Jan Scholz: Glad to be here.

Dave MacDonald: Glad to be here.

Jennifer: We're glad to have you here. Why don't we start off a little bit, and maybe I'll start with you, Jan, just tell me a little bit about your background and tell me about how you came to Rangle and what are some of the things you're working on here at Rangle?

Jan Scholz: Yeah, I joined Rangle here as Director of Artificial Intelligence. I'm trying to build up the AI team here and really teaching all people at Rangle about AI and letting them participate in it. Originally I come from academia, like many data scientists out there, where I studied neuroscience and applied Machine Learning techniques to medical imaging data and that's where I picked up that skill. But what I realized in academia, it's really hard to bring projects to conclusion because people are self-taught. They don't have that industry context of running through projects in an Agile way. And that's when I changed to industry to learn all about developing successful projects.

Jennifer: So a little bit of background then. How about you, Dave?

Dave MacDonald: My background is from industry. I'm here with Rangle as a Senior Solution Architect. A couple of years ago I started a group called Toronto AI. It's just a meetup group. We have about 4,400 members at this point and we do these workshops. Like how it got started was through workshops, teaching TensorFlow to small groups and then to progressively larger and larger groups. Many of those events have been hosted here at Rangle. They have a great Machine Learning team present here at Rangle.

Jennifer: So listen, there's a lot of information ... And I'm going to say it ... Excitement out there about Machine Learning and AI, but I'm going to ask the million dollar question. What is Machine Learning? What is Artificial Intelligence? Are they the same exact thing? Talk to me and demystify for me what exactly these are and maybe Dave, we'll start with you.

Dave MacDonald: Okay. So we might disagree slightly on this-

Jennifer: And that's okay. That's okay. Yeah.

Dave MacDonald: My understanding with Machine Learning is that it encompasses a wide range of techniques for training models to approximate an objective. So that objective could be many different things, with Machine Learning the idea is that you define a task and then optimize towards that using experience based on data. There's many approaches to doing that. More recently there have been neural nets. They've been around for a long time, but it's only recently that they've been used so extensively since I would say probably around the early part of this decade was when they really started to take off.

Jennifer: And talk to me, what is a neural net? Just in case there are people there that are looking for that sort of deep understanding.

Dave MacDonald: Yeah. So a neural net is essentially layers of weights. It's a feed-forward directed acyclic graph that basically you feed data in one side and it goes through a series of transformations to the other side. And so ideally you want to send in data that represents the information about the problem that you're trying to solve. And at the other side you have output units that describe the solution that you want. And then you can take that through an iterative process of what's called backpropagation to feedback air through the model to correct all of those weights. And there can be millions and millions of these weights within those layers to create a better approximation of that function that you're trying to model.

Jennifer: So if I was going to use an everyday example of that, you could say that if I owned a restaurant and I was using AI to help me with ordering products or reducing my inventory of things going bad is I could say here's all of the people that are coming in every single week and the times that they come in and the things that they order and how much their bill was and on the outside of it, I want to have a fundamental understanding of how much inventory or raw food product I need to have on a weekly basis based on that learning over time. Then that's how this could operate and solve that problem.

Dave MacDonald: Yeah. That's actually a great use-case for neural nets, that's getting into the area of forecasting. So often what you want to do is to extrapolate from existing data. If you have historical information, you can use that. So given past weeks of information you can see what happens the following week and then use that to build a model for forecasting. So you would make it conditional on the information that you have available.

Jennifer: Which is pretty exceptional. That's pretty awesome, right?

Dave MacDonald: Yes, definitely.

Jennifer: So Jan, maybe a different opinion? Similar?

Jan Scholz: Yeah, great question. Currently people in the business world throw around those terms interchangeably. AI, Artificial Intelligence, Machine Learning, deep learning, they don't care. I think now in the industry people settled on calling everything AI and there's sort of an insider joke that even statistics like linear regression is now AI, which it really isn't. But people call everything that is somehow remotely smart AI now and to some degree we have to talk about topics in that way. But fundamentally if you look at it, I think Artificial Intelligence is an umbrella term that includes Machine Learning, but also robotics, expert systems and other things as well. And then in Machine Learning, the different Machine Learning techniques, one of which is to do with neural networks called deep learning, but there are also things like random forest models that are not deep learning but they fall under Machine Learning.

Jennifer: Which we do.

Jan Scholz: But in the industry AI is probably the term that is used for most of it now. What is a neural network? Just quickly, it's what's in your head when you learn something. So if you think about learning how to ride a bicycle, you get on the bicycle and you fall over to the right side, next time you try to avoid that by steering the bicycle to the right. And that's exactly how a neural network works. You have a difference between what you expect, you want to ride on your bicycle, and what is actually happening, you fall over, and in your head you adjust your motor response, your arm movement, and your weight on the bike accordingly to avoid that mistake. And that's exactly what the fundamentals of training neural networks.

Jennifer: So, million dollar question then, why Machine Learning? What is the case scenario for businesses to invest? Because let's be clear, it's an often onerous and sort of expensive thing to invest in. So why are businesses investing in this? What is the use case scenario for Machine Learning or Artificial Intelligence in businesses using it to drive their business? And Dave, maybe we'll start with you.

Dave MacDonald: Sure. So one of the areas that I find it can be pretty valuable for businesses is to be able to provide personalized experiences. An understanding of, say, items to recommend on a page to a particular user based off of information that is available from prior interactions with that user. That can go a long way to creating a better overall experience. By showing them better products, it's more likely that they'll get to the shopping cart. So that's an area that can be very relevant in terms of being able to determine what type of inventory to stock shelves. Again, that's getting into forecasting, predicting demands. All of that ties into AI or ML-based techniques.

Jennifer: So we're talking about a fundamental reduction in overall costs or inventory carrying charges or out of stocks or markdowns as an example for retail if we're talking about that, but also more so that if I think about some of the work that Rangle is doing with healthcare as an example and fundamentally understanding the needs of patients as an example and what they require across geos or across different diseases, etcetera, that I know that Machine Learning is helping some of those industries in using predictive technology to create solutions. Maybe you could talk to me a little bit about that, Jan.

Jan Scholz: Yeah. I think there's a bit of a misconception that we think of AI as an expensive or difficult problem, or a difficult thing to implement. Here at Rangle we think about Artificial Intelligence in terms of smart experiences and automating things and it's really not that much different from any kind of the other product development we do here. For example, front ends. We want to identify the pain point that is in the industry, for example, in healthcare or in retail. And we are using all our skills here at Rangle to build the right solution, part of which could be AI and we're using AI to provide a better user experience.

Jennifer: So something that you've both alluded to in your answers is the integral relationship to data as part of this. So talk to me about the relationship of data to Machine Learning and what fuel a machine needs in order to give the right outputs, as you said earlier.

Jan Scholz: Again, a misconception that we want to dissuade here. A lot of organizations that want to invest into AI get stuck at the first step. Is they asking themselves, "Do we have enough data? Do we have clean data?" And until they can answer that question with yes, they are not going to do the AI part. And I think that's turning things upside down because, again, it's important who looks at the AI in the end. Is it useful? Is that output useful? And we want to get to that point very quickly so we get the feedback from the users in the end. So here at Rangle we are trying to come up with solutions to maybe simulate data or get data from a public data source or get third party data or use whatever little data is available to start the Machine Learning modeling process and then get that feedback from the user of that Machine Learning model very quickly. And that will help us to inform actually the data acquisition process.

Jennifer: So that brings us to the other side of the spectrum, which is all of these major institutions, like healthcare institutions or banks as an example, or even retailers or legacy companies, they have this huge historical lake of data. Is there any sort of requirements to implementing Machine Learning or AI within an organization? How do you bridge that gap between the old data and the new data and is there any requirements from companies to make that step or is that something that you're working with companies to solve?

Jan Scholz: Yeah, it's a great question because a lot of larger organizations are sitting on a huge pile of historic data and they feel there's some value in there and that there should be a way to monetize or make that data available for the business. At the same time they are continuing to collect more and more data through their business processes. I think, again, there are two worlds here. There's a big data AI world that looks at large historic datasets such as customer data, sales data, supply chain data and then uses that data to forecast certain KPIs into the future, predict churn, predict cross sell or upsell opportunities and so on, and so on.

Jennifer: That's a really interesting thing. So let's say the business has now said, "Okay, we need to invest in this, this propensity model that we want to do or the modeling of our business that we want to do moving forward." They would either invest in themselves or partner with someone like Rangle in order to make this happen. So what is the key considerations for them at the outset on whether one, they're going to do it themselves or partner with experts? What are those key considerations that they are thinking about or is there any consideration or is it just the, "Yeah, get started today and here's how?"

Jan Scholz: Organizations currently, despite the hype out there in the media, there's no day where you can't see an advert on TV about some new AI solution by large companies and there are new use-cases coming out every day. But despite all the hype what is actually made in reality, the AI solutions that are being developed right now, the market is quite slow still, because fundamentally organizations haven't understood yet the value that AI can provide and they haven't understood a process of building these solutions. Again, that's what I mentioned earlier, we have to think in terms of product development about AI. A lot of companies hire dedicated data scientist teams from academia. These people don't have a background, usually don't have a background in project management or in design thinking and all these things that help you create good products.

Jennifer: So invest for the purpose of utilizing, right? So instead of just investing for the purpose of collecting more data and providing additional insights, it's, "Well, how are you going to action those insights?" Right? So you mentioned earlier, Dave, that you were a part of a meetup and Jan I'm guessing that you have some peer networks locally as well. So talk to me about some of the projects that they're excited about or the things that they're working on or the way that you show you see Machine Learning in practice right now, in not just this community but even globally.

Jan Scholz: Yeah, great question, Jen. There's a lot of startups that are centered around AI and there's some really creative use of using AI. I remember in one of your meetups, Dave, there was a company that presented a use of AI in a more creative generative manner. Do you remember that, where they design logos and business cards?

Dave MacDonald: Yeah, Andrew Martin I think was hosting that. That was using a generative model for creating, I believe it was fonts.

Jan Scholz: Yep.

Dave MacDonald: So that's an interesting area in terms of generative AI. That's an area that I've been kind of fond of myself and we see that in a number of areas in terms of being able to generate, whether it's music, speech, video, those are all different modalities that you can train models to create in very realistic ways.

Jan Scholz: So think of you go to that company's website and you click a button and you get your own personalized, unique business card with its own font with its own logo. That would have taken a couple of hundred dollars to design previously if you had asked a human. Now you can design a million variations on your business card for free. Almost.

Jennifer: And you alluded to something around music. Talk to me a little bit more about that, that's interesting.

Dave MacDonald: So in terms of music synthesis, MuseNet is a model that was recently released by OpenAI and I would check it out. They have a soundcloud, they've been using that to essentially create compositional pieces-

Jan Scholz: Just think of being able to explore the space of music with a click of your mouse. So you could say, you can ask the AI to design you a classical version of a death metal song. What would that sound like?

Jennifer: Iron Maiden in the key of Bach.

Jan Scholz: Exactly. Some of them are actually quite compelling and it's good to see that, traditionally we thought of AI automating the boring tedious tasks, but now it's creeping closer to what we thought is exclusively the human domain, the creative domain. So we saw that with music, there's also interesting AI models that translate between different domains. So you could say, for example, write a script and then the AI would create a movie corresponding to that script. We're going to see a lot more and I think that's an interesting development.

Jennifer: For sure. Yeah. And that's I think where it gets really interesting for someone like me who is in marketing as part of my career, which is it's not just about personalizing for customers and experiences better, but it is also, I came from the book industry and I read an article recently that they took all of the novels that were put into the ether as just script and they said, "Okay, write the perfect novel based on the books that most everybody reads." And essentially they came up with like the key plot lines. You can imagine where that went, right?

Dave MacDonald: So I would say e-commerce is definitely well positioned for benefiting from AI in terms of being able to access better product recommendations and more personalized user experiences. That translates just directly into more revenue for these clients. Also A-B tests are going further than that into multi-armed bandit testing and further into reinforcement learning and AI. That's definitely an area that can be used to determine what to show a user that's visiting a site, and to learn from the population what is most effective. So on one hand there's personalization. On the other hand it's like, "Okay, well we could present the site in one of many different ways, which approach results in the best conversion? What maximizes what ends up in the checkout cart," right? So those are all optimization problems that e-commerce can definitely benefit from.

Jennifer: And if we're going to talk about e-commerce, we're also talking about the consumer. Every now and again I like to throw something in my cart on Amazon that is just like completely out there, like a harmonica on a necklace or something. But what I do appreciate is with a lot of clutter out there, digital clutter, the curation is a benefit to me. If I'm shopping at a retailer and you know that I've been there three times in the last three weeks and you know exactly what I bought, could you surface some contextual suggestions that all the other people who buy these kinds of things would also like? So aside from e-commerce, what other interests, Jan, do you see really benefiting from this in the short term?

Jan Scholz: That's a topic we have discussed internally a lot. What industries should we talk to? Who has a need and where's the success here? And I think the best way of answering it is everybody can benefit from AI. I think AI is not particularly good for a particular industry, it just can solve so many problems. So I would answer that question with the organizations that are fast and nimble can profit the most. So again, we have digital native companies like Google and Amazon that have already invested heavily in AI, but it's really the organizational structure of being able to build hundreds of models and run them and integrate them in their business that helps them to take the advantage, it's not the particular industry they're in. We see basically a split here. Small startups are very good at leveraging AI because they can build from the ground up the infrastructure and the organization around AI. And we have large organizations that have the capital to invest into AI, but they haven't come around yet as an organization to produce AIs.

Jennifer: Got you. So the clients that you're currently working with as part of the AI team here, or Machine Learning team, talk to me about that onboarding process, those initial conversations that you're having. Because I guess what I'm trying to get here is, someone out there is listening to this conversation and they're saying to themselves, "Okay, this is really interesting. I definitely want to pursue this." So in those initial conversations that you're having with businesses, what is a part of that onboarding process? What is required of them as part of that?

Jan Scholz: That's a great question. Having been in the industry for a while, usually the clients are hesitant. That's the first thing. So to people out there who are interested, don't worry, there is a process out there that can help you get through this. So the first question that clients have is do I have enough data? Is my data clean? And so on. And as we've talked about it before, don't worry about it. We have ways to have third-party data, we have ways to simulate data. We can get around that. The next step is what is your pain point? Again, we have techniques like value canvasing, design thinking to help you identify that pain point and that is really a business pain point and not a AI pain point. Because we want to build something that is ultimately of value to the organization.

Jennifer: Well, that is the promise.

Jan Scholz: Exactly. Yeah, that might be the blue sky dream at this point, but don't try to do this as a first step. As a first step find a little use-case. Work on that. Get your data in order. Get your organization around. You know, if you're in a legacy industry, you probably have an IT department that sits on the data and then you have data scientists with the IT department about the data and so on. So there are a lot of organizational things that have to be solved as well. But with that roadmap we can give confidence that you actually can achieve your goal in the future. And the beauty of it, all the rest will fall in place once you have solved these problems first.

Jennifer: So listen, you guys have been in this space for a while now, early adopters to the extreme. Dave, what are you excited about? What are you excited about in this space that you can see coming? Or like what can you not wait to see happen here?

Dave MacDonald: I just attended the ICLR conference in New Orleans and so that was a bit of a glimpse at the future. In terms of being able to see intelligent agents able to have a very solid understanding of ... Salesforce is doing some excellent work in this space, actually. They have an AI team that's been working on developing state of the art systems for understanding of documents. It's known as question answering. So the idea is that you can take a piece of text off the Internet then ask a question about it and have the AI provide the response. That was really fun to play with. Also getting into like generative modeling. I find that to be very fascinating for industries that are more creatively oriented AI is going to play a big part going forward in the future in terms of being able to assist artists to develop their work. Creative studios definitely could benefit from AI.

Jennifer: And how is AI benefiting that community specifically?

Dave MacDonald: Being able to generate a new material at effectively no cost ... Let me rephrase that. Being able to generate novel, new material that is within a particular style, that can be very useful in terms of being able to explore ideas, creating starting points for developing new creative content. That's definitely one area. And then in terms of reinforcement learning as well, there's been tons of work in this space basically to make these agents more and more capable, to be able to learn off less and less data so that eventually will tie into robotics. But that's looking far into the future at this point.

Jennifer: But far into the future these days means like 18 months from now, at least in the way that we think about things in the tech community, right? Just really quickly. What about you, Jan? What are you excited about?

Jan Scholz: Well, it's hard to make predictions, especially when they concern the future. We don't really know what the impact of AI will be. I think there will be some surprises and whatever great AI will come out, I'm pretty sure we will get used to it very quickly and it will be boring. But what I see is quality of life. I think there's a huge potential, and we're talking a lot about technology here and maybe some edge cases here, but really how does AI change the way we lead our daily lives? What do we do everyday? We get out of bed and then we have to put on the coffee machine. An AI could put on the coffee machine for us at the right time. Then we get into our car. We are still driving a car, you're circling around a block to find a parking space. AI could do that for us, that would help enormously.

Jennifer: Right. They'll still have to change diapers like we do though.

Jan Scholz: Probably.

Jennifer: Thank you so much to Dave and Jan for joining me today. It has been a pleasure talking to you, I really, really appreciate your insights, and thank you to all of you for joining us. As always, we love bringing you these insights on What's Your Rangle. We'd love to get your feedback on what we've talked about today and we'll certainly provide some contact information so you can connect with Jan and Dave as well after the show. Thanks so much for joining us, we hope you learned something just like we did today. Thank you again, and please do like and subscribe and share with your friends. Thank you.