AI in Action E179: Nils Lenke, VP and GM Apps BU at Cerence

Thanks for having me.

Sure, happy to. So I studied computer science and communication research in the 1980s and 90s, actually, because I wanted to do something with computers and communication, or say speech, language, and all that. And I did my PhD in AI in 1995. Then right after, I joined a company called Philips Dialog Systems, which was a Philips company obviously, and we built dialog systems using speech recognition for call center automation stuff like that. And then frankly, I've never switched job really. My company morphed, I got acquired and sold and spun out and renamed, but it's still, I've never really moved away. I've done many things, and now I run a business unit within Serence, which is now the company how it's called, and we deal with new revenue share opportunities, new apps that we implement in the car. And then speaking about Serence a little bit, Serence got spun out of a larger company that some of you may know called Nuance. We were spun out October 2019, so we are a little over a year old. We are now an independent NASDAQ-listed company with a headquarters in Burlington, close to Boston. And we built AI systems, conversational AI as we call it, for the photomobility space, mostly in automotive. So if you've ever spoken to your car, if you drive a BMW or a Mercedes or a Volkswagen or an Audi or a Ford or a Geely or whatever it might be, there's a very good chance you've spoken to one of our systems. So we have put speech and other technologies into more than 300 million cars historically, and last year we shipped roughly 50 million cars, so one out of two cars shipped has some Serence technology in it.

That's right. So I've been in BS and in R&D and then product management and all that, and it's been very exciting because it's always been about letting people interact with machines. But for many years we called that just machine learning and stuff like that, and then say in 2010, 2011 we started using neural networks for doing what we are doing. And today we are using neural networks of course in all our technologies, but back then we switched it over step by step from Hidden Markov Models to neural networks and all that. And then it started to be called AI again, right? I started AI in the 1990s when it was symbolic AI, and then machine learning was very different. But today it's the same thing. Many people use AI and deep learning nearly as synonyms today. And so the whole field has changed a lot, and also the expectations. I mean back then it was kind of pioneering to put speech recognition into call centers, but today people really expect they can speak to their phones, they can speak to the television, they can speak to their cars. Having an interaction with a machine is now really normal and nothing really exotic anymore. And that's really what has changed the whole thing, it's being from pioneering isolated solution to something that's a new normal today.

That's totally right. And we've enjoyed that from early on in the automotive space because speech makes so much sense, right? You're driving, you should focus on driving, but you still want to do stuff. And in earlier times it was just operating the navigation system or the radio. Today, people while driving want to stay connected, they want to be in touch with other people through social media, they want to send messages, they want to get information from the internet, they want to purchase stuff. We are building things we call Smart Domains where you can search for restaurants, search for gas stations, make a purchase, you can play games and all that in your car, and speech is just a natural way of doing that. So that helped us a lot early on. And today speech is everywhere in the living room and on your phone and everywhere. And you need to make it consistent, it needs to be a consistent experience. And that's what we are building. We are building the speech experience in the car that naturally also integrates third-party assistants elsewhere, but it's still an OEM-branded experience. Maybe I should mention that you may not have heard about Serence because what we build, we build it with and for the OEM, so they build an experience in their cars. It's powered by Serence, but it's branded by the OEMs.

Absolutely. And we see people actually need this because it's unsafe to do stuff on your smartphone while driving, but many people can't live without being connected for an hour or two. And so we connect them from the car to their systems at home, we connect them to other systems, we connect them to the internet and to all the information they need. So that's really the whole idea.

It's a good question. And let me start with saying it's not true for us. I mean, as I said, we use data science or AI, deep learning, whatever you want to name it. We do that at two levels. We do it at the core technology level. So we build speech recognition, we build natural language understanding technology, we build voice synthesis or text-to-speech, we do voice biometrics. And for all of that, we use neural networks these days, so it's all data-driven. And we have done that for so many years that for us it's an evolution. We don't sit down and say 'hey, a totally new technology'. For us it's an evolution of things we have done for a long time. And then we do the same at the more conversational level. So you're having a conversation with your car, and the car learns your preferences. I mentioned we do Smart Domains, like finding fuel stations while you're driving. The system will learn that you prefer a brand over the other, and that's of course also using AI technologies. And I think the trick here is, and it brings us back to an earlier question, the whole question of saying 'I have a data science project and now it fails' -- it's turning it backwards because you start with data science as a solution and then you look for a problem to solve. But in reality you should start with a problem, and then data science comes in handy to solve that problem. And then you don't work on the wrong problem, and you don't build something that you need to look for a problem for. If you start from the problem, you have a need for that solution, and it's less likely to fail in my opinion.

I think the two questions actually are tied into each other because if you have a problem to solve, you can't just look at the data science problem. You need to look at the organization around it, at the processes, and also see what are the processes I'm going to change, how is this impacting the overall organization, how does control and data and information flow today and how will it be in the future? Imagine you replace a business system by something that's data-driven. Well, you already have data flowing today, and you will have a new flow tomorrow, and it will impact people. It will maybe cost some people their job, it will create new jobs and opportunities for others, it will make jobs more interesting for people. You need to look at the entire picture, and then it becomes evident where you get the data from and who the people are who should help you. Tapping into data is very important. Again, for us it's quite simple. Our technology used to be embedded only in the past. Today we built a hybrid system with the cloud component, so cars are connected these days. And that means a lot of the interaction between the people in the cars -- we get logs of that that we can use for specific purposes, complying with all the rules like GDPR and all that, of course. For specific purposes, we have access to that data. But yes, you're right, in other circumstances it may be a hard job to get the data in the first place. Again, if you try to steal a job from somebody, they might not be helpful getting you the data you need.

It's a good question. And some trends we're seeing happening already. So one trend is clearly connectivity. So some years ago there were hardly any connected cars. Today, most cars hitting the road are connected. So telematics services, being connected to the internet, being able to deploy new services in cars. I mean a car used to be a fairly static object. You would buy a car with a certain infotainment system, it would still be the same after five or six or ten years. That's totally different today. You have the large screens, you have the software running in the head unit, in the infotainment system, with speech and other services, and you can update them. You can update the cloud service behind, you can over-the-air update the components in the car, and keep the infotainment system fresh. That's what Daimler does. They recently launched the S-Class, and there's Cerence technology in there, and it will not stay the same. They launched MBUX with the A-Class a couple years ago, and they're updating it all the time. So systems evolve and cars stay fresh, you know. So infotainment is clearly in itself a huge trend because some years ago you would sell a car by looking at horsepower and complicated braking systems and stuff like that. But today, when you're moving to electric cars, the infotainment component gets a lot more exciting than the technical components inside the car. And then finally, people want to stay connected, so they want to understand is their heating at home working okay, they want to open the garage door while driving home, they want to be connected with friends and family, they want to purchase stuff. They may want to order a restaurant in advance, either a table or a pickup in these days before they get there. They want to pay for fuel in a contactless way without having to enter the gas station. So in-car commerce is another trend we see.

That's true. So there are competing trends here. For environmental reasons and keeping cities also a place you want to live in, there is of course a trend to tokenized shared rides, electrical buses. And by the way, that's something we also deal with. So we showed a nice concept there last year at CES where you could interact with an electrical shared bus from the outside and inside by voice and some fancy displaying technology. So we don't only deal with cars, we also deal with two-wheelers, e-scooters, etc. So there are new trends of mobility. But you're absolutely right, COVID has created a counter trend which is, let's get back into my private car because it's safer. And we see this in tourism also, by the way, building a product that helps you enjoy visiting a city with your car because it tells you about it. And so that's that trend. But if you want to do this with a good conscience, you may want to acquire an electric car, an EV, and then you can do it without thinking about the pollution you're creating and the climate crisis. You know, you can still do it.

I'd agree. I mean, it's really exciting. The acceleration and innovation, the new players entering the field, what people expect of their cars, I think it's unprecedented how quickly things change in the automotive industry. And it's all thinking of let's think in three, five, four year production cycles no longer exists.