Artificial Intelligence in Commerce

Brian: [00:00:25] Welcome to Future Commerce, the podcast about cutting edge and next generation commerce. I'm Brian.

Phillip: [00:00:30] And I'm Phillip.

Brian: [00:00:31] And today we are interviewing Jonathan Epstein, the CMO of Sentient.AI, an unbelievable AI focused company. And we're very happy to have him on the show to talk about AI today.

Jonathan: [00:00:48] Great to be here.

Phillip: [00:00:51] And this is our first episode, actually, about artificial intelligence in depth.

Brian: [00:00:55] Yeah. Full on. And so we would love to have you give us feedback about today's show. So please leave us some feedback in the Disqus comment box on our site. You can also subscribe to listen to Future Commerce on iTunes and Google Play or listen from your Amazon Echo and TuneIn Radio with the phrase "Alexa, play Future Commerce podcast." And with that, yeah, I'm super excited to have you, Jonathan. Maybe you could introduce yourself and give us a little more of the story behind Sentient. How did you get your beginning? How did you get involved? Give us a little history.

Jonathan: [00:01:34] Sure. Absolutely. Be happy to see. So Sentient is we're a mature startup. I guess we'd be called in Silicon Valley. The company is nine years old. So we've been doing AI for quite a long time. AI is not a new thing. It's been in the offing since the 60s or so. But as far as this most recent wave of AI, we started earlier than most. And the genesis, the company was our two founders, Antoine Blondeau and Babak Hodjat who are our CEO and our Chief Scientist respectively, were sitting around and there may have been a shower epiphany involved, but basically the question they asked themselves was "What would be possible if we were able to apply artificial intelligence at truly massive scale?" Now, both of them had worked in the space before. The two of them had worked together at a company called Dejima, which was sold to Sybase and Dejima did a project to the CALO project which ultimately morphed into what became Siri. So they were experienced with AI, and Antoine in particular, was entranced by the notion of could we find a way to apply vast amounts of compute to artificial intelligence. So set about with Babak and with initial seed funding to build out a very powerful AI platform that not only combined a number of different types of AI... I'll go into that a little bit later, but also combined the middleware layer, what we call dark cycle, which allowed these models with these evolutions or these, you know, these problems to be run across a huge number of computers. That network is now about two million CPUs and about five thousand GPUs. So an amount of compute that, you know, you might have if you're Google or your Facebook or your Amazon, but probably you don't have, and if you were to try to buy that amount of compute power on the Amazon cloud, for example, A) you couldn't, B) if you could, it would be really expensive. So that was the core theory. And the company's goal is to take this platform. We're not a platform company. We're a product company. And build products that achieve breakthrough performance or step function performance in an industry at a time. The first the first business of the company was in the trading space. We run a hedge fund. It is the first fully AI traded hedge fund.

Phillip: [00:04:11] Wow.

Jonathan: [00:04:11] AI has been around in financial trading for a while. But usually the AI would give recommendations and then traders would execute the trades, so the difference with what we do is that really the only human input is to say here are the signals you should look at, and they might be complex derivations of various technical indicators. But once that input is given to the AI, in the case of the hedge fund, it works on a field of AI called evolutionary algorithms. Once that input is given, then the machine figures out and essentially evolves and breeds virtual traders that then once they're trained up initially on historical data and then on real time data, but not with real money are finally the best of them are released into the fund and they execute trades. They do the buy the sell themselves. They figure out how to enter and exit positions in the nature of the orders and things like that. So that was the first business we were in.

Brian: [00:05:17] So cool.

Phillip: [00:05:18] {laughter} As if that weren't enough. I mean, fintech is obviously massive, right now, at least here, you know, in the Western Hemisphere... As if that weren't enough, there's more?

Jonathan: [00:05:33] Oh yeah. Certainly and that's what brings us here today. So in terms of other commercial offerings, about three years ago we started looking at eCommerce as, and maybe commerce more generally, as an area that we could make a lot of difference in. We look at the state of eCommerce, and obviously, there's been a huge growth in eCommerce and we're seeing big players emerge and all sorts of interesting models. But in terms of some of the fundamental characteristics of eCommerce, how the customer experience is designed and created, the type of conversion rates that people enjoy, those things haven't necessarily changed so much over the last 10 years. And so we saw an opportunity there for AI to really transform the way people are able to run and execute eCommerce businesses in ways that would increase revenues and profitability and customer satisfaction. So we'll talk about that. Just to mention that we also do quite a lot of research with leading institutions. I should note in saying people are saying, how do you do all these things? Sentient has raised more money than any other independent AI company, so we have raised about 143 million dollars to date. So we're amply resourced to both build a platform, create a number of products, and we do a lot of our platform research with research projects, with people like MIT, MIT Media Lab, Oxford. And in those areas, we've done some very interesting work in health care. We've created a predictor for sepsis.

Phillip: [00:07:08] Wow.

Jonathan: [00:07:08] We're now working on closed container agricultural optimization. What if you could grow any crop anywhere? Our role in that project is finding the ideal recipe, so to speak, the right combination of nutrients and light and water that will create the maximum crop yields for those projects. You know, we're doing a lot of fun stuff we can do. We do quite a lot of primary research. We have applied for thirty one patents. We've been granted 11. We publish papers. It's really a remarkable institution. I came there as actually, my background is in media and in technology. I used to run a number of companies. If you're a gamer, you might be familiar with GameSpot. I was the founding CEO there.

Phillip: [00:07:52] Oh yeah.

Brian: [00:07:53] I love GameSpot.

Jonathan: [00:07:56] In its day it was a breakthrough. You know, in 1996 when we launched it, no one had done that sort of publishing on the Internet. So I've always tried to stay on the leading edge, and I ran GameSpot, and we sold to ZDN. So I ran that for a while. But I'd taken some time away from operations and had become an executive recruiter for a couple of years. And Sentient was one of my clients. And as I worked with them, I guess two things happened. One is I fell in love with the company and its vision and really the premise of what AI can do, not just for finance, not just for commerce, but for so many different industries and how that could change society for the better. Also the financial promise of achieving such goals. And at the same time, they loved the job specification I'd shared for them so much and how well I was able to tell their story and sell the company and recruit a number of the top executives that are there today that they brought me in in the marketing role. So that's how I got there.

Brian: [00:09:00] Nice.

Phillip: [00:09:01] Wow. That is also officially the best intro of any show that we have done.

Brian: [00:09:07] I'd say so. Yeah. That was awesome. Yes. You're playing AI to obviously several industry. Commerce, hedge fund, and several other industries. Commerce is what we really want to drill down on. And you guys have a couple of products that are focused on commerce, Ascend and Aware. So maybe you could give us a quick rundown on, you know, how do they work, what they can accomplish. Yeah, just to give us quick take on those.

Jonathan: [00:09:47] Sure. Absolutely. So let me start with Sentient Aware. Sentient Aware is about a year old in terms of its commercial release, and its first customer deployments happened in November of last year. Sentient Aware is really a platform that uses AI to understand product catalogs at incredibly fine and nuanced levels, both visually and in terms of metadata around products. And it also understands user clickstreams. What our users are doing on your site? And it brings those two inputs together to enable a pretty wide variety of applications that are designed to increase engagement and increase conversions based on, again, those two inputs. So some of the ways that people use Aware, one is this notion of adaptive merchandizing, right? I think with most eCommerce sites, when you go and visit the homepage or the category page, everyone pretty much sees that same page. There might be a recommendation bar at the bottom. The actual core order of the products displayed is based on maybe popularity, some simplistic algorithm, maybe just the order that things are in the database may be promotional or priorities, but by using Aware it creates the potential that you can have each of those pages be tailored to the user and their particular interests. So right away from the start, you were able to create a more relevant and engaging experience for consumers. Another key benefit of it, and this is where we sort of originally the first users used it, is in the area of visual search and product discovery. We're not a click to buy. We don't we don't take pictures of shoes on people's feet and tell you what they are. The notion here is when you have retailers that have, you know, medium to large catalogs or really large catalogs can be very hard to find what you're looking for, particularly with visually driven products. If you search for a black dress or black cocktail dress or little black cocktail dress or whatever on Amazon, you'll get 10, 20, 30 thousand responses. And of course, what that means to one individual might mean something very different to another individual. So, you know, everyone has been doing tagging for a while and trying to use that as a way to help people search. But it's very hard to describe things that are visual in words that are consistent. So Sentient Aware allows eCommerce operators that sell certain categories like fashion and shoes and watches and eyeglasses and home décor, the ability to create interfaces where the products are themselves the navigation into the product catalog. You can see this on ShoeMe.ca, which is Shoes.com, a Canadian site, where you come into a category and your first display the AI basically says, "Hey, here you're looking for a men's boot," let's say, "Here are the ten most distinct versions of that boot. And then you click on one, and it'll display 10 that are like that. And a lot of similarity or visual search systems could do that. We think we do that a little better, but it's with the second click that the magic starts happening because all of a sudden we're starting to understand, you know, all right it was the color and the heel, but not the buckle that you were interested in. And when they click a third time and a fourth time, we are really able to hone in on what is the user intent in the moment and deliver a product that will match that intent and lead to increased engagement and conversion. Sentient Aware is also a recommendation tool, so at its core, we use this huge network of CPU's and GPUs we have to build very complex models around different types of products. We take a catalog and embed it in high dimensionality space, they would say. So we sort of understand how similar any one product is to any other product across hundreds of dimensions, which might be visual or might be things like brand or what...things that aren't necessarily visual. And then we let the user essentially path their way through that huge cloud of product and deliver them a personalized taxonomy of those products, the products that they'll want. Now, most of the applications have been on site so far, but we're now starting to extend Aware's capability, so that if you want to do an email that is trying to bring someone back to the site, you can not just send them the most popular products or the ones that are on sale. You can you can do that, too, of course. But the specific products that for that individual were right for them. So it's really moving us in the area of true personalization. So that's Sentient Aware. Sentient Aware knows what products you should show your customers and helps you increase engagement.

Brian: [00:15:19] Gotcha. So this is kind of an off the all thought. We had Healey Cypher, who's the CEO over at Oak Labs on the show a few weeks back. And he talks about how Oak was applying sort of what the eCommerce techniques and tools to the physical store. I can't help but as you were talking, my brain was just sort of thinking about, you know, man, would it be awesome if retail associates had power like this as shoppers were coming into the store?

Jonathan: [00:15:58] You're right on the money. And in a way, we think of this as what can we do to create the best attributes of a highly trained sales associate in a store and bring that power online? A good sales associate will know the customer. They'll know their habits. They'll know the context of their visit. They'll understand their tastes and preferences and how they might even shift seasonally. So that's what we're trying to do. We do have Shoes.com, which I mentioned, one of our customers who is using this in this store. They're using it primarily for consumer facing, so large format screens in the store. But it's a way to help them navigate the so-called endless aisle.

Brian: [00:16:44] Yeah.

Jonathan: [00:16:45] And get to the products you want. We also work with a company called ThoughtWorks. There are a pretty well known retail systems integrator. And we've been exploring different types of clienteling apps where, yes, if you if you're Nordström to your sales reps are well trained, well paid, and they know their products and they know their customers. But a lot of retail stores don't have that same kind of staff profile, staff turns over a lot. So yes, you can use Aware to arm them or the customer service people on the phone...

Brian: [00:17:15] Yeah.

Jonathan: [00:17:16] With that same knowledge of, "Hey, if you couldn't find this, these would be good things to try" and make the sale.

Brian: [00:17:23] I love that.

Phillip: [00:17:25] I also had sort of a follow up question, because what you said about, you know, a well funded business probably has the ability to do some more interesting things in the space. And, you know, particularly I think you mentioned like Shoes.com or at least their Canadian arm. They're doing some of these interesting things today with Aware. But I can imagine that you would have a phenomenal amount of training, data or data available for a very large catalog on a marketplace like Shoes.com, where you're getting shoes from all kinds of vendors with incredible depth of catalog. Is that something that you could potentially leverage and use with greater efficacy for a shoe vendor than, say, for someone else that you would have to have unique training data against? Does that put you, like, uniquely able to service footwear companies now? Or tell me how that kind of works and how you would apply that sort of to other verticals at Sentient.

Jonathan: [00:18:33] Yeah, absolutely. And it's a very astute question. And our model building pipeline continues to be enhanced. We've really made some great strides there recently. So when we started out, let's just take shoes as an example. Every vendor had their own sort of model developed for them. And there was an initial amount of training. We used some of the people like Spare5 or Mechanical Turk to essentially get things going. And at a certain point, we're able to then use the company's catalog and build out the model for them. But that's no longer the case. Now, we do have a master shoe model for each category and now we're moving to a master category being like heels versus boots versus athletic, sandals, etc.

Phillip: [00:19:23] I see.

Jonathan: [00:19:24] But now we're building the ones shoe model to rule them all. And we've applied those same techniques into fashion and dress. I mean, dresses are a much harder problem than shoes for a number of reasons. The back matters as well as the front. The pictures are often taken on people, where shoe pictures are pretty consistently, you know, generally a certain angle or a set of angles. And they're not on people's feet. So that's taken some more doing. Sunglasses. We have a sunglass model where we'll be deploying pretty soon. I can't name the name, but a pretty significant chain there. And if you think about sunglasses, you have some basic design differences, but a lot of them look very similar, you know, so we had to find really nuanced approaches. It's a model building and then the ability to deliver that, you know, in a cost effective manner. Those are things that we've been particularly good at and continue to get better at it.

Phillip: [00:20:25] And I'm going to sort of help you segue way back to talking about Ascend, but one of the big problems with smaller retailers or retailers that are sort of trying to take the next step to applying data to their decision making so that they're kind of getting away from the gut feeling approach and becoming more data driven. One of those things is they they tend to turn to tools like multivariate testing or or a/b tests to sort of help them make better reasoned decisions. But what they often find is that they just don't have enough data, whether it be the types of tests that they're trying to run or the pages are running it on or the traffic that they actually would need to come to some sort of statistical significance to even make a decision with classical models. And I know that you sort of have a solution to this with Ascend. I'd love for you to pitch it, but the thought that I'm going to ask you to try to answer is under the guise of, you know, is it the type of technology that can be or that is really can be consumed by SMB? Not that we're in SMB focused podcast, but is it something that's sort of broadly available to all of digital commerce? Or is it something that is better utilized with like what you've done with Aware where you have your building models that are specific to a particular vertical that make you exceptional at that particular vertical? Or is it something you grow into? So go ahead and answer before I answer for you. {laughter}

Jonathan: [00:21:57] Yeah. So, I mean, let me start by just talking about what Sentient Ascend is.

Phillip: [00:22:02] Right.

Jonathan: [00:22:02] And setting the stage for that is that. I was a ThoughtWorks conference the other week at a conference called Paradigm Shift, and they had the Founder of Fast Company Magazine as one of the keynote speakers, and she spoke a lot about experimentation and testing and how really it's at the heart of becoming a break through business in the next 10 years. And that the key thing to maximize is insights divided by the sum of dollars plus time. The more things you can learn more quickly... And this wasn't just limited to the website. It was a great way of thinking about it, because we do have access to data now. And so if there are ways that allow us to use the signals we get more efficiently, if there are ways to allow us to use them more cost effectively, then this is a real powerful advantage for businesses that take these up. So Sentient Ascend is the first conversion rate optimization solution, some people say conversion and revenue optimization solution, that's based on evolutionary algorithms. So it's a different form of AI than what drives Aware, which is built on top of deep learning and online learning stacks. And the great benefit that Ascend provides is that on the same traffic that you might have been able to do when a/b test or a few a/b tests or maybe a small multivariate testing, the testing eight different combinations of designs of your website. We have clients that are testing thousands, tens of thousands, hundreds of thousands of different potential design combinations to achieve results that much more quickly. Now, when I say hundreds of thousands of combinations, that doesn't mean you're changing one hundred thousand design elements or copier layout or colors on your Web pages. It means it might be 20 or 30 or 40. But believe me, it stacks up pretty fast in terms of all the different ways you might combine those different changes. And at the heart of this is this notion of evolutionary algorithms, which we use for the hedge fund, also where it's a branch of AI that's built around, at its core, the principles of natural selection. What's fitness? Fitness is what makes something better than another. So when a hedge fund does this trading, this virtual trading, does make money better than the other ones. And in the case of a website, it might be conversion percentages. It might be revenues, might be pages if you're a media site. There's a notion of combination, breeding things together, and that if you have two successful Web designs that are better than others and you combine them in some way, that some of that "children" of those two designs, some of them would be even better than the parents. Some maybe worse, some maybe the same, but some would be better. And then mutation is this other function and evolution. Mutation makes sure that we're exploring all corners of the map, as we say in statistical terms, and finding not just a good design that lists revenues or conversions, but finding the best design. And so by using these principles and the Ascend platform automates the testing process, you'll put in 10, 20 different things that you think will move the needle on performance, and it goes through a process of evolving different designs, and by using evolution, you can test and solve for the best of, say in one of our case studies, twenty nine thousand potential designs while you're only testing about one hundred and fifty of them. So that's a huge beat up over a/b testing. We wouldn't live long enough, well we probably would live long enough, but your business wouldn't be around long enough, to test you twenty nine thousand a/b tests. It's just, you know, you need to go mad separately. It's a game changer for testing and it's being adopted in a lot of the top consultancies and companies are starting to take it up.

Brian: [00:26:15] Absolutely. I think to your point, like a lot of, you know, sort of smaller mid-market players or even mid-market players that are only able to test... They do a few major tests throughout the year, maybe more than a few. But it's nothing compared to what you're talking about, right?

Phillip: [00:26:33] Right.

Brian: [00:26:37] It's exponential figure. And so it's just a very cool piece of technology that it seems like you've got some very cool studies to back it up. It's effective. I definitely really enjoyed your full on demo and honestly, for our listeners, if you ever get a chance to see a full on demo, I'd highly recommend it.

Jonathan: [00:27:03] {laughter} I could provide the URL if you want.

Brian: [00:27:06] Yeah. Go.

Jonathan: [00:27:06] Yeah, you can find it on the Sentient.AI website. But to answer the question you asked. Yes. This works for small businesses or small to medium businesses. It works for really large businesses. We have some customers that have fifty thousand users a month, you know, we have some that have ten million users a month. So yes the ones that have 10 million users can test even more things than the ones that have 50. But no matter what your size is, this method will work better to test more things. And by testing more things, it's like it's like throwing darts at a dartboard. Each test is a dart. And, you know, you might hit the dart board and you might miss the dart board and you might you might get a bull's eye. You might get a one point. You know, each dart is a little different. But if you can throw in the more darts, you know, the chances are that more of them will hit in the score. It's actually not just additive, it's multiplicative. It's the combination. There's a synergistic value in the combination of various design changes that add up. And so we see, you know, we see very exciting lifts in conversions and in revenues, 15 percent, 80 percent, and then the fact that it automates the whole process is, you know, some people love that as well. It also tests across multiple pages, which is not something current most current solutions can do. So I spend a lot of time on this. I was just out of the conference in London and the room is everyone's lining up to test it.

Brian: [00:28:41] So I kind of already know the answer to this. And so I'm going to ask it anyway and then I'll ask a follow up that I don't know the answer to. So do you plan to expand Ascend's feature set and capabilities in the upcoming year? And is there anything maybe that you haven't talked about are released yet that you want to unveil on the show?

Jonathan: [00:29:05] To the latter question I have to think about that. I think I think we tend to be pretty open about our roadmap. So I'm not sure I have any surprises for you there.

Brian: [00:29:18] That's fair.

Jonathan: [00:29:18] There will be an exciting brand new UI come the New Year before, so definitely working on that.

Phillip: [00:29:24] Oh nice.

Jonathan: [00:29:24] But I would say there's three areas for Ascend and the changes we made to it. One is just in terms of core features, the number of platforms we integrate into the number of analytics systems we integrate into. We just actually this hasn't been announced that we are just adding Adobe analytics integration. That should be live early next week.

Phillip: [00:29:51] Nice.

Brian: [00:29:51] Hey!

Jonathan: [00:29:51] So there you go. Currently integrate with... That's very helpful. You don't need to, by the way, to be integrated with a commerce platform to work with any commerce platform, but it can be very helpful as well. So things like that, better looking graphs, core feature stuff is one area. But the more exciting futures are in two other areas. One is, and we've talked about this before, right now Ascend will take your ideas. We don't make up the ideas. The ideas come from the marketers and the consultants in the agencies like that. That's really important. And we need more of them than other solutions. We take those ideas, and we build the best possible, the best web design to achieve your goal, say revenues, improving revenues for your audience or any segment of that audience that you define. The next version, which will come out of next year, uses a different type of AI, something called neuro evolution, where you use evolutionary computational techniques to evolve a neural network. Which is a very deep learning. And in that example, you decided you would say, "Here are my ideas for design changes, but here are the things I want you to consider about every user, what device they are, where are they in the country, what time of day is it where they are, what group they click on and what do I know about them if they register?" And we will evolve a neural network that will understand the importance in the relationship of all those different signals in defining the absolutely best design for that person at that time. So then you get into it. So both Aware and Ascend are focused when we have that vision on how do we really personalize the customer experience for a commerce environment? In the case of Aware it's what products should we show it? And with Ascend it's how best should we show them to achieve the goal of increasing a business? Now Ascend isn't limited to eCommerce. It works great for Legion, for B2B sales, other types of Web businesses, really any business that relies on these digital touch points for the lifeblood of their results. The other area we will be expanding in Ascend is right now it focuses on website optimization. But I think you can see us take it into email optimization and also advertising creative optimization. So stay tuned. We have a very exciting roadmap.

Brian: [00:32:38] Awesome.

Phillip: [00:32:39] Awesome.

Brian: [00:32:39] Awesome. Thanks for talking about the future. It's exciting.

Phillip: [00:32:43] From what we know, or from what I know about Sentient, you guys have a very clear vision for intelligent commerce. And I'm just kind of thinking out loud here. But are you doing anything else in the world of AI, in commerce or digital commerce or maybe even retail that's beyond what you've already laid out that sort of helps achieve that vision? And how do you actually see the job of people working in digital commerce right now, sort of changing and evolving to adapt to making use of the data and decision points that you're providing?

Jonathan: [00:33:25] Yeah, both good questions. Let me answer the second one first. I actually just read an article in Martek Adviser, which took on the question of is AI going to enhance or end my marketing career? So we could probably insert eCom there as well. Certainly, you know, there's a lot of fear that's spread by various analysts, by people like Elon Musk, AI is coming to take our jobs away. And we don't... It will happen in some areas for sure, because things that there are certain repetitive processes that can be done better, faster, cheaper and also learned faster by a machine. But in the area of commerce and marketing, digital marketing, we see AI as really allowing people in the field to focus on creativity, on ideas, on hypotheses, on the high value stuff and focus less on, in the case of Ascend, how to run a test. It allows the best ideas, whether that's what products to carry or what design changes or copy to try. It democratizes the process so that it's not who's paid highest. It gets to decide what to test. The system can test it all and do all those things. So we see these tools, both of them, as augmenting the roles of the people that are running eCommerce businesses or running the marketing for them and allowing them to achieve better results for their business and do stuff that's more satisfying in engaging in their careers. Now, as far as the intelligent commerce vision that we have, it's a pretty big one. It's you know, ultimately it's how do we deliver more personalization and engagement and relevancy to eCommerce and digital marketing in ways that drive not just better business for the companies, but also make better experiences for the consumers? I sometimes think about creating a more efficient marketplace. Efficiency is sometimes sounds like a cold word. It's not meant to be. It's like, if I can get what I'm looking for faster as a shopper when I want to do it quickly, then that's a good thing, right? Yeah, if you waste my time, that's a good thing. If I want to waste time, then you should let me. Similarly, if companies can stock the right products because they had better input... So one of the interesting benefits of aware is our customers that run it... Sketchers is another one, by the way. They expose a much higher percentage of their catalog than they ever did before because because eCommerce sites without Aware generally have pages and pages and pages of products. The ones at the top get seen. There's sort of standardized hierarchies. And in the case of Shoes.com, only 20 percent of their Canadian stock was shown in a given month. Shown. On the website.

Phillip: [00:36:43] Wow.

Jonathan: [00:36:44] And a portion of that was bought. And with us it was 93 percent. When you break down the one size fits all hierarchy and taxonomies that current eCommerce platforms have and allow people to find their own way, they will find their own way. And they'll see your products. Yeah. And then they'll buy the ones they like which may not be all of them but that informs a much better and smarter merchandizing plan for the next season. Right. All right. People saw it all. They had a chance to vote. Let's move in this direction. Those types of things. You'll see us move, in terms of additional products Aware and Ascend are both viewed as platforms and both of them will extend in terms of functionality, as I mentioned, where functionality is not being projected into emails. But you'll see it used for advertising, for retargeting. We love chat bots, but we think that if you're trying to sell someone something versus service them, that just talking to them isn't that great. So how do we use the the APIs of Aware to imbue chat bots with again, an experience that is much more like the knowledgeable store associate bringing you the next set of glasses that they think you're like. And then one of the things is analytics. So underlying this all you'll see us move into how do we take everything we've learned from both these products and other things we can learn and present them in a way that allow eCommerce operators and marketers to make even better and more informed decisions? The important ones.

Brian: [00:38:38] So as you're talking, I'm thinking, OK, you guys are actually going to be competing against some pretty big names with this. Like Adobe. The Adobe experienced manager in Commerce Cloud. Probably Salesforce's new Einstein Commerce Cloud Technology, and also maybe even IBM... {Cognitive} commerce. Could you maybe give sort of like...

Phillip: [00:39:09] {laughter} I feel like you said a swear word, Brian.

Brian: [00:39:15] Potentially.

Jonathan: [00:39:16] Do you mean cognitive commerce?

Phillip: [00:39:20] It's kind of a brand word.

Brian: [00:39:20] Like IBM brand word for sure. Yeah but I mean, like if you go look at Sentient's site, they kind of intelligent commerce, you know, similar to at least I sort of saw some similarities between what you guys are going after and what IBM is going after. Maybe you could kind of highlight some of the differences between what your vision is and where you play versus what IBM's doing with Watson, which is very all encompassing for sure. I mean, they've put a lot of different tools out underneath the Watson name.

Jonathan: [00:39:55] Yeah, I mean, I think the guy that runs the intelligent commerce unit formerly worked at Watson, did a lot of practicum there. So I think a cynical view of Watson might be that it's been more of a brand initiative and a way to position IBM. IBM is a great company. I would never suggest otherwise. But to give them an aura of mystique and a leadership position. Whereas, you know, I think that underneath that Watson label, they took a number of products, some of which were AI, some of which weren't AI, and they have put it into that framework. So I think, sure our visions overlap in sort of the end goal. How do we make commerce work better, more fluidly for both the retailer and the customer? But I think the ways in which we go about it are different. Watson is very focused on natural language processing, they certainly do other stuff as well. But, you know, if you look at their North Face, it's a great application. Yeah, we're trying to do the same thing. How do I find that winter jacket? But it's all about, you know, asking and answering questions, whereas our approach has been visually focused. How do you do a visual conversation? We do that. You know, why did we do that? There's a couple of reasons for that. One is, you know, our guys are involved in the early Siri and they kind of felt natural language has been done, so they chose to focus on something else. {laughter} But more to the point, you know, if you look at the you know, I have children that are in college. But the kids that are growing up, you know, their language is visual, right? It's not so much typing words in the way that it used to be.

Phillip: [00:41:50] That's true. Yeah.

Jonathan: [00:41:50] And on mobile, it's really hard to do. I mean, I guess people are pretty good at it, but it's not that fun. Right? And so trying to deploy conversational interfaces in a mobile setting by typing and chat as well seems limiting. But more to the point, it's like for the types of products that Aware focuses on, these are visual products. Typing doesn't work. Voicing it may not work either because different words mean different things to different people. But in terms of the competitive set, yeah, I think you're right. I mean, we look there's a lot of recommendation engines. There's a lot of visual search guides. There are chat bot companies. But, you know, I think while I can't say we've actually run into any of those bigger guys in any specific customer engagement. You know, I'm sure all our big clients do talk to them. And that is the playing ground where Sentient sees itself. How do we build a business that's worth, separate from our other business units, billions of dollars based on our ability to make things better for this most critical industry we serve?

Brian: [00:43:07] Yeah, definitely. I think one of the things that stood out to me, you know, as you first came on the show, and I think this is the question that I have for a lot of these AI companies is how much of it is essentially hands off?When I think AI I think of being, not a being, but something that can do things that I would do on its own, and that obviously there's different levels of this. But I think what everyone wants is someone that can go out there and actually make the trades and they can make those decisions and then actually make the purchasing decisions.

Jonathan: [00:43:55] Yeah. Or the merchandizing decision or the site design decision. So we focus very much on building autonomous tools. We want our hedge funds to trade on their own. Sentient Ascend, you put in the ideas, but once you do that, you hit start, and it figures in what order to test them and how to do it, and it determines the winner from in that case user input. In Aware, it knows what products to show and does that in real time for each user individually. So it's really important because otherwise it won't be efficient. By using machines, machine learning, AI to handle those kinds of processes, again, you can free up humans for higher value tasks. It's not that those things aren't a value, but for the creative input that informs the models that these systems are based on.

Brian: [00:44:53] Yeah. So it makes me kind of want to step back a little bit even and ask you some questions about AI in general. I mean, obviously, it's a really hot topic right now. I mean, at this point, Google has all but come out and said AI is the future. Other companies have pretty much done the same thing. I mean, heck, Zuckerberg is going to have a personal assistant now that Robert Downey Jr. wants the voice. Right? {laughter}

Phillip: [00:45:27] I mean, let's face it. We all want that right?

Jonathan: [00:45:31] We all want that. Yeah, for sure. But no, I mean, for our listeners, maybe you could... Usually we don't explainify on this show, and I don't want to get too far down this road. But you can just give us a little more about the state of AI right now. You know what I can do. What I can't do. Where you think it's headed. Yeah.

Jonathan: [00:45:51] Yes, I mean, that's kind of a tough question in some ways because, you know, the optimist in me was to basically believe that AI can ultimately do pretty much anything that is definable. I'll come back to your question and say, I think, look, what are the limits? Let's talk about what are the limitations. The limitations on what AI can do. One is the data and how organized the data is. But there's a lot of data. We have a lot of data, but it's not always organized in a way that makes it easily processed. Compute power, you know, obviously gets cheaper all the time. No one else is doing what we're doing. It's pretty hard to aggregate large amounts of compute power unless you're a really big company or you're doing it distributed like we are. And so there's a lot of AI companies out there, but a lot of their models aren't particularly deep or robust as they could be if they run on a larger scale. So I think that that's where they're held back. AI is still... There's still some things that we're searching for. We're working on this as well, which is how can we make the AI learn like people do? Unsupervised, without a frame of reference? Babies don't have a training set. Sure, there's a mom there saying this or that or providing input, but it's not the same. So that's what we would like to see AI get to. But it's not really there today in a big way. Now we talk about AI. There are many types of AI. There are many approaches to it. There's no revolution, this deep learning. There's all sorts of variants there in all of these things. So it's kind of important thing to remember. But look, here's the deal. AI is going to be and in many ways already is in everything we do. And in its best manifestations, you won't know it's there. Uber's got AI. Nest has AI. Siri of course has AI. I mean, all these you know, all these things that are in their own category as breakthroughs are using AI techniques, but it's in support of a larger goal. It's not about we're not trying to create a brain at Sentient. We're trying to create products that deliver value to people and to businesses. And that's where other people are trying to create artificial brains. And we'll see where they go. That will be interesting.

Phillip: [00:48:39] So I don't know, maybe I watch too many Hollywood movies. But isn't the frame of reference kind of the important part here? Because when you sort of take off the frame of reference and you have no context for these bots, like we saw Microsoft's Tay basically became a genocidal racist. And Stephen Hawking thinks that we're creating Skynet. I mean, I kind of keep hearing doom and gloom around the stuff. And on one side, you have people who say it's a panacea. You have the other side who say that, you know, if we use the Internet as training data, it's like they're getting the worst of humanity. It's probably a little bit of both, but I'm just not, I'm kind of conflicted myself on it has broad applications everywhere. But, you know, if we take context out of it, what are we creating? That's sort of the meta question, right?

Jonathan: [00:49:33] I mean, so unsupervised learning doesn't mean you take context out of it. It means that it can learn things on its own. To make the right decisions, I mean, so we think when we think about our systems, we think about essentially a loop. It's called the OODA loop. It's a military, I guess, framework or model. But OODA it stands for Observe, Orient, Decide, Act. And it just loops. You do that again and again and again. And if you think about the different AI technologies, deep learning is particularly good at observing and oriented perceptual issues. What is that? What is that picture? What is that word? And how does it relate to other words on the page or where is this picture of this building? Where is it on that map? And in evolution techniques and in neuro evolution and artificially. There's something called artificial life, a life, that isn't. But we were just talking about, you know, these are really good at making decisions, ultimate optimization tools. But the combination of these things is what allows you to do autonomy and what the military research shows... This is all about, which combat fighters who all had pretty much the same training, the same trainers. Some did much better than others. What made them successful wasn't always making the right decision, but the speed at which they iterated across that loop and learned through the process and AI is really good at that if you have the compute to run it at scale and its speed and you have the different technologies to handle those parts of it. So that's been our organizing framework. Look, I think is it a panacea? No. Is it doom and gloom? Certainly not. Does any of this have the potential to be misused? Absolutely. Starting with fire is usually the one they call out in the AI circles. It's like fire. It's pretty good. But, you know, you can burn things down a bit, too.

Phillip: [00:51:38] {laughter} Yeah.

Jonathan: [00:51:38] You know. And so I think where people get worried is like, is this technology so powerful with the machines can think can take over things and not have the right safety nets. But I do feel like the community as a whole, the big guys, and the smaller guys and people take this with great seriousness. There is a great commitment to how do we work together to ensure that these technologies are harnessed for the benefit of mankind. And personally while there is a chance... I think some jobs will not be needed to be done by humans in the future. But I don't believe that means that people are out of work. Historically, we find new ways for people to occupy their time. And if this really plays out and AI can deliver on the more efficient energy and more efficient transportation, the more efficient shopping and better relationships even with other people, we could see a very transformed society within a shorter time than you think. 30, 40, 50 years. You know, maybe not in my lifetime, maybe so though as it is always changing, but definitely my kids lifetime. I don't know if you follow Peter Diamandis at all, but he was also a speaker at this event I went to, Paradigm Shift, and he has a newsletter. And there is abundance all around us. The news distorts the reality. And AI is part of what's going to bring it all together exponentially in ways that benefit all of us.

Phillip: [00:53:20] So I do want to ask one follow up there just because you sort of touched on it. We have such... It took us a long time to get to a place where we had to make decisions in the realm of biology, of what ethical boundaries were in the realm of biology. Do you see a future where we have to create ethical boundaries for AI research?

Jonathan: [00:53:51] Yeah, I think that work is happening now. I mean, there was the consortium that was just formed by the Microsoft and Google and others, Facebook and others is not the first, but certainly the biggest such consortium. And, you know, that will involve we belong to several such groups. So the dialogs are happening. We have them internal to ourselves. What websites should be allowed to optimize using our technology? Should there be limits? Or should there not be limits? And there are great debates about that sort of stuff? But that's the trivial stuff, right? It's more like for example, we don't do government work. It's not so much to be moralistic, but sort of it is. Like we're not interested. There's so much to be done. We choose not to focus our energies in making better defense systems. Cybersecurity maybe. But not missile defense.

Brian: [00:54:58] Skynet. {laughter}

Jonathan: [00:54:59] That is where before you get concerned. Or War Games with Matthew Broderick, of course. We just have to ask the AI the right question and you can paralyze it. And then the whole thing comes crashing down.

Brian: [00:55:10] Right. Right. Well, you have shed so much light on this industry. It's very exciting what you're doing and where AI is headed. And we're coming up on the end of our time here. But we typically like to conclude our interviews with some practical advice for our listeners. So what are some practical ways that our listeners could apply AI to their business in the short term? And then what are some things that they should do to prepare for the next few years ahead?

Jonathan: [00:55:40] Yes, I think, you know, in the short term, look, there's some great products out there. Ours among them, right? Sentient Aware and Sentient Ascend. If you're a big retailer of visual products, Aware would be an interesting thing to check out and for anyone that's doing website stuff, look at Ascend. I think for customer service, ticketing apps, there's a company called Inbenta that's doing some great stuff. And really able to reduce costs there in your support centers. You should be playing with chat bots. I think chat bots are a little bit, you know, potentially poorly done. They can be cheesy. Yeah, but they're getting better all the time. And so now's a good time to play with them. Pandora Bots is a site to check out if you're interested in that stuff. And there's even AIs for site selection for you guys that have physical retail sites. Use is a company we ran across recently that's trying to use the data sets to understand, given what you're selling, where should you put your next store? So, you know, I think it's be open minded. You know, that's really my advice. Do not be put off by this. You know, the images that some people would like to project about what AI is. Because, look, AI it's a better set of algorithms. Right? They are just aster and they're self learning. So they continue to improve. But it's just using computers to try to simulate human processes. The article I wrote about is AI going to eliminate your marketing career? The answer is generally no, unless you put your head in the sand and don't choose to explore. So the key thing right now is, you know, keep an open mind, find out what's out there, play with it, and you'll see what you like and you'll see what you don't like. And then five years out, it's almost too hard to predict. But I think by that time, pretty much anyone that is leading in the technologies that they would use to run their business will be leveraging AI in one form or another. So I think that will just come to them. But if you don't take a look at it now, you may not be there five years from now to have that exploration. So it's time to pay attention.

Brian: [00:58:03] Great advice. Make sure you pay attention and educate yourself and be ready. I love it. Well, thanks again so much for coming on the show, Jonathan. For our listeners, please, again, would love to get feedback from you. So whatever means you want to provide feedback, Disqus comment box, Twitter, LinkedIn... Just talk with us, interact with us. And we know we'd love to hear more about your thoughts on the show and and specifically on today's show. Also, if you can hop on iTunes and give us a five star rating. We could always use that. But thanks again, Jonathan. And with that, keep looking towards the future.

Jonathan: [00:58:46] Indeed. Thanks, Brian. Thanks, Phil.

Phillip: [00:58:47] Thank you so much.

Jonathan: [00:58:48] To the future.

‍