close all of them. But now which one should we reopen again? You need some pretty good data. These are very simple questions, but the data you need to answer these are actually. Is very hard to come by and is extremely hard to answer.

Yeah. So I have kind of a statement I wanted to make that kind of underscores that point and then lead into a discussion about privacy and security. And that is I'm here in Virginia and I'm kind of a data guy. So I look at the data the government puts out, and I decided early on I got to track these trends myself. So every day I look at the data and I type them into my own sheet so I can do my own analysis. Because the government was not at first saying, here's the change from the day prior or the week prior. Well, since I type it in my own sheet, I can do that. And maybe I do it because it's therapeutic for me. I know I don't have to, but I had to do that. So now that began to lead me to some conclusions that policymakers should know about. I could look at doubling rates like it's, you know, this many days before it doubles in my county. Either the deaths double or the cases double or hospitalizations double. Now we're into it for so long. I'm now asking questions that the data can't answer. I want to know, we've been in this lockdown. We're all doing so much better. There's 2,000 new cases. What was the cause of every single case? Where did that come from? Now my feeling is, you know, I could build a list that's pretty darn close based on guesses and assumptions and, you know, reasonable assessments. But I want the data. And the data doesn't exist. And the, you know, no journalists are investigating this. The medical researchers, if they're researching it, they're not telling us. They're all busy. There's no real in depth research on where did you get this virus? Which I think some of the reason that none of that data exists is people are worried about privacy.

Yeah. And for good reason. I mean, I'm very glad that you don't have that data. I mean, this is extremely sensitive personal data. And it's probably some of the most sensitive data that we've had a public discussion about in the last couple of years. And so maybe to, I think a good thing we can discuss is sort of the approach to these tracking apps that different countries pursued and maybe contrast that to the model that was recently publicized by Google and Apple.

Yeah, I would love your context on that because the Google and Apple one is the only one that I personally know about.

But before I go into the detail, there just sort of as a framework for how we as a company think about, think about this stuff. So we have a team that we call PCL team. PCL is privacy and civil liberties and their job is basically to make sure that the work that Palantir does is actually in sync and is checked against sort of our principles for how we want to act with data, how we think about data ethics and so on and so on. And so if you talk to those folks, they will give you a couple of principles that we hold that are sort of non negotiable for us as a company. And then we orient everything we do in terms of software development, in terms of the work we do with our customers along these principles. And so just to give you a couple of examples, so one of the biggest principles there is what we call proportionality. So when you use data for a particular purpose, you need to wonder, well, does the use case actually justify that purpose? Sorry, does the use case actually justify that collection and processing and analysis of data. So just to give you a super simple example, and this is maybe in the, in the medical space, so say you want to go to pharmacy and buy some ibuprofen. You would probably not want to, you know, surrender your genetic footprint just to buy ibuprofen. That would be.

Oh, you're so right. Yeah, I don't even use like a shopper's card or anything. That's nobody's business to buy nobody's business.

If you're sick, like it's really nobody's business. You just want your ibuprofen. On the other hand, for medical research a lot of people give up their genetic. Give up in quotes, give up their genetic footprint. But they give it up for a good cause. Right? So there's a proportionality where to enable genetic research and cancer research and sort of make life better for all of us. We as a societies agree that anonymized genetic data is a worthwhile trade off. So there are databases that researchers possess that allow them to do their research. And this is sort of something we as societies agreed on is proportional use. So that's one case proportionality. Another case is this is maybe a little sort of German centric, but in Germany there's a prohibition against what we call generalized suspicion. And what this means is that basically law enforcement authorities are not allowed to do scanning, large scale scanning of data just because they have some suspicion that something might be wrong. So they're not allowed to say, well let me scan all of the records I've ever collected on anything in all sorts of investigations to see if maybe, you know, Bob had a little petty crime last week in the supermarket forbidden. And another principle is, I would call this sort of duality transparency and accountability. So it should be clear what data you collect, what you collect it for, what it's used for. But also, and this is again a slightly Eurocentric topic, but in Europe we've now had the GDPR for a couple of years and the GDPR stipulates that individuals so data subjects can actually should have transparency into where the data is used, where it's stored and what it's used for. And they can even request for it to be deleted in certain cases. And so this kind of transparency mechanism is really, really important. And so if you take these principles, what they mean for data processing systems, it means that they actually should implement some of these accountability mechanisms, right? So for transparency, for instance, the data processing system should tell you who accessed the data and for what purpose, where and how is what data used, where is it going, where is it coming from? Can I delete the data? Is there any sort of automated retention policy? So if the data is unused, it can get deleted. And all of these mechanisms actually help you implement and align your technology and your use of data with these principles. Right? So for instance, if, let's say a law enforcement database has a stipulation that you have to delete unused data after some number of days or months, well, you can't do generalized suspicion scanning because the data is gone. So for exactly this reason we have these laws in place and we have these sort of moral principles for ethical dips or ethical ethical principles, I guess, for data use in place to actually align what is being done with data with how we want data to be used.

Okay, now I know, sorry, go ahead.

And so all of that pretext to like let's come back to the actual Covid thing and the Google and Apple model. So if you apply these principles, I think it's really fascinating. In Germany, the German government had proposed a tracking app. I guess I'm not sure if it's called tracking or tracing. Let's call it tracking up a couple of weeks ago. And the idea was that the government would have a centralized database, all citizens would have an app on their phone. And the app would register who's in touch with who and upload the data to the centralized database. And there was huge outcry in the media by privacy advocates, by individuals, by reporters and journalists, by politicians, that this is a non proportional use of data. It is, you know, the central government is collecting all the data. It might even have location data in it, which sort of was unclear at the time. It's probably a sort of generalized suspicion problem here because a huge proportion of people are healthy. Why should they be suspicious and have to submit their data? Right. Seems a little, seems a little weird. And then there's also transparency and accountability problem. You put your data into this government run database, what happens to it? Nobody knows. And so now contrast this to the model that Apple and Google proposed, which is decentralized and so every person is actually in charge of their own data. The data is collected on the individual devices. There's a pretty sophisticated encryption protocol that makes sure that nobody can sort of see each other's data unless explicitly agreed to. And is that proportional? I think so. So if you contrast the use here, the use case, we're fighting a global pandemic, okay, that's lethal and can kill all of us. And I give up. Well, what do you actually give up? If I want to submit my data, it's voluntary, I submit a list of anonymous contacts that I've had over the last couple of weeks and I just give that data to those contacts and to nobody else. I think that's for me personally, totally proportional. It's a great model. And actually privacy advocates agree. They all praised this model and thought it was great. It doesn't have generalized suspicion. You just sort of surrender the data that is actually. Sorry. You submit data that identifies people who were in contact with you as an infected person. Great. No generalized suspicion. And there's great transparency and accountability. I own my data, it's on my device, I can do with it whatever I want. I know exactly who I'm sending it to in case I'm infected. And it's decentralized storage so that, you know, there's no government that can mess with me in any way. So I think this is actually a really great model. And if you sort of go one level more abstractly, for such a tracing app to work, you need a large fraction of people to use it. Right. It's not going to be useful if 10% of people use it because then you don't really get all the infectious, all the infection paths. Right. They just get Blocked by people who don't use the app. And so you need a large number of people to use the app, but in order for a large number of people to use the app, they have to accept it. And they have to accept it in terms of how they think about how the data is used and whether it's secure and whether actually happy to opt into that model. And so maybe this is actually a great example in Germany where this was sort of self regulating, the government wanted some pretty intrusive mechanism. People didn't want it, and by not wanting it, they basically made the government change their mind and adopt the Google and Apple model. So for me, this was actually a really encouraging, encouraging process where the government learned that what they proposed was not acceptable for society and the much better model won in the end.

Right. That is good. I think it underscores something else too though, and that I think we technologists, including you, should do more to educate policymakers on the fact that you don't have to give up privacy. You should continue to articulate the requirement and technologists can help figure out a way to protect privacy and get the job done.

Absolutely, yeah.

Because I tell you there are a lot of worries. But you mentioned there are privacy advocates out there that can help kind of make sure that we're all thinking privacy. I know, me, early on in the first part of this pandemic, I was worried that privacy would suffer. And I have seen solutions though, like the one you just described that make me feel better about this fact that you don't have to let privacy suffer if you're a policymaker. I think a key point is you need to continue to consistently articulate the requirement for privacy.

Yeah, I think, I mean, I don't want to go into sort of China bashing rant here, but the model that is used in China would never work. In Europe, society just wouldn't accept, just wouldn't work. It's not proportional, it's just not the right model. And so I'm actually very glad that we live in a functioning democracy where people can raise their concerns and where eventually policymakers actually realized that what they proposed was not good. And I'm really thankful for the folks from Apple and Google who figured this out. It's quite an ingenious little protocol that they came up with. I'm also really happy that they collaborate on this. I mean, if you think about, we can get to that later, but if you think about it, this is like a fiercely competitive market, right? Mobile phone ads, that's where they make Huge amounts of money. Apple and Google, it's a duopoly. There are two companies and in this crisis they start to collaborate and work together on the engineering proposal and actually on making this happen and ultimately on affecting policy change as well in a much better direction.

Yeah, maybe it does underscore this point that even competitors are human beings and we're all in this together and we should be. There's so many other aspects of privacy and security that I want to talk about. And I also want to talk about data governance. Maybe that's why you guys are concerned with privacy and security so much. As you have been working in the enterprise world for quite a while and there just by nature you have to understand enterprise governance, like Data Province, for example, and data encryption and protection and auditability.

So our work, I would say, goes a little, you know, it predates the idea of our commercial work and foundry. In particular, our work here comes out of the US government space. So, you know, when the company was founded, we worked with intelligence communities in the United States and data governance, that's sort of, you know, that's where you, that's where everything starts. You can't do data analysis without data governance and without data security. Every piece of data that intelligence analysts work with is tied back to the source. So this is sort of the definition of provenance in a way, right? Where is the data coming from? Who told you this fact? Who's your human contact? Or what database did the data come from? And so when we started building our first software platform, Gotham we built basically started with encoding these primitives. So the idea of provenance, where did data come from? Is it up to date? All of these questions. And then on the other side, things like access control. So government data in the United States in particular, in the intelligence communities has extremely strict access control requirements. There are different levels and different. This is sort of complicated multi dimensional matrix that encodes what information is accessible to what person. And so the first thing we did is basically try to encode that matrix in software. And then you can answer complicated questions like who can access data? Why can they access data? It's very hard to answer. So, you know, you want to do an investigation on some case, well, you have to justify that you can access this piece of information before you access it. But then once you've made that justification, you can also audit it. So after the fact, you can say, well, this analyst access this data when, for what, how and why, what did they do with that data? So all of that can be encoded in software and it really helps sort of develop transparency and trust in data use. And so, you know, you asked about data governance. You know, there's maybe more sort of collaboration aspects to data governance as well. So once you have data assets, you wonder, how can I change it? Is there a process for changing it? Is there, is there maybe code review or peer review for changing data assets? Because by changing data, you'd sort of change facts around the world as encoded in some database. So there must be some process for who's allowed to change it, but also how they're allowed to change it. Maybe, and this is something we do in Foundry in particular, maybe you want to answer questions like, how was data derived? So you start with what we call raw data sets, which are data sets that exist in a number of databases. If you start combining these data sets in a data platform, you might want to keep track of how they were derived. So, so was there a join between columns or some aggregation? Was there some filtering done? Did we fix some values on the process? All of these questions, I would say, are part of data provenance and part of data governance.

Yeah, data provenance is extremely important to me and my entire career it's been a key factor and also security, keeping data encrypted, because that way you really make sure that the authorized users are the only ones seeing it. And in a medical crisis, if somebody has my X ray, of course, I want my entire medical team to have access to that and nobody else.

Exactly.

So I just think I see the natural fit between the data governance, data provenance, data security, and transitioning to a kind of a crisis response. It's just a natural that the data is still protected, you still have privacy rights.

I mean, I would emphatically claim you should not give up on these principles in a crisis. You know, these principles, they're principles. They're not something you just give up just because the next best thing comes around. And our strategy here is that we basically try to encode as much of as many of these principles in software and in frameworks to basically put crisis responders in this particular case, but more generally software developers and analysts to put them in Rails that really help sort of, that basically take care of, or at least help implement these governance aspects and let analysts and software developers focus on their job. So, for instance, just one example in Foundry, which is the software that is used by the NIH in the United States and by the NHS in the UK in our support of the COVID crisis, some examples of Rails are data cannot not be encoded in Foundry. So all data that is in Foundry is always encoded. And an analyst or a person interacting with a system, uploading data, or a software developer writing an application, they got no choice, right? They have to deal with the fact that the data is always encoded.

Or let me ask about that, that data is encrypted, but can anybody with the key read it? Or, I mean, it's. You put on top of that some sort of access controls.

Yeah, exactly, exactly. So basically any API endpoint, so basically any endpoint in the software that would allow you to interact with, with data. So maybe you can read data, write data, mutate data, but you can also maybe edit code, you can write analyses. So all of these sort of actions that a human would do with the data platform, they would correspond to a number of API endpoints in the software. And as policy, but also as sort of as constraints in the way we write software, any of these API endpoints must have authorization on it. So it is impossible today at Palantir to release a piece of software that defines an API endpoint that does not have authorization mechanisms attached. And so we basically use these. These are Rails that are built into other development processes. So when we submit a change request to the Foundry code base, there is a test that runs in our continuous integration platform that says, hey, there's an API endpoint, it doesn't have authorization. That thing cannot be released into software. You have to first deal with authorization.

Is that where the logging is coded into?

Yeah. So audit logging helps with the same thing. So audit logging and access control are baked into every API endpoint. So and what you get from this is basically a complete audit trail of any interaction that any user has with the software. So again, it would know Bob opened this analysis, Bob loaded this data set, Bob this and that. Obviously that's not information that's open to the public, but it's open to the, typically the data governance officers and our customers that would be able to do an investigation of the form. Was this data actually handled correctly? Did they access data they shouldn't have accessed? And so on and so on.

Yeah. And the term you used was Rails. Like there's, like, there's Rails that kind of bound you, but still it's up to the end user enterprise on how they operate this stuff, right?

Yep, yep.

Of course there's like cultural kind of issues, policy issues and maybe doctrine, whatever they have to operate it appropriately.

Correct. But again, you know, the software puts limits on inappropriate use. So I think the encryption one is a perfect example. Even if some of our Customers would ask, can we store data unencrypted? It might be faster for processing. We will tell them, no, you can't. The software enforces that. It's always encrypted. It's non negotiable. It's sort of our way. It's our conviction that data must be encoded. Sorry, encrypted. You can't turn it off. Cool.

Great, that helps. Other social, cultural aspects that we ought to talk about. Especially in context of privacy and security and your technology.

It's really beneficial for companies to have self service policies for data access. So you want to maximize your exposure of data to your, to your individual so they can make better decisions. But then there are barriers. Right? So sometimes for policy reasons, internal policy reasons, like HR data, you can't just throw out to all of your employees. But also for regulatory reasons, you have to put boundaries in place. And so a big part of our education there is how to actually square these two maybe contradictory requirements. You want to maximize access to data, that is, that should be open to see for everyone. But you also have to put boundaries in place so that the legal and regulatory requirements are met. But then you can also do things like what are data sharing policies? So if you have a data set, how can you actually share it with counterparts? Maybe they first have to sign an NDA or do a training program before they can get access to data. And also these sort of processes is something that we can certainly help with.

Cool. Robert, I want to transition to another topic if that's okay with you. Yeah, please. And that is. So the global economy is under a lot of pressure right now and I feel like the economy is going to come back. But people are using the analogy of is it a V shaped recovery or is it a U shaped recovery or is it a long L recovery?

If you find out, let me know so I can treat my stocks accordingly. And buy or sell.

Yeah, that's right. Actually though, I think the way I should ask it is I think you guys are in a position to influence that. All of us are, but especially creators and innovators. The innovators like you guys may help speed the economy up. That's just my optimistic dream. Do you have any thoughts on what humanity can do to kickstart the economy and get it moving faster?

That's a really hard question.

It is a hard one. I'm sorry it's so broad, but I thought if you don't have the answer to that, nobody does.

So I'll start with a sort of personal reflection that I had a couple of weeks ago. So if you look at the last couple of decades, basically technological advances allowed us to globalize and basically make all of our processes and how we interact as societies and as humans extremely complicated. So you look at things like supply chain and just in time delivery, where factories, they don't have any stockpile anymore. They just get, you know, lorries in with the parts they use, they put them together and then Lorry goes out 10 minutes later. But also things like communication, global communication is now instantaneous. Like you know, we are having a conversation right now. Previously we would have exchanged letters or maybe taken a train to talk. And so I think a common buzzword in our private sector companies and industry partners is this idea of connected company. So we were basically at the climax of this buzzword of connected company. Everybody wanted to be as connected as possible and just in time delivery and everything is instantaneous. And I think this crisis might put a little asterisk on that and say, well, connectivity might be good, but you probably also need to care about resilience. So that sure be connected, but think about sort of big large scale effects and how they, how they affect your, your operations. So to come back to your question, how can we help? Well, we have pretty deep partnerships with a lot of large scale organizations both in private sector and in government. And I think what we can, what we can help them is actually make some of their operations more resilient. And I think as part of our sort of strategy, we're going to see a shift away from just caring about connectivity and caring about maybe, I don't know what we call it yet, maybe connected resilience, I don't know, resilient connectivity, we'll see. But so to give you some examples, maybe if you have a complicated supply chain, maybe you want to strategically avoid single points of failure. So you have a demand for one particular piece to put together your car. Maybe you have to have at least a number of different suppliers for that, that part. And maybe those suppliers must not be on, I don't know, I'm making this up on the same continent or use the same railroad or the same freight company, whatever it is. So, and again, you can use data to have these insights and to sort of get an understanding into the complexity of your supply chain. Not just one step out, but maybe 10 steps out. Because maybe the bottleneck is not your direct tier one suppliers, but you know, 10 steps in if you can't get the screw that allows your, the supplier of your supplier of your supplier to put Together that thing you can't get your, you know, whatever it is, the door you put in your car. So this idea of resiliency I think is something that we can certainly help with. And then I think that's a longer term thing. This is not to get out of the crisis immediately, but you know, over the next years. And I actually wonder if on that topic, if we will see sort of the resiliency equivalent of the financial crisis and the policies that came out of the financial crisis. But for, for instance, supply chains. So in the financial crisis 2008, governments and societies realized that financial systems were very brittle. They were extremely connected. All of these loans were connected in wild ways. They were due to that connectivity, very brittle. And they imposed constraints on how brittle they can be, how connected they can be, what you can repackage as, you know, mortgage, subprime, something, something. And so I wonder if maybe governments will actually come up with measures and constraints on, for instance, supply chain connectivity. Maybe they will just require that you have as one simple example, redundant supply chains. And I think, you know, that's a space where I think due to our really deep and broad partnerships across different industries, we can be really helpful. So that was a really abstract one. Another one, and I mentioned this before in the Google and Apple example, I, I believe that we will sort of stand together and collaborate to get out of this crisis. Google and Apple are a great example of this. They're fierce competitors, but they collaborated for, I mean, no real monetary incentive. But just, I think maybe, but just to sort of get us out of this crisis faster. And again, we work with a lot of different companies that I believe do have maybe just humane, but maybe also financial incentives to collaborate. So to take one of the examples from before, we work with suppliers of medical equipment and we work with logistics companies. So maybe they should collaborate and actually figure out how should I prioritize distribution of this medical equipment? Should I send my stuff to the United States because they have more dire needs, or should I send it to Norway because they have higher demand, whatever that is. Again, we have these deep partnerships and I think we can probably facilitate some of the collaboration in that space both on a sort of human to human, just bringing these people together, but also on a data driven decision making, workflow type level.

Yeah, I could see that kind of, that need for collaboration across multiple industries. And then also I could see the need for tracking much more data of every element of the supply chain and everything flowing through the supply chain.

We'll see about that.

At the same time we're having this conversation. There's also these global fieldings of 5G that are occurring and also Wi Fi 6 is a new standard which is just incredible data throughput and more devices allowed on your Wi FI networks. And then there's the space based comms that are being fielded now, multiple platforms there. So I think if we do this right, we're going to be able to really help build in what you call it a connected resilience.

I don't know, I don't want to coin a term that I haven't thought about long. So you know, whatever it is, I'm.

Going to Google that term now and if the domain is available I going to register it. And Robert, I really appreciate your time today and context on all of these issues, so thanks very much. And I do want to say one of the key takeaways I have is that any policymaker anywhere in response to a crisis should understand start with the simple questions first, like you point out your medium article and also never compromise your principles when it comes to privacy and security.

I have nothing to add. That sounds great. Thank you very much Bob. I enjoyed this. Thanks Robert.

We'll see you.

Thanks for listening to this OODA Loop production. For the latest analysis on cybersecurity technology and global risks, please visit www.ooda loop.

Sam.