Artificial Intelligence is surrounded by marketing hype, making it difficult to assess what's real and useful. In this episode, we talk with a venture capital investor and two software entrepreneurs to learn what's involved with creating products that rely on artificial intelligence and machine learning. Join us as we cut through the hype of AI.
Ed Sim is the Founding Partner of Boldstart Ventures. He was an early believer in SaaS and led first round investments in market leaders like LivePerson (Nasdaq: LPSN) and GoToMeeting (acq. by Citrix). Over 19 yrs, he has led many seed and first rounds and helped a number of entrepreneurs successfully scale from seed to market leader.
Sean Chou is the CEO of Catalytic, which makes the Pushbot platform. Prior to Catalytic, Chou was the Chief Technology Officer and EVP of Services at Fieldglass. Chou was responsible for the overall development and delivery of Fieldglass solution and services, oversaw the product development, hosting, professional services and marketing departments. He provided the strategy and vision for Fieldglass’ award-winning cloud solution since inception.
Keith Brisson is CEO and co-founder of Init.ai, a venture-backed developer platform that enables companies to create conversational apps. Keith is a software engineer and has extensive experience building conversational applications using modern machine learning techniques. He focuses on and writes frequently about the end-user experience of conversational applications, the machine learning and technology that powers them, and the future of human-computer interaction.
Video Transcript: Building AI: A Software Maker's Perspective
Michael Krigsman: Welcome to Episode #222 of CxOTalk. I’m Michael Krigsman, and we have a really interesting show. We’re going to be talking about artificial intelligence, machine learning, and natural language processing. And, we’re going to be taking a development perspective, from the point of view of people who are actually creating products. And we have with us today two company CEOs, as well as a venture capitalist and investor. So, let’s dive in and Ed Sim, you are Guest Number One, or at least, I am introducing you first. How are you, thanks for being here, and tell us about yourself.
Ed Sim: I'm good! Thanks for having me! I really love your show, and I'm glad to be here with two of the founders in the Boldstart fund. So just real quickly, I'm a founder and partner of Boldstart Ventures. We like to be the first check in enterprise founders with respect to AI and ML, etc. We go for the opportunity to look for AI and ML companies. We look for enterprise businesses solving big problems, and if they happen to use AI or ML, like Keith and Sean are doing, then that's even more exciting for us. So, looking forward to chatting about this.
Michael Krigsman: Fantastic! And, you have been an enterprise investor for a long time.
Ed Sim: Yes, yes. Over twenty years done stuff in the back-end with the SaaS side, and companies like Greenplum, GoToMeeting, LivePerson, along with Init.ai and Catalytic.
Michael Krigsman: Fantastic. Our second guest is Keith Brisson, who is a founder of the company Init.ai. Hey Keith, how are you?
Keith Brisson: Excellent, Michael! Thanks for having me!
Michael Krigsman: So Keith, tell us about Init.ai.
Keith Brisson: Sure. Init.ai provides technologies that help enterprises converse with consumers at scale. So, we provide language understanding technology that helps them automate customer conversations, assist sales, and support agents in live chat, and analyze conversations that take place between their users and their agents. These can be external-facing conversations. With consumers, it can be internal as well, allowing employees to access information from systems of records, CRMs, clients, etc.
Michael Krigsman: So, essentially, would it be accurate to say you’re building a natural language processing/AI - hate that term, AI …
Keith Brisson: [Laughter]
Michael Krigsman: … toolkit that enables large companies to build their own products?
Keith Brisson: Exactly! So, we’re providing the technology that lets it unlock the data and the information within those conversations so they can incorporate them into their workflows, whether it’s communicating with their customers, or making their internal processes more efficient, it’s our belief that we should be the ones to take that advanced technology and bring it to them rather than them needing to do it internally.
Michael Krigsman: Fantastic. Well, I’m looking forward to diving in. And, Sean Chou, you are the third member of our behind-the-scenes in AI panel, and you are the founder and CEO of Catalytic.
Sean Chou: Yup. Thank you for having me on. So, Catalytic: We create a product called “Pushbot,” and our customers using our product can quickly and simply create these process bots that leverage work orchestration, automation, and AI; which I agree with you by the way, as a term, it’s definitely a little bit irritating, but now I think it’s popular and it captures a whole broad range of technology which I’m going to talk a little bit more about. But, with our product, our customers really focus on operational efficiency, reducing the number of dropped balls they might have, and tying together a lot of the people and the different systems within their company.
Michael Krigsman: So, you mentioned the term, "process-bot." What is a process-bot?
Sean Chou: Yeah. As entrepreneurs, we feel obligated to make up terms.
Michael Krigsman: [Laughter]
Sean Chou: [...]
Michael Krigsman: Now that's good, and large companies do the same thing. But seriously, when you say "process-bot," tell us what you mean by that?
Sean Chou: Uhh, so, I say mostly to separate us out from, let’s say, chatbots, which are a really popular concept. We do have conversational interface aspect[s] to our platform, so in that regard, it’s kind of got some chatbots. But, that’s not really the main purpose of what we do. The main purpose of what we do is really; and the main purpose of what Pushbot does, as a bot; is to push and promote a process. So, we like to say, “Processbot,” most[ly] in a way to kind of constrain down what we’re doing.
Michael Krigsman: I think a good place to kick off this discussion is all three of you are very actively involved in the consideration of different types of AI and what does it mean. And, Sean, maybe you can begin by helping us understand what do we mean by the term, “AI,” and what does it actually encompass?
Sean Chou: Yeah. For sure. It certainly covers a lot of different things, and I think with all major new technologies, there’s always this retrospective period where people look back a little bit and say, “Hey. This really looks like it should be under this umbrella,” and you get a lot of repackaging of things that once maybe weren’t part of AI, but now, because it’s the hot, buzzy topic, now get rebranded under AI. But, I think generally, when we think about AI, we think about it in three different categories.
There's really a strong AI, which is to try to create basically machines that are able to think in a general sense, in the same way that you and I are able to think. So, "strong" or "general AI," there are only a handful of companies that really should be considering that. You need a ton of resources; it's the Google, Microsoft, Amazon, you know, of the world that are going to sort of win in that type of space.
The second category is really more "weak AI," or "narrow AI." And, that's not as difficult. It's still extremely hard, but now what you've done is you've set instead of a general, thinking machine, we're going to focus on a specific domain or a specific field. And so, you see a lot of that in virtual assistants like Siri or maybe Ingram, or Clara, you know; these are folks who are saying, "We're going to create AI," and its personality oftentimes, but it's only going to solve a very narrow set of problems.
And then the third category, which I believe Init.ai and I both fall into, which is: The users of the technology and the research has come out of all this primary research on AI. So, we are beneficiaries of the research that’s gone into natural language processes, sentiment analysis, machine learning; all the things that kind of power AI, we take them, we repackage it, we make it, at least in catalytic space, we make it available for the average business so that they’re able to use it in their processes. And then we’re using it for our product in a very, very applied setting. So, you know, it’s not machine learning to be able to act as humans, machine learning will figure out how to improve their business processes.
Ed Sim: Yeah, and Sean, I think that’s a great point. Kind of how we look at the world is applied AI. I mean, AI is such a buzzy word these days. Everyone has AI in their business plan, the kind of time that everyone had “.com” back in the day. And the reality of it is, what business problem are you solving? And applied AI is very exciting because you’re not going to out-Google Google in AI learning, or Facebook with that AI, but how do you leverage best what’s out there and then apply to enterprise data? That’s the data that they don’t have and can get, and that’s what I love about what you guys are doing and what other companies are doing as well is working on that private enterprise [...], and learning from that.
Michael Krigsman: What are some of the really interesting use cases for this type of applied AI that any of the three of you are seeing?
Keith Brisson: I can give a few examples. You know, I mentioned we're focused on making customer conversations more efficient. You know, right now, companies need to hire teams of agents to serve their consumers and support in sales areas. And, it's our belief that we can take technology that happens to be using machine learning to make that process more efficient by offloading some of the routing portions of that to the computer so that the agent doesn't have to go through those routine processes themselves, you know.
So that is, for us, is helping displace the actions that people would not take otherwise. And, kind of a more broad scope, machine learning is incredibly powerful in finding patterns in data. So, you see it in all areas of enterprise uses where you're trying to detect signals in large volumes of data that matter to the business.
In finance, it has been used for fraud detection for ages, where you’re trying to find something that really matters to the business in this massive set of data. And, that’s just going to expand where you have more and more data coming in from different sources. You need to figure out how to extract real value from that unstructured data, make your business more efficient.
Sean Chou: Yeah, for sure. You know, it’s interesting, and I don’t think it’s a coincidence that AI is the trend right after kind of data science. Like, remember … This or a year ago, we’d all be saying, “data science.” But, I think there’s a reason for that. It’s because data science ended up producing so much data and kind of cheap storage, cheap processing power; just created so much data, that data science came in and said, “How do we make sense of this data?” But, we’ve actually reached a point where people have thrown up their hands and said, “We can’t make sense of this data,” and so we really need other things to make sense of this data, and those other things ends up being AI and that actually ends up being a great initial use case for AI, which is definitely one of the areas in which … I think also IT.
The notion of AI being used for human augmentation, as opposed to trying to replace humans, is, I think, still the right current curve for AI. I mean, we both … It’s entirely human augmentation: How do we take processes that have people and make people all more effective?
Ed Sim: Yeah. That’s a great, great point. I mean, I invested in GreenPlum back in the day solving the big data problem. Now, it’s part of the EMC and Pivotal. And, we coined the term, “smart data” five or six years ago, meaning you’ve got all this data … I mean, there’s three certainties in life. It’s taxes, death, and the growth of data. [Laughter]
So, when you go back to the applied AI and learning, that's kind of the next phase with that and the GPUs, and great neural network models, and cloud. I think it makes it more accessible to all of us from that perspective. So, when you look at any manual in the process, you know, including security: We invested in a company called Security Scorecard that allows you to look at the security posture of a company from the outside in. Using tons of data, we give classification, clustering, we give use regression analysis, and all that stuff to come up with scores on security. So, it can be anything from internal data entry, business process, but also externally you can use it for security as well. So, it's a pretty broad field.
Michael Krigsman: So, there are so many different areas and applications where patterns can emerge and technology can help discern those patterns. From a businessperson’s standpoint, how can they go about figuring for their own business, and their own processes, where it makes the most sense to apply these types of machine-intelligent technologies?
Sean Chou: Yeah. I think this is an area where, “You look really young, but I’m not going to make a call on your age.” But, certainly like Ed and I, being old guys in technology, I think we would say if you’re looking at this from a business user perspective, AI is just a shiny object, you know? If you’re looking at how to assess AI companies, I mean, stay focused on the business problem and realize that AI is a solution. I mean, when we set out to make Catalytic, we didn’t say, “Hey, we’re going to create an AI company.” We said, “Hey, we think the problem is that these processes are horrible, awful, and in spite of the technological advances that we have, they’re still horrible and awful, and we need to do something about that.” And sometimes, AI is the right answer. And sometimes, it’s not. So, we have AI for some things, but it’s not; certainly our solution doesn’t revolve around or hang its hat entirely around AI.
And I would say for the businessperson, they should have those things - that same mindset: What’s the problem; what value is going to be created; AI is just, you know, a new big, shiny tool that is hard to ignore.
The one caveat I’d put to that is, AI is … It opens the door to new category or problems that previously would have been almost intractable feeling. Just like trains and automobiles in the industrial age, now allow for low-cost nationwide distribution; AI is going to open up a new category of problems that we’re able to tackle that previously were just like, “We just can’t tackle those issues; [they’re] you know, unaddressable.” So, I think you’ll see a lot of those.
Michael Krigsman: But there’s a point at which I am now confused. So, on the one hand, you mention that AI is just a shiny object. And by the way, in doing that, you have just taken the wind out of the sails of half the technology marketing departments in the country, [and] in the world, okay?
Ed Sim & Keith Brisson: [Laughter]
Sean Chou: I’m going to be getting a lot hate on Twitter; a lot of hate tweets.
Michael Krigsman: But at the same time, you said that AI creates a whole new set of problems that can be solved or types of solutions to problems. And so, how do you reconcile the fact that on the one hand, it's just technology, and on the other hand, the implications are so profound?
Sean Chou: So, if you have just an AI solution out there, and it’s not solving anything, it’s just a shiny object that has no meaning. But, if there’s a new problem that you’re now able to solve as a result of AI, then to me, it’s kind of like what it does is it broadens the scope of types of things that technology can solve. So, in that sense, it is remarkable. But, at the end of the day, it still has to be solving the business problem; and I think one of the big dangers, whenever you have [...] things, is that people just start saying, “We’re an AI company,” but they ignore the business problem that they need to tackle. So, I guess I’m trying to speak out of both sides of my mouth here, but I think there’s truth to this; because on one hand, AI does now allow for solving things that we just would not have thought worth solving; when in fact, going back to what I was saying about it being a response to big data - like, big data: designers felt like we had reached a point where we just didn’t even know what to do with all of this data. So AI actually opens the door to being able to really significantly, meaningfully address terabytes of data that are coming in on a real-time basis. But, you know if you’re not doing that for a purpose, it’s pointless.
Michael Krigsman: So Keith, how do you, with your company and your customers, [...] go beyond the AI-as-a-shiny-object into AI that’s something that’s really useful in a practical sense?
Keith Brisson: Yes. So, what we’re 100% in agreement with [regarding] what Sean said [is] that ultimately, our goal as technology providers should not be providing technology. It should be to solve the business use-case. So, there happens to be the letters AI in our name, Init.ai, but we don’t generally talk about the technology. That’s kind of what we advertise when we’re talking to companies. We talk about the use-cases that we can help them be more efficient in, in the problems we can help them solve.
And, to Sean’s point, I kind of cringe when I see other companies really advertising the technology, focusing on the fact that they’re an AI company because it tends to mean that they’re not focused on solving your actual problems. So, what we always strive to message is how can we make your employees more efficient; how can we increase your revenue, decrease your cost, not what kind of neural network we’re deploying into our system; because that’s not really what matters at the end of the day. So, you know, for us, yeah we’re an AI-powered company, but we’re not an AI company. We’re solving business problems. So, does that answer your question?
Michael Krigsman: Yeah, and I think Ed, and certainly from what you were saying earlier, when you invest, you’re also investing in companies that are solving business problems, although they may be using AI to produce those solutions in a better way.
Ed Sim: Absolutely. And, AI’s like not a panacea in the sense that, you know, robots today or AI is not going to replace every human, right? So, the question is … You know, I’ve been very interested in human augmentation using AI so you do things that are better, faster and more efficient. I think that’s a tough sales proposition; the company comes in and says, “I’m going to replace everyone.” The question then is, you automate things, you do things 5x faster, connect faster, what else can those employees do? And that, to me, is more exciting so …
If I were a business, we talk to a lot of CIOs, I ask them, “Find out two or three pressing problems. What’s a low-hanging fruit inside of that business that can be partially automated?” Some of that can be automated data entry, right? Tons of people are doing that. “Look at anything that’s put offshore.” All that’s stuff that’s put offshore: you remember that big trend back in the day? A lot of that’s going to be replaced with AI, you know? And so, we think about that customer service, right? That’s another area that’s ripe for disruption as well. Accounting, AR and AP can gather that data, load it up, and try to automate.
So what we want at the end of the day is … I don’t care about the AI. If I’m a SaaS company or a software company, I’m going to look at the dashboard that the end-user uses, and “Give me answers.” Don’t give me, “Invest in beautiful data and visual screens.” I want to go to that screen and say, “Hey. These are the three companies you need to call today to see if you get paid.” And that’s what I want. I want answers. So, that’s what I want AI to do for me, not give me tons of dashboards, but give me an answer; make sense of that data, and give me something that I can really trust.
Keith Brisson: One thing to keep in mind, Michael, is that we’re part of this. You know, people are using the word “AI” as though the strong AI that Sean explained at the start exists today. The fact is it doesn’t, you know? We don’t have computers thinking like humans today, and we aren’t going to in the near-term, so we’re not going to be able to replace one-for-one a human with a computer. So what we need to do is take this technique, and apply them to specific verticals; to specific use-cases within them where we can see real value generated today.
So, AI as a term, I think that many people think about it aspirationally for where it’s going, and it is going there. And companies like ours and Sean’s are really going to help get us there in conjunction with the big guys where we’re doing some pure research. But today, we can take pieces of that; take pieces of what is going to get us to that point in the future and solve real business cases today. So, there’s a little bit of conflation of terms in this. You know, it’s a real buzzword, but there’s value there in the pieces and how they can be applied.
Michael Krigsman: We have an interesting question from Scott Weitzman on Twitter, who is asking, “What are the key elements that you can use to classify or differentiate AI on the spectrum from technology to business, in order to help businesses understand the nature of the kinds of problems that AI can solve?” In other words, how do you explain it well to businesses that they get?
Keith Brisson: You know, I think that we’ve certainly faced this challenge when we’re describing how we augment or replace humans in customer support, and you know, we’re very clear with companies that we’re not creating something that’s inimical to humans. What we’re doing is we’re making the business process more efficient, and really focusing on those particular segments where we can apply the technology. And I think it’s a matter of protecting the messages. So, if a company comes to you, and they’re telling you about the AI technology and not telling you how they’re going to help your business, that should raise a little red flag; raise an eyebrow for you and you look in a little bit more deeply.
So, I’d say when you’re trying to evaluate these solution, really consider what business process it’s trying to solve, and the ones that are appropriate for that are ones where there is a fair amount of data that is unstructured and it needs understanding around that. It could be language data; it could be financial data; it could be anything else.
Ed Sim: Yeah. I find it comes into two camps in enterprises, and just depends on the use case. It comes down to, "I can help you be smarter," i.e. there are predictive natures of AI, so whether there's a marketing dashboard, or just helping make decisions faster, so that's one aspect of a pitch. And then you can backfill that with whatever problem you're solving. And the other is, "I help you save a lot of time and money that you can free up your resources [with]." There are two different kinds of approaches. Predictive versus the general saving money aspect, and that's kind of the business problems of how AI helps you.
Michael Krigsman: Or to put it into broader terms, you have the efficiency aspects, and say, the innovation aspect?
Ed Sim: Sure.
Keith Brisson: Yeah.
Michael Krigsman: Let’s change gears here, and talk about when you’re thinking of designing your products, developing your products: Is it different when you start incorporating AI-related technologies from traditional software development? What are the differences? How do you think about it differently, design products differently, skills, and so forth?
Sean Chou: I think there is a difference, but it’s largely in the newness of AI technologies. If I were to replace the word “AI” and just say “database,” for example, when I think about a product, there’s no challenge in where a database lies, architecture, or solutions out of our product or the features that it can enable. AI, I can think of in the same way as a component that I can leverage, but it’s new. So, it’s kind of unlocking a set of features and a set of capabilities that we haven’t … like, patterns just aren’t as well established.
I think some are, and that's why among a lot of AI companies, there's a common set of things that are very common. Like, you'll see sentiment analysis, you'll see categorical processing a lot, you'll see machine learning a lot; these are patterns that have emerged as output, and so those are kind of a little bit - I don't want to say a "no brainer," but established patterns that product architects really use to leverage, whether they're engineering or thinking about a product management perspective. But, there are a lot more use cases that we haven't really quite figured out yet because of the newness of the technology.
Michael Krigsman: How …
Keith Brisson: [...]
Michael Krigsman: Oh, oh please. Please, go ahead, Keith.
Keith Brisson: Yeah. I want to just mention that it definitely requires a little bit of thinking, at least in terms of the implementation level, because there’s generally the requirement for data. If we’re using a solution provider, they can offer that data up-front to help you get started. It may not be a consideration, but data plays a role in any kind of application of AI and machine learning today, and if you talk with a solutions provider, that is going to be a concern, like how do you use data to make this thing more efficient?
And the other thing to consider is that these AI systems and machine learning systems, they evolve over time based on back-cycles, or at least they should. So, you can get an immediacy. The value that you get from these kinds of solutions up-front tends to increase over time, as they see more cases in their then-working back-cycles. So, it's a little bit different in that it's not static; it's something that's evolving over time.
Ed Sim: Michael, can I add one other piece? I think Keith makes a good point from the data itself, and if you're selling in enterprises, any type of solution that leverages their data I think a big question no matter how many folks talk about the cloud is, "Can you drop it on-prem?" And, I don't know if either you've been asked that question but, it seems to me that every company that we have leverages some type of data, and to run kind of AI on top of it is, "Can you drop it?" I can move the data, there are all these regulations, and I need the securest you guys can [provide].
Keith Brisson: Well, we get that all the time, especially since a lot of the innovation in AI has come from large companies that have a tendency to suck up data, you know, the Googles of the world, which most enterprises are not willing to trust with their valuable customer data. So, we get that all the time, but data is a requirement, right? So, our [plan] at least, is to provide our own data to help get things started. So, like in our case, it’s a general language understanding data, but then let companies maintain control over the data that’s specific to them. Our strategy on that is to offer on-premise or virtual cloud insulation so they can maintain control over that. The simple fact is most companies don’t want their data training other companies’ datasets.
Ed Sim: Yup.
Keith Brisson: … especially in regulated industries like healthcare and finance.
Sean Chou: Yeah. Data is definitely kind of the muck-side of AI that people like to not talk as much about, but it is really important, I think, for any company that's really looking at AI and what their AI strategy is going to be over time, which, I think, you know, we all are in agreement there should not be an AI strategy, there should be a business strategy powered and enabled by AI tech. But, you know, you have to get your data policy in order, so they definitely have to have a point of view on where data resides, how to own, and so-forth. And I don't think there's a right answer right now. Certainly, at that point, we get the questions to of, "Hey, where exactly is this data? Where does it reside?" It becomes a bigger problem once you start talking about a global context because then it's no longer just US privacy issues, you have entire governments that have a very strong perspective on where physically data should reside.
Michael Krigsman: Let me …
Ed Sim: We are a company for that, too. [Laughter]
Michael Krigsman: [Laughter]
Ed Sim: That’s probably the big idea, and we do use AI and machine learning to tell you where the data is, where it’s located, and we attach it to a unique ID before people kind of run scenarios on top of that, so…
Sean Chou: Got all your bases covered.
Ed Sim: [Laughter]
Michael Krigsman: That is clearly looking at all the different facets of this in order to cover, as you say, cover his bases. This issue of the data is so important, and would it be accurate or inaccurate to say that for companies that are thinking about AI, that it’s actually a data situation, rather than an AI technology situation or question? Is that accurate or inaccurate?
Ed Sim: I would say it’s pretty accurate, right? Because a couple things: I’m so bullish on the enterprise and leveraging machine learning in the enterprise, because once again, a lot of these folks don’t want to give their data to Google to train them, right? That’s Google’s model. Hey, I’ll give it to you for free, I’ll open-source it and [...] Facebook with that AI, but there is tons of value sitting inside of enterprises, right? And the more data you have, the better you can train, the better you can predict. So the question really is, I’m dropping something on-prem, and I’ve got healthcare data. That’s interesting. But somehow, I can tie that healthcare data anonymously with other healthcare data from other companies. Then you start creating a powerful data co-op and you can make better data predictions. But I think data is definitely a weapon, and it’s definitely a weapon in the enterprise space.
Keith Brisson: That’s definitely a weapon, but it’s not even all that you need. The technology and the skills required to process that data and turn it into something actionable are pretty technical and can be a distraction for most companies. So, for most companies, you can have the data. We also need the expertise to be able to transform it into something that’s actually usable within your business. And that’s something that tends to not make sense for companies to build themselves. I think that’s at least our premise is that we should provide that technology so that businesses can focus.
Michael Krigsman: That’s a really interesting … Oh, please, go ahead, Sean.
Sean Chou: I was just going to add, I think that the data is certainly essential for the majority of AI applications right now, and it's largely a function of where we are. When we narrow AI for machine learning, for developing neural networks, you need data. You need feedback to improve that neural network. But, as we start getting more and closer to more general AI, or even very well understood things like, you don't really need to train sediment analysis that large now because that's a more general application. So, you're already seeing some applications and outputs from general AI research that doesn't really need to be trained.
So, I think that that shift will occur over the next decade or two, where we right now need a ton of data, especially for domain-specific applications, and we're going to see more and more "general" general AI come out that won't need the training as much, or won't need as much data power.
Michael Krigsman: This whole …
Keith Brisson: I actually absolutely agree with that. Data is almost a crutch right now because the techniques around general AI have yet to be developed. So data is replacing the sophisticated techniques that are on the rise.
Michael Krigsman: It seems like it's quite a paradigm shift to some degree for people in the enterprise who are thinking about AI because they are tending to think through the lens of the technology, rather than - because it's so new - rather than necessarily thinking first and foremost about the business problem. And then the fact that so much of AI involves patterns and therefore requires that large volume of data, it requires a different way of thinking about how you go about solving problems. So there is an education process, I think, that is still having to take place, and the result of this is all of the hype in the software and technology industry because you can basically say whatever you want about AI. Everybody's got an AI, right? I've got an AI! [Laughter]
So, I was recently talking on this show with James Cham, who is with Bloomberg Beta, and he was saying we need a framework for making decisions inside the framework of the enterprise; investment decisions. So, Ed, let me toss it back to you. When you think about investment decisions that involve AI, what are some of the criteria and things that you think about?
Ed Sim: Well, I don’t try to make a specific technology decision on AI investing. I mean, first and foremost, we tend to be investing in technically-driven founders like Keith and Sean. So for us, it's deep domain expertise and understanding of the business problem they're solving. If they didn't come to me and say, "We're building an AI company," right, and they said, "We're solving this problem; it's a big, big opportunity," … And by the way, one of our core pieces of technology will be some use of AI, and then we fit it in through that a little bit. But that was kind of the main reason we fund it. So I think that's really, really important. It's just like every other tech trend out there. Back in the day, it was Java. You know, there were Java funds out there. Then it was everything with mobile. And now, it's AI.
But guess what? I view AI as just like water. I mean, it’s just like electricity. Every company in the next five years, every technology company is going to use some form of AI. And so, I don’t view that as separate [...] per-say. That’s why I say “applied AI.” What business problem you’re solving, and how you’re doing it 10x faster and 10x cheaper; if you do both, that’s an amazing order of magnitude improvement.
Sean Chou: Totally. Like is there a web company today, you know?
Ed Sim & Keith Brisson: [Laughter]
Sean Chou: But there were! I mean, ten years ago, or fifteen years ago, time flies. You know, fifteen, seventeen years ago, there were web companies. But there are no web companies today.
Michael Krigsman: Well, if you think about it, when Salesforce started, they said, “We’re doing this Salesforce automation over the web.
Sean Chou: Yup.
Michael Krigsman: … because it was new, and it was unique. And so, your feeling is that given some certain amount of time going forward, AI techniques will be incorporated just about everywhere, essentially?
Sean Chou: Yeah. I'm 100% convinced of that. And we're talking a lot about machine learning and human augmentation, but there's just a lot of simpler things like AI … Applying a lot of AI principles just makes a better product, because you are asking a person to do as it's thinking because the product can do more on behalf of the person. So, you're going to see a lot of very subtle importance about AI, because it's just better product. It just lowers the cognitive computing costs on the human's side.
Ed Sim: I love that Sean. I’m thinking that great AI is invisible to the end user. It just works! It’s really easy and it just works. There’s no friction involved, and I think that’s what great AI is. And that’s why Echo’s kind of cool. I mean, it doesn’t fully work all the time. You just talk to it and it works, most of the time.
Michael Krigsman: [Laughter]
Ed Sim: That’s what the great AI is. And that experience to bringing that to the enterprise, and the enterprise-level problems, that is pretty exciting for me. And no-one’s kind of doing that yet, and we hope to do bits and pieces of that with Catalytic and Init, but I think that’s a huge, huge opportunity.
Michael Krigsman: What are the …
Sean Chou: We work so hard for end-users to say, “Wait, what just happened there? [...]” Like, that doesn’t seem like it was really a lot, but they don’t appreciate how much work that’s going on behind the scenes to make this slow, magical moment happen.
Michael Krigsman: What are the inhibitors, or the obstacles that prevent this kind of adoption that Ed was just talking about in the enterprise? What needs to be in place to make it happen on a broader scale?
Keith Brisson: Personally, I just think it's not our time. You know, there are different companies tackling different components of it, and it's a matter of … a lot of it is integration, so as data becomes more connected between different parts of the enterprise, you enable new ways of taking these techniques and servicing value to people who [...] it.
So, to me, it’s just a matter of time; it’s a matter of these cool things being incorporated into the workflows where you’re not going to really see them.
Ed Sim: I would also say that there's an infrastructure answer to that, too. A lot of enterprises that we talked to; I'm sure we talked to Michael as well; are kind of re-platforming technologies, evaluating how to bring hybrid cloud into the enterprise. If you look at Pivotal, Pivotal just announces 72 million dollars last year from one-third of the Fortune 100 to help them think about infrastructure and cloud. So, once you have an agile kind of platform to build off of, that makes it that much easier to develop and deploy an AI plugin on top of their existing application. So, I think part of that too is just what the underlying infrastructure looks like.
Keith Brisson: Yeah. And, it’s getting there, you know? We talked with enterprises, too. A few years ago, all of their data was siloed in different systems with no interconnectivity. We talked to them today and they actually have internal APIs and ways of connecting data together which actually enable these kinds of applications. So, I think we’re right on the cusp of the point where enterprises have enough, or are getting enough connectivity between their different data stores, that we can [start] with applications really blossom into being every part of the enterprise.
Michael Krigsman: Keith raises a …
Sean Chou: I think one of the biggest inhibitors right now is just the lack of clarity as to what exactly AI is, because, on one hand, you have the extreme end of the spectrum that actually creates fear: the all-Terminator, Skynet-type of AI; all the way to, on the other end, where people have said, "Hey, I've just whipped together this thing, and it's AI," but a technologist will look at that and potentially say, "Well that's not really AI! You're just regular expression matching," right? So out of these extremes of what AI … what's being proposed and packaged as AI that lack clarity, and the buzziness of it all, actually makes it very hard for people to adopt, because they have to wade through all this crop to get to the real business value behind whatever the product is.
Michael Krigsman: Well, like I said before, it seems like virtually every technology company is selling an AI.
Sean Chou: Yeah!
Michael Krigsman: Hey, I have a couple of AI's. Do you want to buy one?
Sean Chou: Yeah! [Laughter] Yeah, yeah. That’s right.
Keith Brisson: Yes. You know, I think that probably the most extreme example is where they try to portray it as a single AI: this kind of all-knowing mind that can perform these superhuman tasks. But the fact is, generally what they have is a collection of machine learning-powered APIs or services that enable individual things, and they may sell it as this kind of super-knowing mind that’s going to form Skynet and Terminator, but the fact is, they’re not. And, as long as businesspeople recognize that those … that all-knowing mind may not exist, but the individual components can provide a real, concrete value, it shouldn’t be a reason to not adopt those technologies just because it’s not living up to that hype.
Sean Chou: Yeah. In fact, the AI industry, or people who have AI, almost do a great disservice by overhyping AI. And so, we try to, the minute we have a conversation with a customer or prospect, demystify what it is we do, specifically, and say, “This is very tactically how we use AI technology,” because we want to very quickly get to the value that we’re added, and not be kind of caught in this whirlwind of AI and this buzz of AI. I think if you let yourself get caught in that, and if you lead too much on it, you’re always going to disappoint people, because you’re not ever going to live up to the science fiction portrayal of AI, at least not for the next few years.
Michael Krigsman: Well clearly …
Ed Sim: The reality of it is, if you talk to enterprise buyers, they don’t say, “I need an AI;” I need AI, right? It’s like, “I have this problem, and if you can save me 30%, or do things faster, then it’s interesting to me. And if you leverage AI, very cool, but that’s not my checkbox,” right? And that’s not the checkbox item. So, I think we also need to think about the budgets: why they’re buying things and what ROI you’re providing. And in my mind, AI can help create incredible ROI, but you’ve got to apply it to something.
Michael Krigsman: We have just about four minutes left. And, maybe we can just go around the virtual room, as it were, and let me ask each of you for your advice to business buyers; to people in the enterprise, who are looking at these technologies and hearing about - again, every vendor is just hyping this to the max, whether there’s substance behind it or not. So, what advice do you have for people in the enterprise for sifting through the hype so that they’ll get something useful from AI? Who wants to start?
Sean Chou: Me. I can start very quickly. I would just say, stick with the basics, you know? Make the business case, look at the value that’s being created. AI, like I’m saying, is not a checkbox. It’s an enabler. Where you see people making claims about AI, I would say, you know, what’s the business value and if AI’s being used to dramatically multiply the business value, or if you can see incremental gains from any other non-AI types of solutions. I think AI’s going to evolve; what you’re looking for the potential to really move the needle. Not a 10% gain or a 20% gain. Maybe AI allows you to tackle new problems, or allows you to get 2, 3, 10x type of gains over a traditional solution.
Michael Krigsman: Yeah, that’s really interesting. What asking that core business question, “What are we doing; what’s the value; what are we trying to solve?”
Sean Chou: Yeah, absolutely. Stick to that.
Michael Krigsman: Keith, your thoughts on advice to folks in the enterprise who are hearing about this technology, and what should they do?
Keith Brisson: Yeah. My advice is to try to calm yourself down when you hear the hype. A lot of it is … It is hype, but there’s real value there. Like, this is a long-term trend that is just getting started, and it is going to transform the way business processes are tackled throughout the enterprise. And, really focusing on these media business use-cases like what Sean and Ed were saying is absolutely the right way to go. It’s not about technology, it’s about the business process. And, you know, I would encourage enterprises to be cautious about the hype, but really optimistic about what use-cases these can enable, because there are going to be entirely new categories like what Sean was saying, and there is real potential there to transform mass parts of enterprise workflows.
Michael Krigsman: So in other words, don’t buy into the hype, but focus on the real problems and practical solutions.
Keith Brisson: Absolutely! But, still be excited. It’s okay to be excited about where this is going because the long-term trend is there, and it is real. It’s just people are just a little ahead of it. So, that’s all.
Michael Krigsman: That’s a great point! And, finally, Ed Sim, your thoughts. You’ve been involved in folks in the enterprise for a long time. What’s your advice for people who are hearing the hype and trying to figure out what to do?
Ed Sim: If someone is selling you AI snake oil and that’s your initial pitch, run for the hills!
Sean Chou & Keith Brisson: [Laughter]
Michael Krigsman: [Laughter]
Ed Sim: [Laughter] The reality of it is, is that for example, you solve a business problem and think about what problem you have, and AI and some form of AI can help you solve that problem and help you do it much faster, and there are tons of companies in every category leveraging AI. So, really make sure you talk with a few different folks, whether it's large companies or startups - hopefully, startups, because I'm …
Keith Brisson: Agreed.
Ed Sim: And, the second thing is, as far as new categories, I mean, Security Scorecard, for example, created a security ratings market overnight, and that wouldn’t have been enabled without leveraging neural network technology, machine learning, rules-based classification - all that stuff - but they’re not going in and saying, “We’re an AI security company,” they’re going in and saying, “Hey, I’m going to help you figure out which third-party vendors that you have, that have not kind of attached their security systems, and might be risky for your company.” AI and machine learning never gets brought up. So, I think the companies that are the best ones are the ones that actually help to solve a problem, and they actually have amazing, amazing technology, but they’re not pitching that as a first kind of entry point.
Michael Krigsman: Great advice! Clearly, there is a unanimous decision here that the way to go is you solve the business problems and find technologies that will enable those problems to be solved in very dramatically better ways.
Everybody, thank you so much for watching! You have been watching Episode #222 of CxOTalk. Our guests today have been Ed Sim from Boldstart Ventures, Sean Chou, from Catalytic, and Keith Brisson, from Init.ai. I'm Michael Krigsman and tune in again next week. You can see all of our upcoming shows at CxOTalk.com/episodes. Thanks so much. Bye-bye everybody!