Artificial Intelligence - Episode 2
Host: Bradley Metrock (CEO, Score Publishing)
Guest: Rand Hindi (Founder and CEO, Snips.AI)
Duration: 18 minutes, 11 seconds
In this delayed episode from September 2017, host Bradley Metrock (CEO, Score Publishing, and host of VoiceFirst.FM podcasts "This Week In Voice" and "The VoiceFirst Roundtable") interviews Rand Hindi, CEO and founder of Snips (Snips.AI). This wide-ranging discussion covers Snips' technology and core product, Hindi's own prodigal background as a programmer and entrepreneur, and more.
Bradley Metrock: [00:00:14] Hi. And welcome back to Artificial Intelligence, Episode 2, for September 2017. My name is Bradley Metrock - I'm CEO of a company called Score Publishing, based here in Nashville, Tennessee.
Bradley Metrock: [00:00:28] My guest today is Rand Hindi of Snips. Rand, say hello!
Rand Hindi: [00:00:33] Hi! How's it going, Brad?
Bradley Metrock: [00:00:35] It's going great, Rand. Thank you very much for joining us today. I'm fascinated by you and your company, and am looking forward to chatting with you.
Bradley Metrock: [00:00:43] Let's just get straight into it: let's start by giving us a brief background of yourself. You've got a fascinating background - share with some of that. And then also please share with us the elevator pitch for Snips.
Rand Hindi: [00:00:54] I was lucky that my parents very soon put me into coding. So I've been coding since I'm 10 years old. I did my first startup when I was 14 - it was a social network back in '99. It was quite interesting at the time to build websites, by the way. I then had a web agency before going to London to study computer science, eventually getting a PhD in machine learning applied to biology. And as I was doing this, it kind of struck me that there was a lot of different things you could do with AI. And one thing in particular caught my attention...
Rand Hindi: [00:01:31] When you think about what's happening today between men and machines, what you realize is that humans need to learn how to use every machine. What that means is that the more machines you want to use the more effort you need to make to use those machines. So the more technology you want to use the more effort it takes to use technology to a point where everybody currently feels completely saturated. So I think that AI can be a solution to that because it can make all of those machines around us able to communicate with us like we do with other humans using language - which is not something you have to learn. It doesn't matter how your coffee machine works, if you can just talk to it and ask it whatever you want. You don't have to learn to use it. When you do that, something interesting happens. The first thing, technology becomes easily accessible to everyone. Anybody who can speak can use technology.
Rand Hindi: [00:02:22] And the second thing is it becomes so intuitive and easy to use that you completely forget it exists. Basically you can think of AI as a way to make technology disappear. So that's the vision we have for our company Snips. We want to put an AI assistant in every device on the planet to make technology disappear.
Bradley Metrock: [00:02:45] Rand, that sounds great. What does that mean for you and Snips in practice?
Rand Hindi: [00:02:52] So this is a vision for our company, right? But what we're offering today as a first product is a voice platform for connected devices. So basically if you're a Maker building on a Raspberry Pi or if you're a company building like a super-advanced device or whatever you're doing really and you want put voice in it, we offer a technology to do that....everything from customizable hot words to speech recognition to natural language understanding, the entire stack necessary for your voice assistants and your products is what we offer.
Bradley Metrock: [00:03:23] I want to ask you about privacy. Privacy is a differentiator for Snips and it's something that is part of your overall angle on the market. Talk to me and the audience about the importance of privacy with voice and AI and how Snips approaches that.
Rand Hindi: [00:03:45] What's important to keep in mind when you're talking about voice in particular is that voice is considered to be very personal, sensitive data.
Rand Hindi: [00:03:52] It is biometric. You can be identified uniquely by your voice and you cannot change it. It's kind of like your fingerprints. You know you wouldn't really think of sending your fingerprints to people in the cloud, right? So why would you do that with your voice? So for us privacy was never like a nice-to-have. It was always part of the equation of building assistants that people could trust. And so the way that we're doing this, and something that we're really proud of, because as far as I can tell we're the only one to be able to do it that way, instead of processing the voice in the cloud, we're actually doing everything on the device that you're interacting with that you're speaking to. Meaning that your voice never ever leaves the actual device that you're using. And so that gives you a few very interesting properties. It guarantees privacy by design. Nobody can hear your voice or access your voice data but you. It also means that you have super low latencies because you don't need network access....so your assistant can reply a lot faster. And it also works completely offline. So what that means is that even if you don't have an internet connectivity, maybe it's like your washing machine in your basement in your house or something. Your voice assistant will work.
Bradley Metrock: [00:05:07] That's cool. So what challenges have you run into in implementing your voice assistant that way where it's all hosted locally on the device versus in the cloud.
Bradley Metrock: [00:05:20] Can you share perhaps one or two challenges that have presented themselves?
Rand Hindi: [00:05:26] It's actually very hard to kind of like cram down multiple deep learning algorithms and models into the equivalent of like a Raspberry Pi. The challenge is first the size of the models. I mean when you think about it you know deep learning and all the machine learning tools were built with the assumption that you had infinite power in the clouds. So whenever we tried to do this on a few hundreds of megabytes of RAM with like an 10-year-old CPU just like single call 1 gigahertz. Very quickly what you realize is that it doesn't work. So you need to think about how to build smaller models. So the training data, the architecture of your deep nets, all these things are important to optimize. But you also need to rebuild all the tools. Things like SensorFlow did not run on a Raspberry Pi. So we had to find ways to tweak it so that it could run on it. So the challenge is both in the machine learning side to build small models that still have the power and performance of much bigger networks and an engineering challenge, where you got to cram this down, trying to squeeze in every single computation that you can so they can fit on the smallest hardware possible.
Bradley Metrock: [00:06:48] Let me ask you something that sort of intrigues me about you yourself and you alluded to it earlier. You started coding at 10, am I right?
Rand Hindi: [00:06:58] Yep that's right.
Bradley Metrock: [00:07:00] And I'm looking at your LinkedIn profile now. You founded a social network at 14, web agency at 15, started your Ph.D. at 21, Forbes 30 Under 30. You've been pretty busy.
Rand Hindi: [00:07:16] I've been lucky that my parents put me into that at a young age. I mean I'm 32 today. If you've been coding for 22 years you know I mean you've done stuff right. It's just that had a lot of time to do stuff.
Bradley Metrock: [00:07:28] Do you advocate for youth following a similar path that you have in learning to code at such an early age and getting headfirst into technology? Or do you advocate, given that you've been doing this for 20 years now, a different path? Share with me some of your personal background and if you would do things the same way or what you tell your kids that you deal with.
Rand Hindi: [00:07:55] It's actually quite interesting because my view on that has changed quite a bit over the past couple of years. I used to believe that everybody should learn to code at a young age - at least everybody who was interested in science. Just like you learn mathematics you should learn basics with coding because that gives you computer literacy. You can understand the world around you which is more and more computer based. But the more I think about it, the more I feel actual coding isn't as interesting as designing software. So basically the coding piece is really going to become simpler and simpler, maybe even automated but conceptualizing the software, figuring out how you're going to use it, figuring out what it does, figuring out the architecture, like systems thinking of software development I think is really where it's interesting. And so rather than teach people how to code you should teach people how to build software even when that means not coding anymore.
Bradley Metrock: [00:08:55] Yeah. It's fascinating to hear you say that. And I completely agree with you I think there's way too much emphasis on coding right now, just to be frank. And with some of the shows that we do with VoiceFirst FM....you know This Week In Voice, this show Artificial Intelligence, and some of our others, it's clear that liberal arts backgrounds are becoming more and more important in this space to designing. You still need people with technical skills. Don't get me wrong. I need people who have gotten their hands dirty. But it's a much more diverse group of talent that seems like voice technology and machine learning and AI need, so it's interesting to hear your answer to that question.
Rand Hindi: [00:09:42] No, I 100% agree. I think as we're progressing with AI, we're realizing that we want to make it more and more human. And so you know things like character designer, what precisely you want to build into your voice assistant, things such as ethics, sociology....all of these things are coming back into play because we want to make sure that if we live in a world where humans and machines co-exist, well, that it kind of doesn't suck, right? I mean that's the bottom line.
Bradley Metrock: [00:10:14] Yeah. You're right. Let me shift gears and ask you since yesterday we witnessed Amazon just absolutely unload on the marketplace, in the event that they did yesterday with numerous new hardware devices and a lot of interesting revelations. From your standpoint and what Snips is doing in your background, what caught your eye in this deluge of stuff that Amazon gave us yesterday? What, if anything, caught your eye? And give me your perspective on Amazon's approach specifically toward this market.
Rand Hindi: [00:10:52] I mean we definitely have to thank Amazon for having revived the voice market. I mean voice isn't new right? People have been thinking about this for years and years and years. The concept of voice interfaces has been theorized and taught of since the 50s, but it never really works. Amazon was the first company that used modern voice technology and put that into a consumer product that had an experience that was very nice essentially. So I think we can thank them for that. Without the Amazon Echo, the whole voice Echo system today wouldn't be that big. Having said that, Amazon has a different motive than other companies might. The goal Amazon has is to be the platform you're computing on. Amazon doesn't have a mobile operating system. So capturing the voice platform on top of which everybody is going to be doing transactions gives them a very unique position compared to Apple or Google....So I think that's a very different thing than let's say a car company who wants to put a voice assistant in the car in which case, yes, the assistant is a lot more verticalized to the actual device and product. Whereas Amazon really has this platform that it sees as you know the way to integrate every single business you kind of want to access by voice and yourself.
Rand Hindi: [00:12:23] I think this is really sort of like you know the end game is everything you're going to ask to a voice assistant you should ask to Alexa. And so what we've seen yesterday, is one step forward in the strategy of creating as many touch points as possible in as many commodity things as possible. All funneled through one consumer experience which is the Alexa voice assistant. So what we do at Snips is a bit different. What we do at Snips we sell a white label voice technology. We don't actually put a Hey Snips in every one of our customers devices. They can pick whatever we want as a name. They can pick whatever they want as you know different priorities and functionalities we're really focused on. Here's a company, they want a voice assistant for their coffee machine, their TV, their car, their whatever....we're going to sell the technology to do that. So it's kind of like a different position. It's the same type of technology while they're in the Cloud One device but very different business positioning.
Bradley Metrock: [00:13:25] Well sure. And I do think that Amazon spamming the marketplace almost with different types of hardware can't do anything but help a company like yours because, for example, the Echo Buttons. Did you see the Echo Buttons? They kind of got made fun of....did you see that? I don't think they should be made fun of it all, I think somebody creative will find something interesting to do with those. But I digress. The more that Amazon mainstream hardware gets out there I think the more it just educates different little pockets of the marketplace and the more and more the marketplace gets educated the more customers are going to be looking for something like what you're doing. That Alexa doesn't cut it, and they don't want that mainstream option. They need something more customized, something more private.
Rand Hindi: [00:14:17] Exactly. It's a different positioning. I think companies who work with us also work with Alexa very often but for different purposes, for different parts of the product. What we guarantee is privacy by design, compliance with the upcoming European regulation on privacy, which is a major deal, by the way and we haven't talked about it but this is a major deal for AI. We offer offline capabilities. And importantly we give you complete control over the experience and the data and everything from your user. So what we see a lot happening are hybrid systems where you might do an integration with the Alexa ecosystem. So maybe a scale or something like that for your coffee machine but you'll still want your coffee machine to be voice controlled without an Amazon device in your house. And I think this is a paradigm that we see a lot. So I don't consider myself to be competing with Amazon. You know I'd consider myself to be competing with their cloud services for voice not with Alexa.
Bradley Metrock: [00:15:25] Yeah that makes perfect sense. Very interesting.
Bradley Metrock: [00:15:30] So I guess my last question for you Rand is....I want to ask you just sort of in general, you've got a fascinating perspective as we've talked about, you've sort of been immersed in technology from a very young age. Paint me a picture of where you think we'll be with voice and AI in three or four years time and where you hope your company fits into that.
Rand Hindi: [00:16:03] It's quite interesting, right? There are multiple scenarios you can think of but I'm an optimist. And so the way I imagine technology is and voice in particular is as an interface that doesn't necessarily replace buttons and screens but makes it a lot easier and intuitive to use those interfaces that we already have. So I don't think that in four years screens will disappear or mobile phones will disappear or physical buttons to turn off your light will disappear. I think those will still exist but voice will also be there in all of those cases so that you wouldn't have to figure out or learn or make any kind of effort to use those different devices. So what excites me about voice is the ubiquity of the technology, the fact that you can apply it as much to your lights as you can as if you're in a meeting room, as you can to your TV or to your smartphone. And so for the company, quite frankly my objective is really simple. We want to be the only company people are going to whenever they want to put a voice assistant in device and doing this without internet connectivity and respect and privacy.
Bradley Metrock: [00:17:22] Cool. Well it sounds like you're well on your way. Everything about what you're doing is so impressive. You yourself are very impressive. Thank you very much for setting your time aside to chat with me today.
Rand Hindi: [00:17:33] Thanks for the interview - it was great.
Bradley Metrock: [00:17:35] For Artificial Intelligence, Episode Two...thank you for listening. And until next time.