Unlocking the Power of Edge AI: Cecilia Dones

Speaker A: Hello and welcome to the Business of Data podcast, brought to you by Corinium Global Intelligence. In this podcast, we talk to senior executives, thought leaders and experts from a range of industries and departments within large and small organizations across the globe. They share stories and experiences that shape their passion for data and analytics and form the future of our industry.

Speaker B: Welcome to the Business of Data podcast. In today's episode, we are discussing innovations in edge AI driven analytics and decision making for data and analytics leaders. I'm, um, really excited to introduce our guest, Cecilia Doneness, marketing executive professor, researcher, uh, and doctoral candidate. Thank you so much for joining me, cecy. I'm really pleased to have you on the podcast today.

Speaker C: Well, thank you so much. I'm so glad to be here.

Speaker B: Great, great. Um, now, before we dive into the topic, I was hoping you could tell our audience a little bit about your background, uh, your journey in the industry, uh, and what kind of gets you excited about, you know, working in the industry these days.

Speaker C: Yes, um, it is a nonlinear and maybe sometimes, depending on the day of the week, it feels like a circuitous journey. Um, the way I describe myself is that I've always been compassionately curious. And what that means is I want to know everything and anything about how people work. And my entire story has been around asking questions about how and why do people do the things they do and utilizing data and all the tools in my toolkit to help tell that story.

Speaker B: Wow, that's. That's fascinating. I. I would like to know that myself, actually. Um, um, so I've been paying attention to your LinkedIn profile of late, and it's largely because you've been, uh, publishing this, um, really interesting series of, uh, essays about edge AI, amongst other, uh, topics as well, uh, which I found really interesting. Uh, it's one of the reasons I wanted to speak to you today. And, you know, as technology, uh, even the technology we have in our pockets, um, it develops, you know, we're hearing more about this, um, A.I. kind of getting closer, uh, to us. Um, I thought, uh, perhaps for our audience of data and analytics leaders, if you could just kind of explain what edge AI is, um, least from the way you see it, and, uh, why it's becoming increasingly relevant in the world of data processing and analytics.

Speaker C: Yeah. Uh, so, first of all, thank you very much. Uh, I do have a LinkedIn newsletter. It's called, uh, what's a CMO to do with AI? And the inspiration for that was actually conversations I would have amongst marketing peers, marketing leaders, CMOs, regarding how do I Apply all of this wonderful technology that enhances the consumer experience and into what I already offer, what I already do and deliver for consumers. And without getting too technical and without getting, um, a little bit too much and overwhelming with all of that, how can I apply and learn along the way? And so that's really the reason why I started the newsletter to write down some of those conversations and the reason why I got really excited about Edge AI is that if we were to take a step back and um, look at the framework of AI technologies, um, oversimplification, um, hashtag warning, oversimplification, um, if I were to create a framework around how to describe AI, we have all the way on the left hand side, narrow AI. So very point solution, very, let's constrain the context of that AI application, help me solve this very specific, specific type of problem. And that's where the majority of all of the research and work and operationalization and uh, commercialization of AI sits today. So it's quite narrow. The things that would get into news where you have thought leaders that are brilliant, um, and can see much more of the future than I can, and I humbly admit I can't really see the future. Um, what they are arguing is that the next phase of all of this is AGI, so artificial general intelligence. And this is where the machines begin to actually really mimic the way that humans think and the way that humans experience the world. And so that is where, um, we start having all of the popular media, oh no, is it going to be like Terminator or is it going to be like the movie her or something like that? And then the third bucket, which I would argue is all the way on the right hand side, is super intelligence or super AI. And so the idea is that yes, we have a way of understanding our world today, and that's very human centric, but we are building machines that it is quite possible. I, um, wouldn't put bets in either direction, but it is possible, um, that these machines will be smarter in a different way, in an alien way from us. And so that's called super intelligence. And many people think about it today. But are we quite, quite close to it? I would argue no. Um, people can debate that as well. So within those three buckets we are firmly within the narrow AI space. So generative AI. Um, what's beautiful about it is that it, like the name suggests, generates things. And so if you are in a creative pursuit or you're in an exploratory, um, mindset and you're trying to figure Things out. That type of AI could be quite helpful. Where edge AI, I think is also quite applicable, especially within the marketing and advertising, uh, discipline is that at the end of the day we have consumers that are setting the expectation of personalization. You should know me, brand, I've exchanged information with you, brand, um, you should understand my preferences, you should understand my wants and needs and interact with me in that way, build a relationship, relationship with me in that way. So that personalization, um, expectation is juxtaposed to another expectation consumers have which is, ooh, we're very, very much into privacy. I want to make sure that you only have the data that is required for me to have this transactional relationship with you, brand or company. I don't need you knowing, um, you know, what I ate for breakfast or uh, you know, um, what the weather is like outside for me today, or uh, any other kind of information that is super to the actual relationship at hand. So we have an expectation of personalization and we also have this expectation of privacy. And so now we have this paradox, these two tensions that are driving the consumer expectation of experience. And this is where edge AI can be quite powerful in terms of an up and coming and almost there type of technology that I would argue data leaders should definitely think about, um, not only from a data architecture and a CDO perspective, but I would argue as um, leaders who are in charge of delivering consumer experiences and really understanding the customer and consumer. At the end of the day the CMOS should be really interested in these technologies. So what is AI? I would not, uh, consider it uh, as part of cloud AI. No, it's not cloud. Um, cloud services light. No, no, no. Um, cloud is really about centralized processing. High latency requires a significant amount of connectivity which um, is great. And for many organizations they've already done the cloud migration and done all of those things. Generative AI, very resource intensive, very focused on creation. Great, fantastic. Um, edge AI is all about personalization at the endpoint. So what does that mean? It means the analytics and the data sits at the edges at the endpoint devices. Why is that fantastic? Well, if you're interested in privacy, you're not sending data into a cloud. You're not doing operations in the cloud. Fantastic. Low latency. Okay. I have resource constraints. I don't have that many cycles to run analytics, so I can't run generative AI models on my endpoint device. Fantastic. Very privacy focused. I already talked about that. And it doesn't necessarily always require connectivity from the Internet. So if you want to deliver hyper Personalized experiences at endpoint devices for consumers and customers. Edge AI, I would argue, as long as the developments in those areas continue to develop in the trajectory that they are, is a potential technology that could really deliver a significant amount of value to consumers.

Speaker B: Yeah, it's really fascinating how far and how fast this kind of technology is developing. Um, you highlighted there that Edge AI improves um, or has the capacity to improve data privacy through on device processing. Um, how does that shift in data localization impact privacy strategies for data leaders? What kind of impact do you think that has when they're designing data architectures for example?

Speaker C: So I think um, the reason why I got really excited about this particular technology is um, what we do know that is uncertain is that as leaders the level of regulation, um, what that actual regulation looks like, we don't actually have all the answers. Uh, these are things that we're all figuring out. And for example in California, Newsom decided to veto a particular um, AI bill. And so we thought it was going to go through. People were beginning to think about how do we prepare for a future in this way. Um, he decided not to. And so in terms of regulation this is an uncertainty. So if I were to take a proactive or maybe an offensive stance in terms of architectures as a cdo, I would say gosh, I want to practice as much data minimization as possible and I want to minimize how much data is moving anywhere um, within my ecosystem, especially data that is obviously regulated and considered private to consumers. And so the beauty of having that hyper localization of not only data storage but data processing at the endpoint device allows us, ah, as CDOs and data leaders to be proactive, uh, regardless of what the regulation looks like, to keep consumers data safe and literally on their own devices.

Speaker B: Yeah, it feels like a bit of a, um, trying to anticipate that regulation and when it's going to come down. It's interesting to hear you say that, um, it seemed to be a surprise. Do you think that um.

Speaker C: Well there are four and anyway it's a constantly moving dynamic.

Speaker B: Right?

Speaker C: Absolutely. Um, so trying to keep up and all of those things. It's exciting, it's good, it's happening, that regulation is happening, but how it actually manifests itself, uh, that is definitely a roller coaster ride and we're all on it, right?

Speaker B: Yeah. And it's a, it's a, it's a guessing game really until it comes through. But um, I think to me it seems that one of the key advantages of Edge AI is kind of Minimizing the um, data transfer to centralized servers. How does this reduction in data transmission help to enhance security? And um, what should analytics leaders focus on to safeguard sensitive data?

Speaker C: So from a data perspective, as long as the data is not going back into the cloud and always staying within the endpoint device, I um, wouldn't be as concerned. From an analytics perspective, I would say, yeah, um, there are trade offs. So while the data doesn't leave the device and the processing happens on the device, you have to figure out, well, when we're thinking about some of these large models or some of these larger applications, what components of that analytics, uh, application should stay in the cloud and be pushed to the endpoint device and how much of that personalization or processing should happen at the endpoint device. So exactly what's the right mix of that hybrid architecture? I think that's something we're going to have to learn into. To be very honest. Um, I'm not seeing too many case studies or too many applications in market yet. Um, but that is something we're going to have to figure out. So it's not an or situation, it is definitely an and situation. So it's not cloud or Endpoint or it's not cloud and edge. It's not cloud or Edge, it is uh, cloud and Edge. It's just figuring out for your organization, for the resources you have, for the value and experience you want to bring consumers and customers, what's the right blend? And then for an analytics perspective, another consideration is, okay, um, if I had a more traditional environment I would worry about, okay, I built a model, it's an analytical product now it's uh, in production. I have to address model drift and I have to address um, any kind of degradation amongst the data set itself and all of those normal model ops or AIOps type of things. When you have Edge devices, um, you have challenges with heterogeneity when it comes to the data itself. Meaning not every endpoint is going to have all the same data. And so, oh no, now I have to figure out how do we deal with potentially different data formats, different data qualities depending on the end user. Um, you're also have to going to deal with, okay, well I'm doing some centralized processing in the cloud cloud and pushing down into the device, still keeping the data safe. Um, but model drift, how do I deal with that when it's happening at the edge and happening in a distributed way? How do I make sure that the fulfillment of the promise of the value to the consumer is consistent enough amongst consumers? Knowing that the analytics will not look exactly the same because some processing and some of the data storage is happening at the endpoint. And, and then also realizing, okay, well there's always the unknown unknowns, any kind of exogenous factors that are always difficult to anticipate. How do you deal with that from an analytics perspective? So in transparency, at least in these early days, I see some challenges trying to figure out the hybrid architecture that works best for the organization maturity and use cases. Um, from a data perspective I think more of the onus is really going to be on the analytics side trying to figure out, okay, we're creating some of this variability at the endpoint that's delivering hyper personalization to the consumer and customer, which is good, but it is additional variability. How do I address that, how do I deal with that? I'm very excited for the upcoming papers, academic papers and case studies and talks about the techniques that will come to address some of that variability.

Speaker B: Yeah, it's fascinating. Um, you mentioned in particular the opportunities for kind of personalization, uh, at the edge. Um, what other opportunities do you see, uh, for organizations to leverage this technology to improve the way that they do analytics, decision making, so on.

Speaker C: Yeah, so I think uh, one of the bigger, uh, one of the, one of the differences I would say between cloud versus edge is cloud. We can definitely do things faster but it's still some, for the most part, some form of batch processing. When you're talking about the edge, you're really trying to be as near time as real time as possible to exactly what's happening at ah, that end point. And so when you have applications that are really sensitive to what's happening in near time and real time, I think that's where you're going to find the kind of use cases, um, at least for right now. And so when I think about things like um, the automotive sector, um, people who are working on autonomous vehicles, it might matter a whole lot what's happening around that autonomous vehicle, um, at that end point in that moment of time. And so from a safety perspective, oh gosh, edge AI probably a good idea. Um, if you are in manufacturing, it doesn't matter what goods you are manufacturing. If you have a factory, um, where you're either utilizing industrial robotics or maybe you're building personal robotics, um, that area of um, of, of industry automation could be a very good place where you need edge AI. So if something's happening on the manufacturing floor and you have people and machines and these things can be quite dangerous for humans at times. If something goes wrong, you probably want Edge AI type of analytics to help support those humans and also support the machines themselves to make sure that um, if there is something going wrong, you can respond very, very quick, uh, to the situation. Um, I would say in the health care sector, this is where I get very excited. So we've already been trained to utilize healthcare devices, personalized devices like Fitbits for example. Um, and so what if those devices got really smart in a safe way, again, making sure that uh, your personal data doesn't go back to the cloud and your personal data stays on your device. But what if, um, we can bring some of these more advanced analytical uh, techniques and technologies onto that edge device such that you get personalized recommendations, you get additional guidance, um, regarding your daily activity, for example. So in healthcare I can see this being quite beneficial and quite therapeutic because again, at the end of the day we're using biosensors to sense ourselves. Um, and so that's another area where I think there could be a lot of value, um, and then where I get very excited because I think this touches uh, marketing land and advertising land. I, uh, think in augmented reality, um, those technologies could definitely be a use case for Edge AI. The reason why I say that is with AR experiences, um, we have issues with uh, because of the hardware and because of the current development of technologies. They're typically low latency type of uh, experiences. And so it would be actually very helpful if you had Edge AI technology so that you are processing sensor data, um, that renders ah, an augmented reality experience locally on devices. And so I think this is like an exciting area of potential immersive, um, communication and immersive consumer experience that could be coming for the marketing space, um, ahead of the kind of virtual reality type of environments. Um, I have yet to see hardware uh, that completely covers your eyes that doesn't eventually give you next rate. Um, so my personal opinion is that between um, virtual reality versus augmented reality, what's going to be more ubiquitous for the consumer experience, what's going to be more ubiquitous? And in terms of scale and viability, I'm arguing that augmented reality is most likely to be first, which will lead us into mixed reality experiences, extended reality experiences, where again, um, uh, virtual reality will eventually be part of it. Um, but it's definitely not at scale today. So I think there's tons of opportunity for Edge AI use cases across different sectors and in different functions.

Speaker B: Yeah, I think some of the examples that you gave there are really fascinating. And it's got so many applications potentially that uh, it's kind of uh, mind boggling. I really appreciate what you said there about VR. Ah, I have some experience with like commercially available VR and yeah, it's amazing for a little while but then you start to feel a bit motion sick. And I think they've got a few things to work out there.

Speaker C: Yeah, um, there's some academic literature and apologies and I should have uh, researched the names appropriately. Um, but there is some early research that indicates that um, virtual reality devices, when you utilize them, actually changes for a period of time. Your depth perception. Once you take the device off and you deal with the pain in your neck and your head is okay, it actually changes how you perceive how near and how far uh, things are. It goes back to baseline. But that change in perception, uh, for some individuals, um, could be quite disorienting. And so that's a potential side effect of utilizing VR devices that at this point in time, obviously we'll figure out how to address those things in the future. But at this point in time could be a limiting factor for some individuals, uh, to choose whether or not to utilize that technology.

Speaker B: Wow. Um, yeah. And one of the things you mentioned in your essay was about behavior analysis. And um, you know, I think that that's something which I found particularly interesting. I was wondering if you could just speak to that a little bit, what you think some of the opportunities are there.

Speaker C: Mhm. So it goes back to the consumer expectations surrounding. Okay, I, yes, brand, yes company, you should know me, you should offer a level of personalization. But I don't need you to know everything about me. I need you to know the me or the Persona of me that I'm presenting to you in a way that makes sense for the state of the relationship we have uh, transactionally today. And so when it comes to the kinds of things that um, would lead to improving those types of experiences, imagine um, with Edge AI devices, so endpoint devices, uh, sensors that we wear on our clothing, on ourselves. You um, could have potential models. And again in a safe way, in a regulatory compliant way, in a consumer provides consent way, you can sense potentially, um, whether or not that consumer is responding to a brand, um, with a first order effect of behavior, but maybe a second order effect, maybe there's an emotional response, um, or maybe a potential psychological response to a brand. And while that may have been more difficult in the past because collecting all this data, obviously there would be concerns around privacy, there would be concerns about consent, et cetera, et cetera. Again, when you're solving for that problem by keeping that information, that data that should be, should belong to the consumer at the endpoint device, you do away with that. So you actually give an opportunity for the actual models themselves to be smarter and figure more things out about the individual. The additional challenge with that is, okay, great, fantastic, we get the analytics smarter. So it starts to nuance the difference between really bored or regular amounts of board for any particular individual. Oh, okay. Now that we can nuance it, you also have to make sure as brands, as organizations, delivering experiences to customers, when we are able to nuance it, can we differentiate the consumer experience at that end point such that we are responding to the nuance in what the analytics is suggesting? So uh, the reason why I say this is it's great to measure things and to analyze things with more granularity, faster, more precise, all the wonderful things we want in analytics. However, if we don't change the way we respond to the consumer or we don't provide enough personalization or variability in that consumer experience, then it begs the question, okay, what's all this precision for? What's all this personalization for if they're all getting the same thing anyway? Um, and so both, um, both capabilities have to grow hand in hand. And sometimes I think, um, organizations can struggle, um, because one capability, uh, is growing faster than the other capability. Um, it's great to keep growing, but if it's too lopsided, then you're still not delivering the full value you can to the consumer and back to the organization. Um, so I would argue if you measure better and you analyze better and you analyze faster, then you also have to deliver faster. You also have to create more bespoke experiences, um, in a meaningful way that's personalized as well.

Speaker B: Yeah, absolutely. So what do you think? What's the state of um, you know, edge, uh, sort of AI at the edge now, uh, in terms of the extent to which it's being used, um, you know, um, in industry and enterprises. And how do you see that, um, changing in the, in the near future?

Speaker C: Yeah, um, as I alluded to earlier, I do not have a crystal ball. Uh, so I, I do not purport to be able to predict, um, the future, uh, with high levels of certainty. I can be confident because I'm pretty confident in my own skills, in my own inference skills. Um, we'll go with, I don't know about accuracy of that. Um, so I feel that um, edge AI is actually a, ah, natural evolution of our AI journey as we're developing these technologies. Meaning, um, we have IoT devices already in place. We already have like smart fridge and smart uh, washing machine and smart homes and et cetera, et cetera. So these endpoint devices already exist um, in an ecosystem and we already have um, AI devices. So the next way with agentic AI coming along after um, generative AI as additional capabilities. So I feel that um, EDGE AI is just a natural extension of where AI is going in terms of additional capabilities and the hardware infrastructure that we already have in place. So when we think about people who are worried about building the future cities for example, where you have autonomous cars, um, or people who are trying to move into solution, um, uh, solutions for um, uh climate change and how we address all of those resource constraints, ah, related to all of that and uncertainty related to all of that. People who are worried about the power grid. This is again a very natural extension of the technologies that will help us to solve some of those problems. When we think about industry, um, 4.0, um, and all of the work, um, that the uh, World Economic, um, um, Federation and all of the those types of organizations are worried about the future of labor, um, again this is another extension of uh, where EDGE AI is not um, it's not trying to um, it's not orthogonally disrupting the trajectory of development of technologies. It's actually somewhat along the path. And so from that perspective I'm comforted that we're on a trajectory. And so while I cannot predict the future, um, this is a uh, piece or a step along that path. I'm not too, too concerned from that perspective. If I were a sitting CDO inside of ah, a Fortune 50 or Fortune 100 organization and I'm thinking about okay, we went through this generative a high situation. Some people would argue it was hype, some people would argue no it's really changing the way we're doing things. Okay, great. All right. We're seeing advances from an analytics perspective, in a hardware perspective. Okay. Agentic AI will be the next thing. So now we have people who are um, very much interested in how much automation, how much uh, uh, goal focused AI can we have as opposed to something more narrow which is what exists today. Okay, great. And do I need to invest in that or do I need to invest in edge? The way I would address this if I were sitting in those shoes, um, I would look at the level of exposure I have to endpoint devices, the endpoint experience. So if you are a, if you're in sitting in one of these very, very large organizations and you're very worried about localized energy. Okay, so maybe technologies like this and being aware of how it continues to develop and becomes more commercialized might be an area that you are interested in. If you are in robotics, manufacturing, anything that has to do with weather, um, and disaster response, I would be very much interested in learning about these technologies and making sure that my organization is ready for. Doesn't have to do it today because there's lots of priorities, but is ready for, is along the path of our AI strategy, is along the path of our, ah, AI trajectory.

Speaker B: That's fantastic. And, um, I think I might have to get you back for another conversation or some other kind of, uh, media talking about, uh, the potential impacts of EDGE AI on, uh, environmental issues and on the power consumption, all that kind of stuff. But that may be a topic for another time. Uh, well, thank you so much for sharing your insights with us today, Ceci. It's been a fascinating conversation, um, to, uh, our listeners. We hope you enjoyed the conversation today. Remember to check out, uh, Ceci's LinkedIn for more articles and thought leadership on this. Uh, and as well as that, uh, you can also check out her substack, it's called Authentic Interactions, uh, for all kinds of musings around AI, ah, and her adventures, uh, with the Internet and other such topics. Thank you so much for joining me, uh, Cecy, I've been so pleased to speak to you today.

Speaker C: All right, thank you so much.

Speaker B: Well, that's all for today, so be sure to join us next time on the Business of Data podcast for more discussions about shaping the future of data and analytics.

Speaker A: We hope you enjoyed the episode. Be sure to subscribe to the Business of Data podcast wherever you're currently listening. And keep up with us on socials and for more content, visit us on, um, businessofdata. Com. Be well, and thank you for listening.