Interested in connecting with Eric Daimler? Reach out to him on Linkedin!

About Conexus: Conexus was founded to deal with one of the biggest problems plaguing the majority of businesses today — data deluge. Every business is now a data-driven business but they are few means to manage data efficiently with minimal time and cost.

The Conexus solution uses new math developed at MIT to create new algorithms that establish relationships among large, disparate sets of data resulting in seamless data integration and interoperability which is accomplished in a short time period at a mere fraction of the cost of today’s cumbersome, manual integration projects that can take years and waste billions of dollars.

Key Questions and Topics from this Episode:

(00:54) Intro to Eric Daimler

(04:47) Intro to Conexus

(06:16) What types of Applications have Conexus been involved in?

(10:14) What is an AI system?

(14:13) What are the components of an AI system

(18:57) What is the industry and customer focus at Conexus

(21:11) What challenges did you experience going to market?

(23:30) What market trends have you seen in AI?

(27:09) What are data lakes?

(30:12) What’s the best first step for companies to utilize their existing data?

(33:42) How will AI affect the workforce?

Transcript:

 - [Announcer] You are listening to the IoT for All Media Network.

- [Ryan] Hello everyone, and welcome to another episode of the IoT for All podcast on the IoT for All Media Network. I'm your host, Ryan Chacon, one of the co-creators of IoT for All. Now, before we jump into this episode, please don't forget to subscribe on your favorite podcast platform or join our newsletter at iotforall.com/newsletter to catch all the newest episodes as soon as they come out. But before we get started, does your business waste hours searching for assets like equipment or vehicles and pay full-time employees just to manually enter location and status data? You can get real-time location status updates for assets indoors and outdoors at the lowest cost possible with Leverege end-to-end IoT solutions. To learn more, go to iotchangeseverything.com, that's iotchangeseverything.com. So without further ado, please enjoy this episode of the IoT for All podcasts. Welcome Eric to the IoT for All Show. How's your week going so far?

- [Eric] It's going well, thank you. Good to be here.

- [Ryan] Yeah, it's great to have you. So let's start this off by having you give a quick introduction about yourself to our audience, any background information, anything exciting about your kind of past history in the space that would be relevant for audience?

- [Eric] Sure, I guess I'm most often known as serving as an AI authority in the last year of the Obama administration. I've worked with the Science Advisory Group which I'm excited to know is now elevated to a cabinet level position under the Biden administration. It's exciting place to be then, and I'm sure it's an exciting place to be now. I am fortunate to have spent my entire career, 20 plus years in various capacities around what is AI as a venture capitalist on Sand Hill Road for a time. I've started several companies. And in addition to working in Washington, I started as a researcher with the University of Washington, Seattle, at Stanford University, then at Carnegie Mellon University where I was faculty after getting my PhD in computer science. So I've seen this from a lot of different perspectives over many years, and that's the perspective I bring.

- [Ryan] That's awesome, that's very exciting stuff. We don't have too many people that have the connection to the government, like experience in the government, like you do. Can you tell me a little bit more about that, what the objective in the role there was and kind of just the overall like purpose of the position?

- [Eric] Sure, the Science Advisory Group is a part of the Executive Office of the President adjacent to the west wing, which is really the, you know, the Hollywood narrative, the very big building next to it has the National Security Council, the Domestic Policy Council, it is the Office of the First Lady, the Office of the Vice President, the Council of Economic Advisors, and then the Office of Personnel Management maybe comprises of the biggest in terms of square footage. This is all within the security perimeter. It also holds the Science Advisory Group. They started under Nixon to provide science advice on everything from space to water, to issues in life sciences and then a growing influence around information technology. So there was a chief technology officer as a new role established under the Obama administration and then a chief data scientist as a new role started under the Obama administration. I was just happy to for my period of time be the authority on AI, the authorities in computer security and mine was around AI and automation and software and hardware, so it touched robotics. The particular job was speaking modestly on behalf of the president to coordinate the executive. So this is really a bipartisan issue. It's obstensively a nonpartisan role where everyone would want our tax dollars to be spent effectively and all of the million or so people that work across the U.S. government to coordinate on their initiatives in research, in deployment. So from the state department to defense of course, to health and human neuro services and transportation, how do they think about AI? How do they think about robotics? How do they think about their data collection and its use? Where do we need to do more research, where there's security vulnerabilities and so forth. So the president has a view and it's our job to communicate that out.

- [Ryan] That sounds absolutely fascinating and probably a very exciting environment to be in. That's very cool. So tell me a little bit more about Connexus, your company now. Tell us obviously what you all do, the focus there, and then also I'd love to hear a little bit about background on the founding of the company, kind of the story behind it and the problem or opportunity you saw in the market to obviously warrant the company being started.

- [Eric] Sure, yeah, Connexus is one of the more rare stories you'll hear this year. And it's because the discovery actually is in math to tell the, to hear that the MIT say, Massachusetts Institute of Technology, the institute would say that this is the first ever spin out from their math department. That's, right, that's quite a statement. So even though I spent time as a computer scientist and even though I spent time in AI, this is upstream from that. It's even more fundamental from that. There's a blockage around fully taking advantage of AI and that is in the collection and integration of data. Structured data, unstructured data, just all types of data. And this was born of that need.

- [Ryan] Oh, okay, gotcha, very cool. So talk a little bit more to us about the role you all play in the market, Applications, anything that kind of brings us full circle to some real life examples of solutions that you you've been a part of developing and how your technology and your offering kind of plays a role in the market.

- [Eric] At an abstract level, you might think that a company with an infinite balance sheet, it's a dream and very smart people from any number of places might, and the abstract create an optimal IT infrastructure. This might be what some of the listeners can imagine, or what they go for, what their customers might think they go for. And that's just not how reality plays out. Because the reality is companies are looking about to implement what's needed for their business at any particular time. Nobody can really see years out. They don't know how their company is necessarily gonna grow. So in one particular instance, we worked with a large ride sharing company. This ride sharing company has an effectively infinite balance sheet and some extremely smart people. And yet, they do not have at all an idealized IT infrastructure. That IT infrastructure in their particular case grew up a jurisdiction by jurisdiction. Anybody else can think about this by department by department or by customer need, this was by city. So this particular ride sharing company would look at the city of Los Angeles and want to say, well, I wanna do a business case for the rates versus writer satisfaction, or the weather versus drivers satisfaction, whatever sort of business analysis that one's AI algorithm might engender. But in order for them to do that, they could do the city of Los Angeles and then they would have to do a statistical comparison to the city of San Diego. They couldn't do the whole state of California, let alone all of North America or all of the world. This impaired their business decisions. It also slowed their business decisions. So they implemented our solution looking really around the world, how does this get solved? How do we integrate these hundreds of databases all into one to make better business decisions that are faster, more accurate and, you know, really represent the responsiveness that we seek. And the answer was that they had to look more fundamentally than the computer science solution. So some of your listeners might be familiar with RDF and OWL, those were solutions in the early 2000s to try to deal with this day-to-day louche that we often read about in the press, the doubling of data every very shockingly short period of time. That continues but it's been a failure because it really doesn't scale. It just fails to keep up with the size of databases today. So this is gonna be flipped around the world. They found us and to have them tell it, we were the only solution available to have them integrate the totality of their data because we were fortunate to be recognized as the leaders in enterprise software taking advantage of this particular discovery in math.

- [Ryan] Fantastic, yeah, I mean, the Applications along with the story of coming out of MIT as the first company from the math side is absolutely fascinating. I don't think unless I was having a conversation with you today, I would have known that or even a guest that otherwise. You just would assume with this, the high profile MIT that other businesses have been built out of the math side. So that's really exciting, and congrats on all the success you guys have had so far.

- Thanks.

- [Ryan] So I wanted to kind of bring this a little more high level here for a second and see if you could talk to our audience a bit about an AI system in general. Kind of what an AI system, what it means to have be, or have an AI system, how an AI system works, just less technical and more kind of on the layman terms of just kind of what that is when people hear about it and why it's important.

- [Eric] You know, I think it's really terrific to delve into that question, thanks for asking it. It's important that all of us get involved in this conversation around AI and not just think it's for people like me that grew up in a basement in front of computers, you know, it's for everybody. And the term is a funny one because it's begun to emerge that people think it doesn't really exist, there is no really artificial intelligence, and I find it that to be confusing, what we used to call generalized AI versus specialized AI. But these sorts of distinctions, they really don't matter to the 99% of us that are not AI researchers day to day. I find that these definitions are more helpful if we just look at what's useful for somebody's day-to-day existence. And I like to look at the very beginning. So how is the data collected? How is it sensed? This show is about Internet of Things. This is data from those Internet of Things. You know, my dishwasher somehow is connected to the internet for some reason so it collects data, it's for maintenance and that's a collection of data from different sources. It could be the LIDAR on top of the car, it could be the air quality in the rooms we're in, that's called the collection of data, sensing data. And then that data is transferred to thinking, planning, cognition, we might say. That's where traditional AI would live and if we really wanted to get pedantic, we could say that the popular term of machine learning is a subset of AI and the popular term of deep learning is a subset of machine learning, so there are non machine learning AIs, but that is not useful for the 99% of us that are not AI researchers. And we can expand that further, we have the sensing, we have the planning, but we then have to act. We have to do something with the data. So if my automated car gets together, collects data from the LIDAR, then processes that in the totality of its system inside the car, It then has to decide if what is in front of it is a shadow, is a crosswalk, is a person, and what the actions are, if it should slow down, if it should wait until it collects more information, or if it should just continue at the same speed for whatever reason, that's an action. So you have sense, plan, act, and then what's unique about AI, so it's not just the thermometer from the 1970s, it learns from the experience, learns from the experience. So it will start to learn. Now, AI systems don't spontaneously learn. And this is almost a misnomer or misunderstanding about what the learning is. My automated car is not gonna go down the street at an intersection, see that crosswalk for the thousandth time and suddenly start spouting Mandarin. It doesn't learn Mandarin, right? It's funny when I say it that way, it's so obvious but people get this misinterpreted. It becomes a better driver on that intersection, on that intersection. Now, the learnings might be then fed back into the manufacturer and the manufacturer could use that as learning to increase the efficiency of the algorithms in general, but the car itself is only gonna learn that intersection. So sense, plan, act, learn from the experience, that I think is a useful definition for nearly all of your listeners.

- [Ryan] That's fantastic. And when it comes to all the different components and pieces that make up an AI system, which ones would you say are the most important and kind of that really drive the success of an AI system when it's deployed?

- [Eric] When I was in computational linguistics for a number of years and I found that there was a certain dynamic that plays out more broadly in information technology in general. And it's this idea of waves of tech and then waves of application of tech. So we would learn the new machine learning techniques and then we'd apply them to language and we get all really excited and have some breakthroughs and then we'd reach the limit of that and then have to go back to researching language again, semantics, semiotics, syntax, what have you, and then a new technology wave would come along and we would apply that again. I find the same with AI and maybe even enterprise software where if you look at the system that is AI that we just distinguished sense, plan, act, we might say that many of the technologies that we now live with deep in the infrastructure of our world, products from companies like Oracle and Microsoft and even Salesforce to some extent, those were really innovative products 20 years ago, 30 years ago, 40 years ago, right? But the innovation has been felt instead on the consumer side where that used to be the laggard but now, we have some fantastic consumer oriented technology devices, but now enterprise software is really lagging. So I might say that whole system is just up for some new developments that are terribly exciting. In particular, as we started this conversation, look before the learning algorithms. So there's a lot of seemingly magical items coming out of deep learning in the case of DeepMind's protein unfolding, they deserve all the praise that you see in the popular press. One of the companies I sit on the board of, Petuum, they have a fantastic suite of AI products that can democratize its application across industries. But for many venture capitalists, they're gonna be disappointed with the returns in that domain because the bottleneck is upstream from that. There's a lot of data collected that's not used, dark data.

- Interesting.

- [Eric] There's a lot of data that exists in these databases that's not adequately brought together. There's the refactoring of databases, keeping the semantics while the structure changes is where careers go to die. The years are invested, hundreds of millions of dollars, it's really, it's where the calcification of old industries begins to show up, and that's what I'd say is an exciting place to look. This is deep in the plumbing of our world and how it operates. It impairs some of our logistics systems. And maybe I can give a story about that. During the the early stages of the COVID crisis, we were working with a logistics company. You read in the news recently about that large container ship that went, it's going say sideways, it's going diagonal in the Suez Canal. Well, those ships you read about have thousands or even tens of thousands of containers on them and there are thousands of these ships around the world. And it was hard for us to imagine that when one of our clients or rather one of our clients' clients, each one of which has tens of thousands of employees and thousands of ships with those tens of thousands of containers, they don't know immediately where their stuff is. They wanna ask a business question, where's my stuff? And so in the early days of the COVID crisis, where's the personal protective equipment? Where's my PPE? And should it go to San Diego or Santiago, but I don't even know where it is. That business question ideally would be answered, again ideally, in a matter of hours, instead it takes days. That's an expensive question to ask in other words. So we help solve that problem, integrating the databases to make those business decisions faster, cheaper, and allow the business to become more flexible as another example of bringing data together. That's an exciting place to look.

- [Ryan] Yeah, let me ask then from a listener standpoint, what type of company, individual, organization, industry do you all kind of focus on, if any, and at the same time, who is out there listening should makes the most sense to kind of reach out and engage you if they're interested in kind of just getting a better sense of the work you do and how your offering could work with what they're doing, like do they need to be in a certain stage of development? Do they need to be focused on a certain kind of problem? How does that usually handle it on your end, like that typical customer engagement and just any advice out there for people that are very intrigued obviously by this conversation and wanna kind of connect in that way?

- [Eric] Well, for us as a firm, we generally work with larger firms with complex data infrastructure, the ones that have thousands of tables, tens of thousands of tables, sometimes it's measured in size of the database, petabytes of data, but it's really the complexity of the data set. You might say it's physically heterogeneous but the data models are homogeneous. That's a terrific scenario in which we can provide the quickest value. Specifically in Microsoft SQL environments, that's a good place to be looking to the quickest return. I can offer in general the guidance for people around the foundational technology for your listeners to just listen for in other environments which is the discovery of category theory. So categorical mathematics is the branch of mathematics that powers what we do. We're not the only company that will be doing this, we're not the only company today that does it but we're just the leaders in enterprise software taking advantage of that. But category theory as a branch of math is the math of the 21st century. That's where I suggest to people look in general for themselves, for their companies, for their children. You might say the more math, the better. But in 10 or 20 years, calculus, geometry, trigonometry, that's gonna look a little bit like speaking Latin to each other.

- [Ryan] Gotcha, okay, interesting. Now, I wanna take, kind of shift topics here for a second and talk a little bit more about when you all were coming to market and kind of forming the company, what were some of the biggest challenges that you all saw in kind of promoting and educating the market on the value of what is that you're doing? Are there anything that you say stands out to you that was a relatively noticeable challenge and kind of how you guys overcame that?

- [Eric] Yeah, the problem for us is we compete against not doing and we compete against manual. So if you look at the revenue of some large consulting firms and not, I'm not just talking about Deloitte, Accenture, McKinsey, what have you, but you look at Tata, Wipro, TIPCO, those companies have grown their revenue right in line with the rate of data growth. Those companies exist because of this problem in data interoperability, because that process is horribly manual using tools like Informatica or Ab Initio which are themselves are good companies, but it's a terribly manual process that seeks to be automated. But if a company has grown up with this business process of just putting a decision through even a manual operations, shifting out of that can be, you know not worth the effort, not worth the risk, so that was a challenge. Even the canonical 10 times benefit or even a 100 times benefit can be a misleading metric if what you're competing against is a given business process. And you can know this intellectually, I knew this intellectually, I've taught this for goodness sake. But coming upon it firsthand, it's funny to see in action where just the nature of large organizations is no one person is in a position to disrupt the business process, that no one wants to change their workflow, that's a lesson to learn.

- [Ryan] I understand that completely, absolutely. That's great, that's great. I wanted as we're kind of talking about those challenges, I guess with obviously your extensive experience and knowledge of the market for many years, what are you seeing as some of the biggest trends in the market right now as it relates to AI and even IoT to some matter?

- [Eric] IoT I think is being misinterpreted in its impact. It's not about collecting more data by itself that is the value, it's the relationships between the data. So it's not, we worked with this utility company in Europe that had wind mills. I came to find out there are five large windmill companies in the world. They're more but there's five companies comprise 90 plus percent of the market for wind mills. Now those wind mills for you and I are interesting only to the extent that they supply power. They're green power, that's fine. But we don't care about the nuances but there are sensors on those windmills that will detect the temperature. It's four degrees centigrade in Europe. It's 40 knot winds expected, that's data, right? So that's data, we're sensing, sensing, planning, acting, learning from the experience. So we're sensing the data, collecting the data, that's nice, but what's much, much more valuable and what's unique to a company, this is where every company is becoming a data company, is the unique knowledge that a company would acquire when it is four degrees Celsius and there was predicted to be 40 knot winds, then something happens. Something could be, there's a chance of freezing, there is increased efficiency on the windmill, there is a shorter maintenance when you, what happened, whatever, that's knowledge, that's valuable. The data collection much, much less valuable. It's the knowledge from the relationships of the data, the sharing of knowledge, that's the future. And then going one step further, it's the patterns in the relationships of the knowledge, those are the people that are gonna own the future.

- [Ryan] That makes a lot of sense. I mean, we talk a lot on here about the ability for companies to collect data in ways that they have not been able to collect data before due to the advancement of technology as it relates to IoT. But it's interesting your approach because, and which makes a ton of sense which is the data collection is becoming a bit more trivial now and it's actually what you do with that data, how you interpret the data, the decisions you make on that data, basically that knowledge like you're talking about is what dictates the value and the success of the entire process to collect the data. And it's very, that's, a great way to kind of explain it, what you're saying because it's hasn't really been explained that way to our audience I don't think as of late.

- [Eric] Yeah, the models, the schemas, that's where every company is gonna derive value. I say that every company is an AI company because a lot of other things are just becoming commoditized. Whether you're a public utility or you're a mining company or a sort of shipping or logistics company, many of the physical goods are becoming commoditized. And like we're saying, the collection of that data can become commoditized, but the relationships, the knowledge that's becoming the goal.

- [Ryan] Gotcha, one question I would ask is, it's not really related but it's just a term I hear often and I'm not sure if our audience fully understands it 'cause we don't cover it very much. But data lakes, can you talk a little bit more about what data lakes are and kind of what their purpose is and kind of your overall just view of them?

- [Eric] Yeah, data lakes were trying to compensate for the failures of data warehouses. Now, a lot of these terms I don't like because just like we started off talking about AI, the terms themselves can have specific meanings but it's the interpretations of the terms that really matter. So I will say in data lakes what it's come to mean is magic. I'll bring all my data together in one place. So there was an example earlier in the COVID crisis where a c3.ai got a lot of press for creating a data lake about a COVID information. But really what they did is they just they brought everything together essentially on one hard drive.

- [Ryan] All right, okay.

- It's not, it's doesn't make your analysis necessarily any more useful or available, I guess I'll say that, it doesn't make it any more available. Because when one is trying to do an analysis, make better business decisions, all of your data needs to be integrated. There is no AI without integrated data. I'll just say that again, there is no AI without integrated data. You have to have fully integrated data. If you have data that you collected but it's just somehow sitting around, that's dark data. That's bad, right? So the data lake was an attempt to first collect all the data essentially in one big hard drive, one big lake and then we'll worry about the integration later. We'll let the data scientists do their integration later. But that is the proverbial kicking the ball down the road to let the data scientists do the work. But the hard part isn't necessarily using these increasingly sophisticated off the shelf learning algorithms, it's actually collecting and integrating the data. That's the hard part, that's the manual part, that's the part that takes years. It takes a commitment from senior management and unfortunately, today it takes the manual effort of many of these consulting firms who are costing tens of millions of dollar