Breaking Down A.I. | Interview with Rett Crocker | Part 1

DISCLAIMER: Below is an AI generated transcript. There could be a few typos but it should be at least 90% accurate. Watch video or listen to the podcast for the full experience!

Alexander Ferguson 0:01
Welcome to UpTech Report series on AI. I’m Alexander Ferguson. For our third series of deep dive interviews. I sat down with Rett Crocker, CEO and CTO of Udu in Raleigh, North Carolina. Rett has designed and developed over 100 games for mobile devices, personal computers and video game consoles. He’s also invented multiple programming languages, game engines and multi user content. And he’s created innovative software technologies in fields ranging from speech synthesis, to adver gaming to collaborative education. And the first part of a conversation, I asked read about the power of AI, the different facets of artificial intelligence, and how it’s evolved over time. I start by asking Rett one simple, but important question, what is AI to you?

Rett Crocker 0:50
A loaded term, that doesn’t mean very much. I think that most people that use the term artificial intelligence are using it. Not disingenuously, but certainly it’s marketing, more than anything else. To me. There are certain certain very specific technologies, some of the deep learning stuff that Google’s done. And, you know, some stuff being done here locally, that is really some fundamental in fundamentally interesting math and like good core computer science. But really, I mean, it’s interesting, our particular uses for it are, for example, are almost all just pretty much stats 101, it’s basic regression analysis with, yeah, we use some neural networks here and there. But you know, that’s not, you know, that’s not that hard. But a lot of people that are like, oh, yeah, AI, this AI that they really are just using it for marketing purposes. The real key, like if I really, truly had to define it, and want it to be not sort of making fun of that, is trying to use a number of different machine based techniques to try and predict outcomes. And that’s, if you can do that, well, it’s very valuable. But there’s a ton of problems with doing it. Well, for a while there, people were calling it predictive analytics and, and not using the AI term, which I kind of preferred. The predictive is also a bit of a loaded term. Nobody really uses it anymore, because a variety of reasons. But there’s no question though, that you can do some pretty interesting, pretty straightforward statistical modeling that gets you interesting results.

Alexander Ferguson 2:53
What are some interesting examples of AI that you’ve seen in business and everyday life?

Rett Crocker 2:59
The examples of that are really the predictive type functionality, where you’re saying, I don’t know how much John Smith is going to spend at my Veterinary Clinic next year. But I can build a model, a statistical model, that makes sense that, that tells me with some degree of confidence, that that’s how much they’re going to spend next year. But the, you know, real case is probably something that is that we’re using every day, Google Maps when you ask it for the location of the the local movie theater is basically doing sort of AI work, particularly since you didn’t write you didn’t type in the exact name of that particular movie theater. Google’s AI had to say, well, that thing that you wrote is probably, I’m predicting that it is this particular movie theater. And so I’m going to give you that address. Think back to the beginning days of email. We had to train our email programs to recognize spam. So we’d see oh, look, there’s another email that mentions Viagra, this is spam. Click that button. And you’re basically training a little AI, that’s in your email program, what you think spam is, and your version of spam and my version of spam might be different. They’re probably the same. They are now because Google solve that problem for all of us. And we all have the same spam filter. But really that little Bayesian algorithm there. That’s AI straight up because it is it’s just doing statistical analysis.

Alexander Ferguson 4:32
How does artificial intelligence power modern consumer technologies such as Siri Alexa or google assistant?

Rett Crocker 4:41
So first off, there’s sort of two categories of those right? So category number one is the elections of this world. Those are effectively the same or slightly advanced versions of the text adventure games from the 80s Right? And, you know, you basically say, you know, look, don’t look at door, you know, go east, pick up, you know, sword, whatever those are, that’s basically what Alexa does. It is a programming language to sky with a voice interface. And it is. So you have to speak the exact incantation in order to trigger Alexa and have it go do something. Siri and Google’s assistant, they call it nowadays, they’re different. And IBM Watson, the some of the stuff they’re doing there is different. They’re single sort of state versions of that, where they can answer a thing at a time. Occasionally, they can contain and hold some context and be able to answer more things like you can say, you know, I don’t know, what appointments do I have on Thursday? Isn’t there one that takes place in Chapel Hill, and it says, Oh, I wish the context was appointments. So I’m only going to look on appointments on Thursday. So I’m only going to look through appointments on Thursday, whereas many of them can say do versions of I mean, you can do this quite easily. Give me all the appointments on Thursday that are in Chapel Hill, done. That’s a multi multiple variable problem. But that’s all one problem. The real, the real trick challenge is multiple problems connected together, which is part of what we’re trying to build. And you can do that with you do quite easily. But you end up having to do it in this query language that’s effectively JavaScript. Or a very, very, very light version of JavaScript. But so that’s too technical.

Alexander Ferguson 6:52
How do you define the different facets of AI?

Rett Crocker 6:57
I think there’s probably too many to list because there’s a I mean, if we just look at the the like, language part of it, that’s probably, you know, 10 different sub disciplines right there. I mean, there’s speech synthesis, synthesis. I mean, nowadays, Google and Siri, they generate the voice that you hear is not some recording of a human. That and they do that through deep learning. There’s, under underneath it all is the deep learning stuff to begin with, you know, which there’s, that’s a broad category, too. There’s like a bunch of different types of that. There’s natural language generation, figuring out what to say, there’s, you know, the understanding piece of it. And the sort of context piece is completely not fully solved, I thought it would be, as I said earlier, but it’s pretty amazing to me, you know, how many different sub bullet disciplines are just for that piece? I was there was a company I was looking at earlier today. Can’t remember their name off the top of my head, but they had a chief language officer. They’re AI related, and they 100% have somebody whose sole dedicated purpose is language. And it’s not surprising.

Alexander Ferguson 8:17
What were your early impressions of artificial intelligence? How have these changed since then?

Rett Crocker 8:24
So game development inherently has sort of an AI problem if you’re doing single player games. So I built a number of AI’s over the years, with always with the same sort of principles that I used with have used with you do, which is, if you’re building an AI for a game, you actually don’t want to make it too smart. You. There’s a friend of mine, one of the original cofounders of Yudu, long, long ago, a guy named Richard Richard Harris, he’s in Scotland. He and I, and Frank Bozeman, who was the other sort of technical co founder, I guess, the three of us, had decided we wanted to do something, we started meeting in various cities around the world, and brainstorming ideas and stuff like that. And there’s a phrase he used to describe you to at one point that I think is just brilliant, and it sort of goes back to the way I thought of AI. And back when I did games, which is not artificial intelligence, designed stupidity. You basically allow in games what that means is that you’re you make it basic, insanely basic. It just knows how to I am enemy, I run towards bad guy or the player and I shoot, that’s all I do. And then what ends up happening is the human because we’re sort of narrative beasts. We apply our own narrative to it and we imbue them with thinking, even though they don’t actually do anything that intelligent, they’re just running at you and shooting but because of the fact that they’ve got a few, like things where they like a little bit of randomness, it makes us think, Oh, they’re ducking and dodging and, you know, all that. There are definitely approaches that people are using now that were basically discarded many, many years ago. I mean, neural networks to a certain degree, are an example of that. Because, you know, we’re sort of in the right place, right time, because we have processing power to bear. We have a lot of processing power anyway, we have a massive amount of memory structures, we have cloud based computing. There’s all these technologies that allow us to do the things we do. I used to say, when I was first trying to introduce you to people and you know, with all the connected API’s and everything, when somebody like a VC in particular would say, oh, yeah, well, why now? Why this? You know, what, you know, all that. The answer is a because you couldn’t do it 10 years ago, because cloud based computing didn’t exist API’s to a large degree didn’t exist, at least in the form they were then. And, and then, you know, the the bigger thing is like some of the core techniques that we used to build YouTube didn’t exist.

Alexander Ferguson 11:27
That includes the audio version of this episode. To see the original and more visit our UpTech Report, YouTube channel. If you know a tech company, we should interview, you can nominate them at UpTech report.com. Or if you just prefer to listen, make sure you’re subscribed to this series on Apple podcasts, Spotify or your favorite podcasting app.