>> Narrator: Live from Las Vegas, it's theCUBE. Covering AWS re:Invent 2019. Brought to you by Amazon Web Services and Intel, along with it's ecosystem partners. >> Welcome back, about 65,000 here in attendance, at AWS re:Invent 2019. You're watching theCUBE, and I am Stu Miniman, the host for this seg, and happy to welcome back to our program two of our CUBE alumni. Sitting to my right is Abby Fuller, who is the principal technologist for containers and Linux, with Amazon Web Services. Sitting to her right is Deepak Singh, Vice President of Compute Services, also with AWS. Thank you so much for joining us on the program. >> Thanks for having us. >> Thank you for having us. >> Stu: All right, so as I said, both of you have been on the program, and boy your team's been busy. I mean, one of the things I love, first of all, there is a roadmap for many of the things that are going on. So, we do understand what's happen in the future, but, Deepak, maybe just tell us a little bit about your group and kind of the main focus, and let's start there. >> Deepak: So, my group goes beyond containers. It includes things like Linux systems, our high performance computing organization. But for the purposes of re:Invent, let's stick to the containers org. The containers org owns all of AWS's containerized products. So that includes ECS, EKS, Fargate. We also own our service mesh offering, which is App Mesh. So the way I like to think about it is, it's the right way to build applications in the modern era group, and it's a team that stays quite busy, because this is such a hot space to be in. >> Stu: All right, so we're going to talk mostly about containers, but your shirt is talking about the Linux piece. Tell us what your shirt says. >> Deepak: Ahh, yes, this is the only right way to spell AMI. Unfortunately, my previous, when I was in New York, Corey was at the table interviewing me, and I wore this just for him. >> Stu: So, so, so, if it is AMI, then we're going to spend some time talking about EKS. >> Yes. (Abby chuckling) >> And Esses. >> Yes, which one? (Deepak laughing) We will figure that. For AWS is AWS, I think, is how we will do it. So, absolutely, we're not going to talk about ontological arguments in there. But, Abby, a whole lot of new services in the container space. I want to put a pin and put Fargate to aside for a second. >> Abby: Sure. >> Cause lots of things we want to dig into there. But a lot of other things have been announced, in like the last month or so. Maybe, give us a little bit of a view. >> Yeah, I think a couple big ones for us. So, Fargate and Spot, so run on spare Fargate Capacity for up to a 70% discount off of standard Fargate pricing. (mumbling) things like vulnerability image for scanning for images on ECR. We launched, over the last few days as re:Invent, a capacity providers for ECS, which let's you run, split your traffic between on-demand and spot instances in the same cluster. We also launched something called Cluster Auto Scaler. So, some finer-grained control over how your cluster scales in on ECS. >> Stu: All right, want to take a quick step back. So , Fargate, announced a couple of years ago. >> Deepak: Yep. >> Was only first supported on ECS. Definitely, I've talked to lots of customers, very excited about it. >> Deepak: Yep. >> Maybe talk to us a little bit about how Fargate fits in the whole container discussion. >> Deepak: Yeah. >> And we'll hit with the news. >> Yeah, and, actually, a good way to think about it is from a native US standpoint. If you're a customer running containers, the way we think about our services is: You need a place to store those containers, so that's ECR. You could use your own registry, you could pick a third party one, that's fine. But most of our customers just use ECR. Then you pick your containers carrier. That's either ECS or EKS depending on your preferences. And then you need to figure out where you want to run your containers. And, of course, when we launched ECS five years ago, at re:Invent, there was only one way to do it: On EC2 instances. And two years ago, we added in what in our mind is a cloud native natural way to run containers, which is Fargate. So Fargate serves as a runtime compute engine for containers, and you can pick your scheduler on top of it, and go make hay with your applications. So that's kind of how we think the hierarchy works, and it works pretty well for most customers. They'll start off often with EC2 and move to Fargate over time or mix and match, and it's kind of fascinating to see how many customers of ours have decided they want to be all-in on Fargate. Which is a great place to be for us. >> Stu: Okay, but the big news which actually got a good cheer in the key note yesterday, is Fargate for EKS. So what's the importance of this? >> Yeah I think (mumbling) I think it's saying we've been talking to customers about for a while and it's the ability to run your Kubernetes pods on Fargate Capacity. I think it's really speaking to folks love Kubernetes as a tool and as a community, but it can be a pretty significant lift operationally. And with Fargate they can use APIs that they want or the open source tooling that they want but they don't have to worry about provisioning and managing that EC2 capacity. >> Stu: All right, so Deepak I actually was having a conversation with a good AWS customer, yesterday, and he said he actually started out on Kubernetes before EKS existed, on AKS. And migrated over to AWS when EKS became available. And he said Fargate really interests me, but one of the main reasons he does Kubernetes is he wants to have some portability, has some concerns that, he knows what services he uses and how if he needed to move something there, what do you say to customer that says Fargate's interesting me, but I'm concerned I'm going to get locked in if I buy into this model. >> I would say that he shouldn't worry about it, because of two reasons: maybe more than two. One is: the unit in Fargate that you interact with and work on is the same unit that you interact and work on with Kubernetes in general. Which is the Kubernetes pod. It's the broadspec, it's just a pod, no difference. You can take that same pod and run it on Timbuktu cloud and it will still run. So that's part one. The other one is that he's using the same tools, he's using coup CDL. And in fact you can mix and match your Kubernetes casters. You can run 95% of the application on Fargate, and five percent of it on EC2. All they are doing is changing the part annotation, and if you decide you want to run none of it on Fargate, you just flip that and suddenly everything is running on EC2 capacity. So actually think there's that much to worry about, because it's just the same pod. It's still the same tooling, the operational model is a lot simpler. >> So Abby, we've talked to you at DockerCon, and KubeCon, simplicity is not the word that we hear when we talk about this whole container space. >> Abby: Sure. >> Traditionally. How are we doing overall? I mean, I'm watching the community here, and it's like, wait, Fargate sounds cool but where's my persistent volumes? You know, where are we in, you know give us a little bit of the road map as to where we are to make this, you know, simple and managing more of my environment. >> Yeah, I think the way that I like to look at it, right, is that we've spent, and it's not just us, but we spent a lot of time looking at things like patterns and abstractions that help make these work flows easier for developers. And I think one of the launches that's interesting in that vein is the ECS CLI version two, which we launched a few days ago. And that will help you deploy like a production ready containerized application. It'll help you with the CICD angle, it'll help you with the monitoring and the observability. So I think it's about abstracting away, and adding patterns on top to make some of these common operations and work flows really modular and repeatable, and extendable. And then it's about having the ability to customize where I need to. So being able to run on Fargate, but also to use work loads running on EC2 where I need to, and being able to mix and match, and to focus my energy where I really get any benefit from customizing, rather than having to do the whole thing from the ground up. >> Stu: You know, feedback I've gotten from my friends and the app dev community, is that hybrid is more and more becoming a standard deployment model. Obviously things like outposts and some of the other solutions from Amazon are extending the AWS model of doing things, but many of them also look at just Kubernetes, >> Deepak: Yep >> as a layer to do that. How should we be thinking of this from your solutions? >> Deepak: Yeah, so I thought without both, though, if you noticed in Andy's announcement yesterday, among the list of services available on day one were ECS and EKS. And actually app meshes well weren't on the list, but app meshes available on our post on day one as well. I think when we think about customers who want to run and stay in their own capacity and their own data centers, because EKS is built on (mumbling) Kubernetes with no modifications, the same application, as long as they're running on upstream Kubernetes, on their side, will just run on EKS. And there's a number of models that work there. A great model is the kind that SisCo is running, where they will manage it for you in both places. They become the first person you call, and on AWS it's just EKS. And on premise (mumbling) it's what SisCo has decided to build. Our pro-serf team will also help you by example. So I think there's a number of modes that work there but the key part, and it's the reason why we have stayed with (mumbling) stream Kubernetes, is we never want to make someone say, oh we can't use EKS because they're (mumbling). Somehow modified Kubernetes, and I think that is super important for us. >> Stu: Yeah, I mean Abby I know you're an active participant in the community, what do you say to people that look at Amazon, Deepak you talked a little bit about Fargate. You don't need to be concerned to the same images, so speak a little bit, maybe if you could, to Amazon's community participation, and what you're generally hearing from your customers. >> Abby: Yeah, so I think the root of it right is that we're all building with the same building blocks. I think something that Amazon has been really strong at is open sourcing primitive. So, Firecracker last year, I think was a good example. And we, I think we do really well with saying we built this to solve a problem for us, but we think you might want it too. And in terms of community support, we have been open sourcing more over the last year, we open source our road maps in November last year. We run developer previews off the GitHub road map, App Mesh has a public preview channel as well, so we've been trying to involve the community participation earlier and earlier in our product development life cycle, so that, especially with things like service mesh, where it's really pretty new, we can make sure that we have the voice of all our users and our customers, and there, as early as possible. But to get their hands on keyboards to try it out as soon as they can. >> Deepak: And actually a great example of that is, a word that Weave Works has done. Talking about people who can run Kubernetes on AWS and on premises, they have this project called "Weave Ignite" where they're basically running Kubernetes on Firecracker on premises. And then on AWS a customer just runs on EKS, as an example. And that, I think that part has been not everybody realizes that this is possible. But I think the fact that people are doing it is, excites us a lot. >> Stu: All right, I know you're both meeting with a lot of customers this week, maybe Deepak start with you. Any surprises or any misconceptions other than I know there a lot of people wearing teal shirts, with a certain pronunciation. But bring us inside some of the mind set of your customers here. >> Deepak: So actually, our conversation is very consistent. I think the community as a whole, our customer base has a whole, they all want to get to the same place. How can we move really quickly? How can we give our developers the ability to be more productive? Without putting our company at risk, having the right level of governance? Having the right controls, in place? And I think that's mainly consistent theme across the board. I guess the one thing that would be hard to remind people of a little bit, is a lot of people often think Fargate sits on top of ECS and EKS, it sits below that, and actually the fact that now there is an EKS Fargate, people understand that more quickly. Before that it was a little trickier. But other than that, I think our customers almost all. They come from different places, have very similar problems, they want developers to move quickly and develop deliver business value, and platform engineering teams that we speak to want to figure out how to get out of the way. And that's been great! >> It's interesting, Abby, I love your view point from the developer community Andy talked on stage about very much, to do true transformation, there needs to be the leadership driving things down. I'm curious what you're seeing, customers you're talked to, people you had, cause many of these tools we're talking about, you know, started in the developer world. >> Yeah, I mean there's been, like an increasing amount of curiosity around the cultural side of it. So how can I get my team to work like that? How can I get my team to ship more safely, more quickly, but getting operations out of the way? And I think you see more and more interest in that. So how can we build the tools that work the way our developers do? So we get all the thing that we want, so security and compliance and availability. The developers get what they want, which is easy work flows that match the way they want to work. So you see a lot of curiosity around that. So how do we get to the place where we can run everything on Fargate, and benefit from all the new serverless, severless style (mumbling). >> Stu: All right, real quick just give you the final word. Any websites, or events, or things that people should know when they want to learn more and get engaged? >> Yeah, I think I'd send people first and foremost to the GitHub public road maps. It is the easiest, fastest way to let us hear your voice, and what you want to see us build next. I think especially these next couple weeks coming out of re:Invent, as people start to get their hands on what we announced, think I'm really curious for them to take that back, and then be like, this is great, but here's what I want to see next. And I'd love to see that happen on the road maps. >> Yeah, about a month or so ago, maybe a couple months, we started a dedicated blog for containers on AWS site. One of the nice things about it is a lot of the contributors to that blog site are principal engineers, and engineers in our organization. For example, one of our, the principal engineers in my org are Malcolm Featonby, has a whole blog post on how should to think about scaling and best practices. I think I would encourage people who've now seen what we have, all the new services we're developing, and that's where you'll get the details on how you can use them, how we built them, and I encourage everybody to go to that blog site and check out what we're doing. >> Stu: All right, Deepak, Abby, congratulation to you and your team, great progress, and really appreciate (mumbling) are able to look at the road map, and definitely hope to catch up with you both soon. >> Abby: Thanks so much! >> Thank you so much. >> Stu: All right, I'm Stu Miniman, and back with much more, right in a second, thank for watching theCube. (Techno music)

>> Live from the MIT campus in Cambridge, Massachusetts, it's theCUBE, covering the 12th Annual MIT Chief Data Officer and Information Quality Symposium. Brought to you by SiliconANGLE Media. >> Welcome back to theCUBE's coverage of MIT CDOIQ here in Cambridge, Massachusetts. I'm your host Rebecca Knight along with my cohost Peter Burris. We have two guests on this segment. We have Cortnie Abercrombie, she is the founder of the nonprofit AI Truth, and Carl Gerber, who is the managing partner at Global Data Analytics Leaders. Thanks so much for coming on theCUBE Cortnie and Carl. >> Thank you. >> Thank you. >> So I want to start by just having you introduce yourselves to our viewers, what you do. So tell us a little bit about AI Truth, Cortnie. >> So this was born out of a passion. As I, the last gig I had at IBM, everybody knows me for chief data officer and what I did with that, but the more recent role that I had was developing custom offerings for Fortune 500 in the AI solutions area, so as I would go meet and see different clients, and talk with them and start to look at different processes for how you implement AI solutions, it became very clear that not everybody is attuned, just because they're the ones funding the project or even initiating the purpose of the project, the business leaders don't necessarily know how these things work or run or what can go wrong with them. And on the flip side of that, we have very ambitious up-and-comer-type data scientists who are just trying to fulfill the mission, you know, the talent at hand, and they get really swept up in it. To the point where you can even see that data's getting bartered back and forth with any real governance over it or policies in place to say, "Hey, is that right? Should we have gotten that kind of information?" Which leads us into things like the creepy factor. Like, you know target (laughs) and some of these cases that are well-known. And so, as I saw some of these mistakes happening that were costing brand reputation, our return on investment, or possibly even creating opportunities for risk for the companies and for the business leaders, I felt like someone's got to take one for the team here and go out and start educating people on how this stuff actually works, what the issues can be and how to prevent those issues, and then also what do you do when things do go wrong, how do you fix it? So that's the mission of AI Truth and I have a book. Yes, power to the people, but you know really my main concern was concerned individuals, because I think we've all been affected when we've sent and email and all of a sudden we get a weird ad, and we're like, "Hey, what, they should not, is somebody reading my email?" You know, and we feel this, just, offense-- >> And the answer is yes. >> Yes, and they are, they are. So I mean, we, but we need to know because the only way we can empower ourselves to do something is to actually know how it works. So, that's what my missions is to try and do. So, for the concerned individuals out there, I am writing a book to kind of encapsulate all the experiences that I had so people know where to look and what they can actually do, because you'll be less fearful if you know, "Hey, I can download DuckDuckGo for my browser, or my search engine I mean, and Epic for my browser, and some private, you know, private offerings instead of the typical free offerings. There's not an answer for Facebook yet though. >> So, (laughs) we'll get there. Carl, tell us a little bit about Global Data Analytics Leaders. >> So, I launched Analytics Leaders and CDO Coach after a long career in corporate America. I started building an executive information system when I was in the military for a four-star commander, and I've really done a lot in data analytics throughout my career. Most recently, starting a CDO function at two large multinational companies in leading global transformation programs. And, what I've experienced is even though the industries may vary a little bit, the challenges are the same and the patterns of behavior are the same, both the good and bad behavior, bad habits around the data. And, through the course of my career, I've developed these frameworks and playbooks and just ways to get a repeatable outcome and bring these new technologies like machine learning to bear to really overcome the challenges that I've seen. And what I've seen is a lot of the current thinking is we're solving these data management problems manually. You know, we all hear the complaints about the people who are analysts and data scientists spending 70, 80% of their time being a data gatherer and not really generating insight from the data itself and making it actionable. Well, that's why we have computer systems, right? But that large-scale technology in automation hasn't really served us well, because we think in silos, right? We fund these projects based on departments and divisions. We acquire companies through mergers and acquisitions. And the CDO role has emerged because we need to think about this, all the data that an enterprise uses, horizontally. And with that, I bring a high degree of automation, things like machine learning, to solve those problems. So, I'm now bottling that and advising my clients. And at the same time, the CDO role is where the CIO role was 20 years ago. We're really in it's infancy, and so you see companies define it differently, have different expectations. People are filling the roles that may have not done this before, and so I provide the coaching services there. It's like a professional golfer who has a swing coach. So I come in and I help the data executives with upping their game. >> Well, it's interesting, I actually said the CIO role 40 years ago. But, here's why. If we look back in the 1970s, hardcore financial systems were made possible by the technology which allowed us to run businesses like a portfolio: Jack Welch, the GE model. That was not possible if you didn't have a common asset management system, if you didn't have a common cached management system, etc. And so, when we started creating those common systems, we needed someone that could describe how that shared asset was going to be used within the organization. And we went from the DP manager in HR, the DP manager within finance, to the CIO. And in many respects, we're doing the same thing, right? We're talking about data in a lot of different places and now the business is saying, "We can bring this data together in new and interesting ways into more a shared asset, and we need someone that can help administer that process, and you know, navigate between different groups and different needs and whatnot." Is that kind of what you guys are seeing? >> Oh yeah. >> Yeah. >> Well you know once I get to talking (laughs). For me, I can going right back to the newer technologies like AI and IOT that are coming from externally into your organization, and then also the fact that we're seeing bartering at an unprec... of data at an unprecedented level before. And yet, what the chief data officer role originally did was look at data internally, and structured data mostly. But now, we're asking them to step out of their comfort zone and start looking at all these unknown, niche data broker firms that may or may not be ethical in how they're... I mean, I... look I tell people, "If you hear the word scrape, you run." No scraping, we don't want scraped data, no, no, no (laugh). But I mean, but that's what we're talking about-- >> Well, what do you mean by scraped data, 'cause that's important? >> Well, this is a well-known data science practice. And it's not that... nobody's being malicious here, nobody's trying to have a malintent, but I think it's just data scientists are just scruffy, they roll up their sleeves and they get data however they can. And so, the practice emerged. Look, they're built off of open-source software and everything's free, right, for them, for the most part? So they just start reading in screens and things that are available that you could see, they can optical character read it in, or they can do it however without having to have a subscription to any of that data, without having to have permission to any of that data. It's, "I can see it, so it's mine." But you know, that doesn't work in candy stores. We can't just go, or jewelry stores in my case, I mean, you can't just say, "I like that diamond earring, or whatever, I'm just going to take it because I can see it." (laughs) So, I mean, yeah we got to... that's scraping though. >> And the implications of that are suddenly now you've got a great new business initiative and somebody finds out that you used their private data in that initiative, and now they've got a claim on that asset. >> Right. And this is where things start to get super hairy, and you just want to make sure that you're being on the up-and-up with your data practices and you data ethics, because, in my opinion, 90% of what's gone wrong in AI or the fear factor of AI is that your privacy's getting violated and then you're labeled with data that you may or may not know even exists half the time. I mean. >> So, what's the answer? I mean as you were talking about these data scientists are scrappy, scruffy, roll-up-your-sleeves kind of people, and they are coming up with new ideas, new innovations that sometimes are good-- >> Oh yes, they are. >> So what, so what is the answer? Is this this code of ethics? Is it a... sort of similar to a Hippocratic Oath? I mean how would you, what do you think? >> So, it's a multidimensional problem. Cortnie and I were talking earlier that you have to have more transparency into the models you're creating, and that means a significant validation process. And that's where the chief data officer partners with folks in risk and other areas and the data science team around getting more transparency and visibility into what's the data that's feeding into it? Is it really the authoritative data of the company? And as Cortnie points out, do we even have the rights to that data that's feeding our models? And so, by bringing that transparency and a little more validation before you actually start making key, bet-the-business decisions on the outcomes of these models, you need to look at how you're vetting them. >> And the vetting process is part technology, part culture, part process, it goes back to that people process technology trying. >> Yeah, absolutely, know where your data came from. Why are you doing this model? What are you doing to do with the outcomes? Are you actually going to do something with it or are you going to ignore it? Under what conditions will you empower a decision-maker to use the information that is the output of the model? A lot of these things, you have to think through when you want to operationalize it. It's not just, "I'm going to go get a bunch of data wherever I can, I put a model together. Here, don't you like the results?" >> But this is Silicon Valley way, right? An MVP for everything and you just let it run until... you can't. >> That's a great point Cortnie (laughs) I've always believed, and I want to test this with you, we talk about people process technology about information, we never talk about people process technology and information of information. There's a manner of respects what we're talking about is making explicit the information about... information, the metadata, and how we manage that and how we treat that, and how we defuse that, and how we turn that, the metadata itself, into models to try to govern and guide utilization of this. That's especially important in AI world, isn't it? >> I start with this. For me, it's simple, I mean, but everything he said was true. But, I try to keep it to this: it's about free will. If I said you can do that with my data, to me it's always my data. I don't care if it's on Facebook, I don't care where it is and I don't care if it's free or not, it's still my data. Even if it's X23andMe, or 23andMe, sorry, and they've taken the swab, or whether it's Facebook or I did a google search, I don't care, it's still my data. So if you ask me if it's okay to do a certain type of thing, then maybe I will consent to that. But I should at least be given an option. And no, be given the transparency. So it's all about free will. So in my mind, as long as you're always providing some sort of free will (laughs), the ability for me to having a decision to say, "Yes, I want to participate in that," or, "Yes, you can label me as whatever label I'm getting, Trump or a pro-Hillary or Obam-whatever, name whatever issue of the day is," then I'm okay with that as long as I get a choice. >> Let's go back to it, I want to build on that if I can, because, and then I want to ask you a question about it Carl, the issue of free will presupposes that both sides know exactly what's going into the data. So for example, if I have a medical procedure, I can sit down on that form and I can say, "Whatever happens is my responsibility." But if bad things happen because of malfeasance, guess what? That piece of paper's worthless and I can sue. Because the doctor and the medical provider is supposed to know more about what's going on than I do. >> Right. >> Does the same thing exist? You talked earlier about governance and some of the culture imperatives and transparency, doesn't that same thing exist? And I'm going to ask you a question: is that part of your nonprofit is to try to raise the bar for everybody? But doesn't that same notion exist, that at the end of the day, you don't... You do have information asymmetries, both sides don't know how the data's being used because of the nature of data? >> Right. That's why you're seeing the emergence of all these data privacy laws. And so what I'm advising executives and the board and my clients is we need to step back and think bigger about this. We need to think about as not just GDPR, the European scope, it's global data privacy. And if we look at the motivation, why are we doing this? Are we doing it just because we have to be regulatory-compliant 'cause there's a law in the books, or should we reframe it and say, "This is really about the user experience, the customer experience." This is a touchpoint that my customers have with my company. How transparent should I be with what data I have about you, how I'm using it, how I'm sharing it, and is there a way that I can turn this into a positive instead of it's just, "I'm doing this because I have to for regulatory-compliance." And so, I believe if you really examine the motivation and look at it from more of the carrot and less of the stick, you're going to find that you're more motivated to do it, you're going to be more transparent with your customers, and you're going to share, and you're ultimately going to protect that data more closely because you want to build that trust with your customers. And then lastly, let's face it, this is the data we want to analyze, right? This is the authenticated data we want to give to the data scientists, so I just flip that whole thing on its head. We do for these reasons and we increase the transparency and trust. >> So Cortnie, let me bring it back to you. >> Okay. >> That presupposes, again, an up-leveling of knowledge about data privacy not just for the executive but also for the consumer. How are you going to do that? >> Personally, I'm going to come back to free will again, and I'm also going to add: harm impacts. We need to start thinking impact assessments instead of governance, quite frankly. We need to start looking at if I, you know, start using a FICO score as a proxy for another piece of information, like a crime record in a certain district of whatever, as a way to understand how responsible you are and whether or not your car is going to get broken into, and now you have to pay more. Well, you're... if you always use a FICO score, for example, as a proxy for responsibility which, let's face it, once a data scientist latches onto something, they share it with everybody 'cause that's how they are, right? They love that and I love that about them, quite frankly. But, what I don't like is it propagates, and then before you know it, the people who are of lesser financial means, it's getting propagated because now they're going to be... Every AI pricing model is going to use FICO score as a-- >> And they're priced out of the market. >> And they're priced out of the market and how is that fair? And there's a whole group, I think you know about the Fairness Accountability Transparency group that, you know, kind of watch dogs this stuff. But I think business leaders as a whole don't really think through to that level like, "If I do this, then this this and this could incur--" >> So what would be the one thing you could say if, corporate America's listening. >> Let's do impact. Let's do impact assessments. If you're going to cost someone their livelihood, or you're going to cost them thousands of dollars, then let's put more scrutiny, let's put more government validation. To your point, let's put some... 'cause not everything needs the nth level. Like, if I present you with a blue sweater instead of a red sweater on google or whatever, (laughs) You know, that's not going to harm you. But it will harm you if I give you a teacher assessment that's based on something that you have no control over, and now you're fired because you've been laid off 'cause your rating was bad. >> This is a great conversation. Let me... Let me add something different, 'cause... Or say it a different way, and tell me if you agree. In many respects, it's: Does this practice increase inclusion or does this practice decrease inclusion? This is not some goofy, social thing, this is: Are you making your market bigger or are you making your market smaller? Because the last thing you want is that the participation by people ends with: You can't play because of some algorithmic response we had. So maybe the question of inclusion becomes a key issue. Would you agree with that? >> I do agree with it, and I still think there's levels even to inclusion. >> Of course. >> Like, you know, being a part of the blue sweater club versus the (laughs) versus, "I don't want to be a convict," you know, suddenly because of some record you found, or association with someone else. And let's just face it, a lot of these algorithmic models do do these kinds of things where they... They use n+1, you know, a lot... you know what I'm saying. And so you're associated naturally with the next person closest to you, and that's not always the right thing to do, right? So, in some ways, and so I'm positing just little bit of a new idea here, you're creating some policies, whether you're being, and we were just talking about this, but whether you're being implicit about them or explicit, more likely you're being implicit because you're just you're summarily deciding. Well, okay, I have just decided in the credit score example, that if you don't have a good credit threshold... But where in your policies and your corporate policy did it ever say that people of lesser financial means should be excluded from being able to have good car insurance for... 'cause now, the same goes with like Facebook. Some people feel like they're going to have to opt of of life, I mean, if they don't-- >> (laughs) Opt out of life. >> I mean like, seriously, when you think about grandparents who are excluded, you know, out in whatever Timbuktu place they live, and all their families are somewhere else, and the only way that they get to see is, you know, on Facebook. >> Go back to the issue you raised earlier about "Somebody read my email," I can tell you, as a person with a couple of more elderly grandparents, they inadvertently shared some information with me on Facebook about a health condition that they had. You know how grotesque the response of Facebook was to that? And, it affected me to because they had my name in it. They didn't know any better. >> Sometimes there's a stigma. Sometimes things become a stigma as well. There's an emotional response. When I put the article out about why I left IBM to start this new AI Truth nonprofit, the responses I got back that were so immediate were emotional responses about how this stuff affects people. That they're scared of what this means. Can people come after my kids or my grandkids? And if you think about how genetic information can get used, you're not just hosing yourself. I mean, breast cancer genes, I believe, aren't they, like... They run through families, so, I-- >> And they're pretty well-understood. >> If someone swabs my, and uses it and swaps it with other data, you know, people, all of a sudden, not just me is affected, but my whole entire lineage, I mean... It's hard to think of that, but... it's true (laughs). >> These are real life and death... these are-- >> Not just today, but for the future. And in many respects, it's that notion of inclusion... Going back to it, now I'm making something up, but not entirely, but going back to some of the stuff that you were talking about, Carl, the decisions we make about data today, we want to ensure that we know that there's value in the options for how we use that data in the future. So, the issue of inclusion is not just about people, but it's also about other activities, or other things that we might be able to do with data because of the nature of data. I think we always have to have an options approach to thinking about... as we make data decisions. Would you agree with that? Yes, because you know, data's not absolute. So, you can measure something and you can look at the data quality, you can look at the inputs to a model, whatever, but you still have to have that human element of, "Are you we doing the right thing?" You know, the data should guide us in our decisions, but I don't think it's ever an absolute. It's a range of options, and we chose this options for this reason. >> Right, so are we doing the right thing and do no harm too? Carl, Cortnie, we could talk all day, this has been a really fun conversation. >> Oh yeah, and we have. (laughter) >> But we're out of time. I'm Rebecca Knight for Peter Burris, we will have more from MIT CDOIQ in just a little bit. (upbeat music)