The Six Five On the Road at AWS re:Invent 2022. Patrick Moorhead and Daniel Newman sit down with Susanne Seitinger, Director of AI & ML, AWS.
Their discussion covers:
- AWS end-to-end data strategy
- How SageMaker & machine learning is leveraged in the data strategy
- Responsible AI
Be sure to subscribe to The Six Five Webcast so you never miss an episode.
You can watch the full video here:
You can listen to the conversation here:
Disclaimer: The Six Five On the Road is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded, and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we do not ask that you treat us as such.
Patrick Moorhead: Hi, this is Pat Moorhead. We are live at AWS Reinvent 2022. We are on the road doing what we love to do and that’s talking tech and talking with the awesome people who bring it to the world and society. Daniel, what an event. I mean, it is amazing. I mean, I could barely get through the halls today, walking from meeting to meeting. It was great.
Daniel Newman: I love it. But I love that by the way. If you want to litmus test for how much demand and interest there is in technology and how enterprises are looking at technology to solve its most complex problems right now. Whether that’s the economics of running your business or being able to deliver the customer experience of the future, you come here at AWS Reinvent and you just kind of feel the energy. And by the way, you look at the partners, the ecosystems, the customers, the names that are here, the people that are on stage. So much energy. That’s why we love being here. And of course we love having the chance to sit down with some of those people here on the Six Five on the Road.
Patrick Moorhead: No, absolutely. And the cloud’s 15 years old. It’s a teenager, kind of gangly. We haven’t figured out all the challenges, but also haven’t figured out the way to fully take advantage of it. And the biggest way that enterprises are taking advantage of the technology today is basically big data and AI and ML. And that is leads me into introducing our amazing guest. Susanne, how are you?
Susanne Seitinger: Hi. Good to see you guys.
Patrick Moorhead: Great to be on the show. First time Six Five.
Susanne Seitinger: Yeah, my first time. First time Six Five attendee, first Reinvent. Yeah. Super psyched to be here with you guys.
Patrick Moorhead: Yeah. Maybe a great place to start is maybe talk about what you do for the company.
Susanne Seitinger: Absolutely. I lead product marketing for AI and machine learning services and get to experiment with some of that cutting edge technology you were just talking about.
Patrick Moorhead: No, I love that. I mean, I spent 15 years in product marketing.
Susanne Seitinger: Oh, wow.
Patrick Moorhead: And I love it. I’m a product person. I don’t have a PhD like you, but I still manage to figure out product marketing. But working with the engineers, running the business. I mean, it’s a noble cause and I love that.
Susanne Seitinger: It’s all about empathy, right? That’s what I love about it. You have to put yourself in other people’s shoes and that’s really what it’s all about.
Daniel Newman: I love her humility. She lie, I’m a MIT PhD, but here we already know how much experience you have.
Patrick Moorhead: I know. How about that?
Daniel Newman: This is the fun of our show, but-
Patrick Moorhead: … state school, I pretend to be better than I am. But no seriously, product management is like you have all the responsibility and none of the authority.
Susanne Seitinger: It’s my favorite place to be.
Patrick Moorhead: No, it’s like you’re in between customers, engineers and coming up with these amazing ways of actually communicating to other humans very complex things and topics.
Susanne Seitinger: I mean, honestly, I think if there’s one useful thing that getting a PhD actually helped me with is translation. It’s not so much about the actual rigor. All that stuff is really helpful, but it’s really about translating ideas and concepts and you have to be able to do that and that’s what product managers have to do all day long. Whether it’s internally to engineering teams or externally to customers. And it’s a lot of fun and it’s challenging, especially the more complicated the topics become.
Patrick Moorhead: Oh, you want to-
Daniel Newman: Yeah. I was just actually going to, we don’t talk to a lot of first timers.
Patrick Moorhead: I know.
Daniel Newman: And so impressions, I know you’ve worked for big tech companies, you’ve got some background, but as you’ve gotten out, walked the floor, the announcements, what’s your first impression of Reinvent?
Susanne Seitinger: I mean, the scale is unreal. The scale is unreal. So I mean, you were talking about the cloud is only 15 years old. It seems like cloud has been around for much longer. If you look at the scale of this event, it’s been absolutely overwhelming. And I’m so encouraged to see so many people back after the pandemic. I had no idea what to expect. But everyone is here from around the world. So I think the international piece of it, the scale of it, and then the sheer fun that people are having, I really love all the wordplay with Reinvent replay. I mean it really is the embodiment of kind of experimentation and the future. Really loving it.
Patrick Moorhead: If you get the chance, you need to hit Remarks. If you haven’t been to that, or maybe you were there before, Daniel and I in the Six Five, wasn’t it great?
Susanne Seitinger: Oh my God. My kids were like, wait, what? You’re working, I’m not so sure you’re actually working. Mom, you took a lot of pictures of us.
Patrick Moorhead: Right, exactly. The robot dog that shows up at every trade show.
Daniel Newman: I came home and told my kids that we’re going to be living in some interplanetary space station in like 50 years because they were talking about these. And I’m like, this stuff-
Susanne Seitinger: It seems too real. Right?
Patrick Moorhead: So we are here to talk about-
Susanne Seitinger: Oh right, right.
Patrick Moorhead: Exactly.
Daniel Newman: He is going to help us.
Patrick Moorhead: It sounds good. And it all starts with data. I mean, it’s your garbage in, garbage out, but it’s got to be good. But maybe we start at that fundamental level of what kind of data is being processed, manipulated today. Back in the old days, it was great. Hey, SQL database. We could fit it nice. Perfect. That is just not the case today, is it?
Susanne Seitinger: It’s so true. I mean, 80% of all of our data is actually unstructured. It is not neatly organized. It is not ready to be applied, used in a machine learning model. So there’s a ton of work that has to happen with the data leading up to actually being able to do something meaningful. And we forget how much effort that is. And it is absolutely critical in order to then reap the benefits of machine learning and AI.
Patrick Moorhead: Well, unstructured data, it’s unstructured, but there’s hundreds of different kinds of data and different strata inside of each of those. And you actually had a big announcement here.
Susanne Seitinger: We sure did.
Patrick Moorhead: To hit two other elements of data out there. Geospatial?
Susanne Seitinger: Absolutely. And this is a topic that’s close to home for me. We were talking about looking back, and my background is in smart cities and I’m truly passionate about how technology can be applied to cities. And one of the biggest barriers for a lot of thinkers in that space is dealing with geospatial data. And it’s problematic for so many reasons. The scale of the data-
Patrick Moorhead: Just for the sake of the audience, could you explain what-
Susanne Seitinger: What does that actually mean? You’re right. Yeah. So think about everything that has to do with maps and mapping and imagery, satellite imagery, all the things we’ve now become accustomed to in dealing with different apps. It’s all of that data that actually locates things in a place and connects data to a particular location in time. And so that’s what geospatial data means.
Patrick Moorhead: No, I appreciate that. And you talk a little bit about the announcement. I mean, how absolutely do your customers benefit from what is it and how do they benefit from it?
Susanne Seitinger: Yeah, so think about all the problems and challenges that governments face, that critical infrastructure companies face in understanding and managing infrastructure in actual environments that are complex, messy, difficult to actually grasp and understand. And so our new geospatial capabilities embedded in Amazon SageMaker will allow all of our customers to use and access their geospatial data much more readily. It is going to facilitate using the data, making it actually manageable, meaning turning it into bite size pieces so they can do their different transformations on that data.
It’s going to allow them to access publicly available data, other satellite imagery or other data, and correlate the data much more readily and easily. And then it’s going to allow them to actually use it as part of their machine learning work and connect it to their actual problems, which might be everything from thinking about how to prevent the next wildfire to planning the next city and thinking about the best way to route their bus lines, for example. So a lot of opportunities for them from the beginning through the end of the machine learning life cycle.
Patrick Moorhead: Cool. So you started alluding to, you talked about bus cycles, you talked about smart cities. So I’ll give you the open flexibility to answer this with any industry you want, but when you look at geospatial, I get it. But people out there kind of like, okay, apply this for me. So what are some of the ways that you could apply this geospatial? Give us a couple use cases.
Susanne Seitinger: Yeah, I mean think about, for example, on the West Coast for example, when all the different fire departments are trying to figure out how to deploy their resources, they have a lot of information, historical information. They know where things are or might be, but they don’t know exactly how to predict where they want to deploy their resources going forward. They can use these tools to actually implement this and do this in a much more fine grained way and actually end up using their resources more effectively and protecting lives as an end result indirectly.
And this can help them along that path. Think about city planners. City planners have a long, long span that they’re sort of planning out and they have no way really of visualizing, predicting, experimenting with different outcomes. It’s very expensive to build a city. Really if you want to understand different potential outcomes, you need ML to start to really infer what might happen. And so having access to these kinds of geospatial capabilities is completely game changing because it allows you to create connections between data that might never have been connected, whether you’re comparing educational outcomes, income, other access to infrastructure. You can do all that with these capabilities.
Patrick Moorhead: What other I’d say a great use case and my brain went immediately to city planning and the environment, changes in it and farmers and just the changes that happen and how to manage that. What other types of challenges are you seeing come top of mind for your customers and what are you doing about it?
Susanne Seitinger: As we go beyond, as we’re building cities, for example, a big topic is often equity and thinking about how to deploy technology in a way that is actually equitable, is accessible for all people. So a lot of the work we’ve been doing in the last months and years is really around what are the implications of deploying this technology when it comes to ethics and responsibility? And that’s a shared challenge with our customers. That’s not something we own on our own, but it is something where we’re trying to figure out how to embed tools in Amazon SageMaker, but also provide other guidance and resources so that they can actually basically from the outset of engaging with machine learning think about the downstream implications and how they’re aligned with their own values as an organization, which can be different depending on the use case, the scenario, the geography.
Daniel Newman: Yeah, it’s kind of interesting. There are so many different ways you can apply it. But you actually said something about the equity side then you talked about responsible. Let’s unpack that one a little bit. So-
Susanne Seitinger: We’re going to go there.
Daniel Newman: Well, we’re at this kind of intersection where AI is being really quickly deployed and we as a society almost come to accept that tech will come first and regulating bodies might come later. We’re seeing it with vehicles and that’s probably the most well understood example of it. But the way it’s being implemented, every app that we use every day, how it decides what we see from an advertising standpoint to how companies see us by the way. So a lot of people are looking at this responsibility thing. What are your thoughts on this whole responsibility and what is you, your team, what is AWS’s approach to driving responsibility in AI?
Susanne Seitinger: Yeah, I mean I think it’s a great question and it is a complicated multifaceted question. Primarily, we’re starting from a really humble position. We understand that this is a complex topic that is going to require so many different stakeholders. Stakeholders in research, stakeholders in industry, stakeholders in government, different partners in the ecosystem around the cloud are going to participate in creating what this future needs to be. And it is going to happen differently in all different parts of the world.
And so that’s kind of the starting point. On top of that, the way we’ve been thinking about this is really from a shift from the theoretical conversations about this to what does it mean in practice? How do we actually build? Everyone here at this conference is a builder at heart. You got everyone here, we’re all builders at heart. We love to make things and try to make the world better with the things we make. And that always has consequences. And so thinking about how we build tools for example, within Amazon SageMaker, that can actually help you monitor machine learning models over time and understand if they’re drifting and aren’t doing what you expected at the outset. So how do we create tools-
Patrick Moorhead: What do you mean by model drifting? Well, I think I know which is new data comes in after it’s been trained and it moves in a certain direction. Is that-
Susanne Seitinger: That’s exactly right. That’s exactly right. It stops to behave the way you had expected. The inferences stopped being what they needed to be or what you expected them to be based on what you had set at the beginning from a quality perspective. And so now you-
Patrick Moorhead: [inaudible 00:13:48] dots by the way. Definitely not going that direction.
Daniel Newman: I wasn’t saying we get specific. I was just saying that there were a couple of really great drifting examples of what happened when some bots got deployed and absolutely nobody watched them for like five minutes.
Patrick Moorhead: Okay, now I know what talking about.
Susanne Seitinger: No, it’s true. We need to be careful. These things can happen quickly, but at the same time we need these tools so that we can empower our teams to really work with the insights as they’re happening.
Patrick Moorhead: Well, and I saw an announcement that she made here. It’s basically data masking where you’ve got a amazing set of data, maybe it’s health data and you don’t want have any PII, nothing that what-
Susanne Seitinger: Absolutely.
Patrick Moorhead: But people can do research on it at the same time. And that is powerful because in many ways there are some technologies that we don’t use that could have incredible societal positive benefits. And setting up ways to, first of all, tools inside of SageMaker that can increase the chance of it being right. And I think it’s an education. And if you think about it, we have a contract, a societal contract, and we ultimately arrive there but we’re kind of crashing. And by the way the sign wave gets, as we move forward, it gets better. Understandable. Sometimes we even shift as society. If you’d have told me 30 years ago that hey, in real time I’m going to post where I am and pictures of my family and all of this stuff, I would’ve said there is no way. There’s no way that’s going to happen. But there-
Daniel Newman: Twitter feed.
Patrick Moorhead: Well, benefit macro social media of being able to keep with family and friends in real time and it’s really a trade off and I’m just glad you have tools and you’re putting it in there to help people do better things, have less people crash into the guardrails and make that sine wave get a lot better.
Susanne Seitinger: I mean, as machine learning becomes more widespread in scale and more people within companies are using it, it really is everyone’s job to also think about these things. And for example, if you think about the administrators who are working on governance, and one of the things we’re trying to help them with Amazon SageMaker role manager is to help them deploy the right access controls to people in an instant and not be concerned that someone might get access to data to your earlier point on healthcare that they shouldn’t have access to.
But then also not stop them from innovating when that’s their next task and instead of waiting for a week or two, but until they get access to the tools that they need. So really building that into the way we think about all of machine learning is critical. And the same goes for other tools like Dashboard that really helps them bring in data from clarify and model monitor in the same pane of glass. So they’re seeing it there at a glance. They know if something’s wrong and then they can go reach out to the other ML practitioners and figure out what’s going on and really address it more quickly and really also apply the same standard principles across the organization. So that’s the other critical benefit here is that you’re creating a collaborative environment and the best practices are disseminated immediately and not dependent on kind of training and painstaking outreach, which is really the only way to do it now.
Daniel Newman: So you alluded to it, I want to make sure I get this though. Transparency’s a big thing and a lot of people are trying to like, well, what’s in the model?
Susanne Seitinger: Right. Right.
Daniel Newman: And of course people that have not done a PhD and studied AI, I understand how it works, but I couldn’t go into the actual tools that Python or whatever, see what’s built and be like, oh, I get what… How do we help? How does AWS help so that people know that not only does the model work, but it’s actually being done with integrity, it’s done with in a way that is transparent and ethical?
Susanne Seitinger: I mean, that’s a great question and it has to do with this notion of actually getting the information and insights necessary as a customer to be able to work with the services that we provide. And one of the things I’m super excited about that we talked about here at Reinvent are our new AI service cards and these AWS AI service cards are what we think the first three that we just launched are kind of a first step down this path of providing transparency to our customers around our AI services.
And it really digs deeper through quantitative insights and guidance around the actual application of the technology to help customers figure out how to build responsibly. And so there’s a transparency aspect to that and a trust building aspect to that that is so important because basically we’re putting out there to have a conversation about what is the best approach and really ground it in the use case, in the application. These things, that’s where the rubber meets the road and that’s why we want to make these service cards really actionable and very specific to certain scenarios. So folks understand that what we’re talking about in context, it’s not separate from the context of the use and the actual building.
Patrick Moorhead: That’s fascinating. I mean, sometimes when we go and have these talks, you’re not too sure kind of where it’s going to start and where it’s going to end. And Susanne, I just want to thank you for just illuminating on exactly… People understand now a lot more what you’re trying to do, how you’re differentiated and the problems that you’re solving. And I really love this element too of the guardrail system.
Susanne Seitinger: Right? I love that. Yeah.
Patrick Moorhead: And you look at the way that security is, I mean a lot of people used to say, hey, just lock everything down, make it as hard as possible for users to get anything done. And then users just went all the way around it and then they realized, okay, we got to pull back and make it so smart. So they’re not hitting either the guardrail of not getting anything done or being insecure. And I kind of see a very similar vein for an AI and ML getting that work done but also having a responsible guardrail on the other side and moving it down there. And I like that it’s integrated into essentially your AI, ML, and IDE. So it’s data and the algorithms all the way from data proc coming in all the way to the very end and running the model on an endpoint.
Susanne Seitinger: Absolutely. And then turning it right back into a virtuous cycle where the teams internally are then taking those insights from customers and putting it right back in the development cycle so that they can apply all the new lessons learned from the actual real world scenarios and use it to just continue to get better over time. And not making it something separate but making it really a core part of their engineering process and not sort of this extra check the box exercise. It’s really part of the core effort that they have going on as they’re building.
Daniel Newman: You can definitely tell Susanne’s an empath. No, I’m being sincere. This is the kind of role that you really need. Because everything about it is really naturally not empathic. It’s just data that’s being put into a model. So what I’m saying is it’s really good to think that people that are behind building it want to add that human element. And empathy’s the one thing I know you’re trying to build into systems that you can’t quite get.
Susanne Seitinger: Right. Well, not yet. Not yet. Well, and like you said, behind all these systems are actual people and that’s why we always talk about people, process, and technology because it’s those three things together that actually is how the world works.
Daniel Newman: It’s still the thing, people process.
Patrick Moorhead: You can just do a mic drop right there. Susanne, thank you so much for-
Susanne Seitinger: Appreciate you guys.
Patrick Moorhead: … coming on Six Five and we hope you’ll come on again. Be a regular.
Susanne Seitinger: I hope so. I hope you’ll have me back.
Patrick Moorhead: Yeah, that’ll be great.
Susanne Seitinger: Love it. Let’s keep going. Thank you so much.
Patrick Moorhead: Thank you. So this is Pat and Dan and Susanne signing off for Six Five on the Road at AWS Reinvent 2022. We’re talking about ANML. If you like what you heard, hit that subscribe button. Have a great day and we’ll see you later and tune into all the other videos we did with all the other awesome people out there. Take care. Have a good one.