Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Leveraging AI for Player Engagement with Industry Veteran Stephen Peacock
69 Plays8 months ago
In this insightful episode of the Player Engage podcast, host Greg delves into the fascinating world of gaming AI with industry veteran Stephen Peacock. With over two decades of experience, Stephen has been at the forefront of technological advancements in gaming, from his early days at AudioBrain to his transformative roles at Amazon Web Services and now as the head of games AI at Keyword Studios. The conversation covers a range of topics, including the rapid emergence of AI tools, the impact of generative AI on game development, and the ethical considerations surrounding AI in gaming.
Key takeaways from the episode:
- Stephen's unique approach to testing AI tools using his own headshots.
- The evolution of AI in gaming and its application in areas like LiveOps and community health.
- The future of AI in gaming and how it's lowering the barrier to entry for game development.
To get a deeper understanding of:
- How Stephen's personal experiments with AI tools provide insights into their capabilities.
- The role of AI in enhancing player experiences and maintaining healthy gaming communities.
- Stephen's perspective on the ethical sourcing of AI and its parallels with other industries.
For a full exploration of these topics and to gain valuable knowledge from a leading expert in the field, tune into the Player Engage podcast episode featuring Stephen Peacock. Discover the intricacies of AI in gaming and how it's shaping the future of player engagement. Listen now to learn more.
Recommended
Transcript
Introduction to Player Engage Podcast
00:00:00
Speaker
Welcome to the Player Engage podcast, where we dive into the biggest challenges, technologies, trends, and best practices for creating unforgettable player experiences. Player Engage is brought to you as a collaboration between Keyword Studios and HelpShift. Here is your host, Greg Posner.
00:00:16
Speaker
Hey, everybody. Welcome to the Player Engage Podcast. Greg here.
Featuring Steven Peacock - Tech Leadership in Gaming
00:00:20
Speaker
Today, we're thrilled to feature Steven Peacock, a tech leader who's significantly impacted the gaming industry. With a career spanning over two decades, starting as the first field engineer for AudioBrain, then following that up with transformative roles at Amazon Web Services,
00:00:35
Speaker
and now the head of games AI at Keyword Studios. He continues to shape the future of game development. So join us as we dive into a conversation full of insight from a visionary who's at the cutting edge of gaming technology. Get ready for an inspiring session. Thank you for joining us today, Stephen. Do you have anything you'd like to say? No, but I should carry you with me on the road to hide me. That was a pretty good intro. Thank you. I'll try and live up to it, at least for the rest of my career in games.
00:01:03
Speaker
Beautiful. Let's start off kind
The Rise of AI Tools in Gaming
00:01:05
Speaker
of just what's new in the news, right? It's February 1st today and AI tools are emerging at a crazy rate. It seems like about a year and a half ago, we were introduced to open AI for more of a public side where consumers can start using it. But today, I follow you on LinkedIn and you're trying all these different headshot types of AI tools and other types of AI tools. So, you know, February 1st, 2024, what are some of the AI tools that you've played around with recently? And do any of them leave a mark on yours? It's just kind of fun to play with.
00:01:33
Speaker
Yeah, I do like playing with them. And I don't use my headshot out of vanity. It's the only head I have multiple photographs of. And the more you photographs you have, the better the training. So yeah, I've been playing with them. I tried one the other day. His name's escaping me. I'll try and look it up.
00:01:57
Speaker
while we're chatting here, but it's got quite a nice system for both creating you as an actor and then also a flow to put you into some small roles. It did actually create a video of me as the winner of the Isle of Man TT.
00:02:15
Speaker
saying something absurd. It wasn't bad. My family wasn't impressed. They told me to keep my day job. But it did at least look better than the last ones I put up, which I put up a set of headshots and everyone said, hmm, could be related. I can see the resemblance. So it's fun to see where these tools have got to, you know, on the tech side as well.
00:02:38
Speaker
when people are first starting to play with these, I like to encourage them to do things like ask it about yourself. Right. It's key to it's important. This is why the headshots and doing doing yourself or like being so self-centered is kind of good because most people are an expert on themselves. So if you get a photograph of yourself, you look at it and go, yeah, that's not quite right. Or if you ask it for your resume, I mean, according to last time I tried chat GPT,
00:03:07
Speaker
I was way more impressive than even you made it sound. ChatGPT actually credited me with co-founding Activision, and I don't know if I'm- Congratulations. Yeah, I mean, it was a huge part of my career that I apparently do not remember. But it's always fun to play with these different tools and see where they're getting to. I'm making light of some of their capabilities, but these are amazing pieces of technology.
00:03:33
Speaker
of just how much improvement and how much advancement has been made in just the last couple of years. I've been doing this for five years, and in the last two or three years, it just seems to have been like a Cambrian explosion is a term I've heard. It's like, oh, wow, all these different things coming out suddenly.
00:03:53
Speaker
So it's been very exciting. I love your two comments because I think it's really important for people to understand. Like first of all, you're testing with your own headshot. You're probably going to be your harshest critic, right? So if you want to judge how well a tool works, test it on yourself because if you don't think it's going to look good, I mean, you're good and you're going to be your harshest critic. And then the second part of that was if you check yourself, like throw your name in the chat GPT and see what it has to say about yourself.
00:04:17
Speaker
you'll understand what type of BS might come out of that tool. So next time you use it, you might be a little more wary on that. So I think those are really two great takeaways that in light, it's a funny thing to say, but it's a really great way to test out these tools and see what their capabilities are.
Integrating AI into Game Production
00:04:32
Speaker
Let's take a step backwards about your role, because I think it's a cool role, it's a cool title, and I think I want to understand and our audience will want to understand what it means. So what does it mean to be the head of games AI for a gaming studio?
00:04:44
Speaker
Well, I read sort of an article about the new
00:04:51
Speaker
hot title being the head of AI. I've been doing that for a couple of years. And the question was like, what do they do? So it's very interesting. And I think it's the best job in the world. So I've been doing it for a couple of years. I was doing it for a few years at Amazon as part of the AWS Game Tech for a few years before joining Keywords. And so it very much depends on where you are, right? At Amazon, the role was about identifying
00:05:17
Speaker
where this very exciting technology could fit into game production. Keywords, actually one of the key reasons I wanted to join Keywords,
00:05:28
Speaker
I wanted to get closer to actual production, right? I wanted to be working with artists, programmers, you know, all the creatives that are using this, all of the people at the back end who are working on live ops or trust and safety or community health. I wanted, you know, I'd been advising where these tools could fit into their pipeline.
00:05:55
Speaker
But as a technology provider at Amazon or anywhere else, you're kind of at an arm's length. You're a technology builder. You're not the user. And so for me, the opportunity to join Keywords was quite irresistible because Keywords is both, right? Keywords is a technology company.
00:06:15
Speaker
But it's also got all these amazing creative people and people who work in game production to build games and run games. And so the exciting thing about my new role, because I've only been here a couple of months,
00:06:28
Speaker
is how much I'm learning about AI as Thorber hits the road. There's the theory of, oh, this is going to be great. It's going to identify all the toxicity in the conversation. Or this is great. It's going to do translation, do it real quick, or create art, or whatever it is that the technology people believe their product is going to do. And then there's the reality of how does it actually do in real
00:06:53
Speaker
real-life AAAA production. How ready for battle is this technology? Is the quality there? Is the reliability there? Is the controllability of the tool there? Those are very interesting conversations. As head of gaming AI at Keywords,
00:07:12
Speaker
big part of my job is the same as it's been since my first days at Alien Brain, really. It's looking at a new technology, a new capability, trying to fit it into production, where it's going to be beneficial. Don't try and over-hype it. Make sure it's set expectations right as to where it can and can't fit. And I've always been a big believer that there's a space
00:07:36
Speaker
for a technology. It doesn't have to solve every problem to be useful. It just has to solve enough problems to be useful. And so often, if you're on the tool creator side, your marketing department and your sales team come in, it can sometimes run away from them. And it sounds like it's a tool that solves all the problems.
00:07:59
Speaker
Yeah, you're saying don't overhype it. It's crazy. That's what all is about right now. But it's fun. I mean, I understand it being a fun role, but I also could see it being a mind going crazy role because I mean, at keywords on the create side of the company, right, there's 23 to 25 studios of creatives creating games. And I have to imagine
00:08:24
Speaker
With 25 studios, number of individuals at each, you're probably looking at way too many AI tools. I'm sure everyone wants to look at something or test something or play around with something. So like, do you have a quick litmus test to say, hey, is this a valid type of technology or not? Or are you just kind of going with open eyes and arms and test it out?
00:08:45
Speaker
There's a couple of ways to look at these technologies when you're considering them. The first thing is, is there an actual use case? Because there are a lot of technologies, and this is just the nature of technology, especially with the way the software industry has moved, it's become very much
00:09:05
Speaker
research projects are getting launched as product. Research papers become startups incredibly quickly. And then there's this phase where they go looking for a market fit. They go looking for a reason to exist. Software used to be harder to build, harder to ship, harder to get awareness. A decade ago or two decades ago, it was much harder. So you didn't just
00:09:30
Speaker
throw stuff over the fence and see if anyone was interested. It was too much of an investment and you had to figure out how to get them the CD and pay attention. It was just much harder to reach people. Nowadays, we do have this environment where research papers become startups and they're looking for market fit. You hear them talking about that all the time. The first thing for me is
00:09:54
Speaker
is where is the actual need on the floor, right? There are some tools that break the paradigm so much that when you look at it, you think, well, maybe this could completely change how we approach this problem. And so you have to step back and look at the bigger challenge that the group is trying to solve. But often where these tools are really
Evaluating AI Technologies for Gaming
00:10:17
Speaker
valuable is the being able to pick up on work that people really don't want to be doing.
00:10:24
Speaker
looking for the grind. I know some people like to play grindy games, but a lot of people, when you ask them for the grind in their work, they'll tell you about some tasks. And I found that very often those are the kind of tasks that computers are actually pretty good at. When you ask someone what they do and don't like about their job, they rarely say, oh, I wish I didn't have to do all this really creative stuff. I wish I could balance more spreadsheets.
00:10:53
Speaker
things like this. So it's often a good place to look for those use cases. So when I'm looking at technologies, it's that, but also where are they in their evolution as a technology because it's a difference between if you're looking to adopt technology as a very small indie and you're looking to be very experimental in the kind of game you're building and you're looking to take risks and experiment, you might be looking at technologies that
00:11:21
Speaker
really don't fit into AAA production for a whole host of reasons, right? It's now, it's a major IP with, you know, an install, you know, a massive established fan base who know what they expect from the next iteration of that IP. And they know what gameplay they expect. And it's a public company. So they have to be more careful than a small indie. And so it's really about
00:11:48
Speaker
Have you got your T's and C's set up? Have you got your business model? Are you built for scale? Is there reliability? Is there enough control? Because when you start to get to large scale production, where a lot of these startups struggle is that
00:12:05
Speaker
They work in a particular case, so long as you squint and look at it just from this direction, it produces really great results. And that's not just an AI challenge. I mean, software has done that for ages. But in large scale production with hundreds of people on the team, these tools have to be very scalable, very reliable. And that's one thing I look at when we're looking at new technologies.
00:12:31
Speaker
So there's two things you said that really stood out to me, and I think it's great points to be made. I think the first one is how quickly companies launch these days, right? Companies are launching with a product they think is cool looking for a problem in the marketplace. And when you're looking for a problem and you're going to create a problem, your product's probably not going to catch on because there's no real problem to solve. The other one I like that you pointed out is
00:12:54
Speaker
people work their day jobs and they want to get rid of their grind, the grindy work, right? And being from help shift, right? We sell bots and we don't want to replace everything with bots. We want the grindy work to be replaced with bots, resetting users' passwords, doing all this stuff. So I think it makes a lot of sense when you look at what is that low-hanging fruit? We don't want to replace everything. We don't want to replace most humans, right? We want to get rid of the grindy work so you can focus on things that are more important and bigger picture. But this makes a bigger question, right?
00:13:24
Speaker
I am fortunate enough to be hosting this podcast, but I am not nearly as smart as I look, right? One of my favorite days in gaming was 9999. That's because that was the day that the Dreamcast launched. And it was tagline, it's thinking. And now I'm talking about help shift and the concept of AI. And I know we don't do gen AI. So there's different types of AI. And people say AI now, and I think they automatically think, oh,
00:13:50
Speaker
chat GPT, right? And I don't think that's the case, right? So maybe you can give us a high level understanding of what I would say the different types of level of AI are, right? There's deep learning, there's machine learning, there's now generative AI. But for people who aren't as smart as you, what's kind of the difference between these? So the way
00:14:10
Speaker
The industry is, when I first got into this four or five years ago, people were talking about ML and deep learning all the time, right? And AI was not really mentioned. And especially in games, AI very much referred to pathfinding and rule-based systems that made Pac-Man, et cetera, feel intelligent. And that's really the definition of AI. Those were correctly labeled as AI.
00:14:39
Speaker
they ask, AI is any system that appears to be doing things that requires intelligence, right? So right from the earliest days of video games, we've been programming state machines and pathfinding systems that can make the NPC appear to be intelligent, right? I mean, maybe not in the earliest days, but the ghosts in Pac-Man reacted to what you were doing, right? Later on, the NPCs
00:15:05
Speaker
the guards patrolling, whatever you were trying to stealth infiltrate, they turned if there was a noise, right? And they would walk towards you and kind of investigate where you were hiding. Those were mostly state machine-driven systems, and they are, by definition, AI. They are artificially intelligent. Then machine learning is where a system is being trained not by a
00:15:32
Speaker
a detailed set of rules and states, but by data. It's being shown a lot of data, and it's finding those patterns and learning that behavior. Machine learning was the first time that term was used. It was actually in a paper about a checkers game, and the creator wanted to thought of machine learning as the ability to create a game that could learn to play better than him, right? So, exceed its creators'
00:15:59
Speaker
understanding of how to play checkers. I'm not sure if it ever did, but it got really close to his skill, beat him occasionally. But that was the first use of the term machine learning. So that's not really explicitly programming behavior into a system, but rather showing it a lot of data and letting it figure it out.
00:16:18
Speaker
Deep learning is a subsystem of that, which uses neural nets, which are essentially algorithms that were inspired by the way our brains work. And these can get really, really big. And in recent years, they've become foundation models. There's been this convergence of having enough data with all of the information on the internet, having enough compute, especially with all of the
00:16:44
Speaker
the work that NVIDIA has been doing with GPUs and having enough money. You know, the money has been pretty cheap for the last decade and all those things sort of converged. I like to think sparked by some work done by ex-gamers building AlphaGo and beating the World Go champion, which really woke the world up to what these deep learning models could do.
00:17:07
Speaker
That sort of spurned a huge influx of interest in AI again. That was five, six years ago. I forget the exact time. But what happened recently was open AI, not to put all the blame on those, there were others as well. They sort of came along and they started labeling the next generation of deep learning.
00:17:32
Speaker
these foundation models and the next step in the evolution of these models as AI. It's a little bit like Tesla inventing electric vehicles and
00:17:48
Speaker
and deciding to call them all transport or vehicles. It's technically correct, but imagine if Tesla had started this trend where we all called our cars vehicles. Yes, a Tesla is a vehicle, but it's a very small subset of the vehicle landscape. It's a car, more specifically, it's an electric car,
00:18:10
Speaker
more specifically, it's an electric car with a certain set of self-driving capabilities. And calling
Understanding Generative AI
00:18:17
Speaker
everything AI is kind of like
00:18:20
Speaker
The new trend is to call every car a vehicle or all electric cars vehicle. So it's a little bit confusing and it can get contentious. Should we be calling it AI? But I think that boat has sailed, right? It's sprung into the public imagination. People have suddenly captured the term AI. And what they're almost always talking about is quite advanced deep learning, but only us nerds need to fret about that.
00:18:47
Speaker
So I'm thinking as you're talking through tests, it's like the iPad or iPod, right? The original iPod. Everyone thought that was the first MP3 player that ever came in. Everyone called every MP3 player an iPod. But so you're safe to assume as an end user myself, right? Like, is Gen AI just a conversational front end for me to look into something that's using deep learning?
00:19:09
Speaker
Generative AI is a use case for deep learning and these foundation models. They've been used. You hear a lot of people talking about how
00:19:21
Speaker
deep learning in these foundation models are helping with protein folding and they're helping with gene analysis and lots of interesting things in medical space and in the financial space and they're helping to keep us safe and secure when we're online doing banking or playing a game. They're
00:19:44
Speaker
They're helping understand what's being said and identifying whether it's toxic or not. And those are all uses of these foundation models that don't actually generate new content, but they're using these models in a way that is really beneficial and is integrated into our entire lives, from the personal assistant on our desks or in our kitchens. Those are all using deep learning.
00:20:12
Speaker
A generative AI is perhaps a subset of deep learning or a use case of deep learning, right, where those models are being used to create something net new. And that's relatively new. And so that's where chat GPT can write a poem for you, or mid-journey can create a piece of art for you, or, you know, there was computer vision for years and decades, right?
00:20:35
Speaker
You know, still does and is like critical to your Tesla again, right? It's as it's driving down the road. It's got those cameras there. You've probably seen the videos of all the squares around the people in the cars and all the objects in the scene. That's computer vision, right? That's deep learning models at work.
00:20:53
Speaker
helping your car not crash or helping detect cars in your blind spot or whatever technology you have in your car or helping your smartphone focus on the eyes of the person you're taking a photograph of. That's all deep learning, but it's not generative. Generative is a new wave, a new use case and a very, very interesting one. It's not one that
00:21:17
Speaker
AI has been capable of until quite recently. GPT has been out for four years, I can't remember exactly, four or five years since it first came out. It caught fire when they put the chat interface on it 18 months ago, but it's been around a while. It's very new space, but it's very exciting. It comes with its own challenges.
00:21:40
Speaker
Yeah, I realized this after saying it, and I sound silly, right? It's generative AI. It's generating new content based on historical stuff and things it knows in its database. So that makes a lot of sense. So this next question, just so everyone's clear, isn't necessarily about Gen AI. Gen AI, it's all about all AI. But let's talk about AI and the evolution of it in gaming. There's a lot of cool things happening in gaming right now that's using AI, and not just Gen AI. Just you talk about things like multiplier. You talk about things like just
00:22:09
Speaker
all this new content that's coming up. What do you see happening in the next decade? What excites you about the potential that AI has? What worries you, if anything?
00:22:17
Speaker
I tend to be more of an optimist on this kind of thing because I've seen a lot of waving around of worst case scenarios. What if this technology does this X thing and often there isn't really a great business case for that or society just reacts. Society always surprises, I think, in the long arc of time, right? Society tends to react in a way and fix these problems. But I think the future of
00:22:46
Speaker
AI games have been a very critical part of the evolution of AI, and AI has definitely helped with games, with the evolution of games. There have been games
Enhancing Game Experiences with AI
00:22:58
Speaker
all the way along that have used AI, because even from Pong, the paddle on the other side of that game appeared to be intelligent. So it's been there from the beginning.
00:23:07
Speaker
But I think it's really exciting where AI can be applied, especially in areas like LiveOps, in community health, in community management, keeping us safe as we play. There's kind of the before gen AI and after gen AI, but all this technology that was before gen AI is stronger now than it ever was, because it's also benefiting from all this technology. So there's a lot of use cases, even if we put aside
00:23:37
Speaker
AI creating new content. There's so many interesting use cases in games for these foundation models and matchmaking, pairing us with people so we have more meaningful matches, managing which server we're connected to. Almost every part of our game playing experience can be enhanced with
00:24:01
Speaker
AI with these foundation models and with these new capabilities that are coming from deep learning. So I'm very excited by what it's going to do. But it's going to lower the barrier for entry for game development, again, as game engines did before them and as
00:24:20
Speaker
you know, more accessible programming languages did before that. Every time there's a new technology, more people get to be creators. If you think about the user-generated content boom that we've been seeing, you know, for like a decade now, you and I were talking previously about, you know, the modding community, right? You used to need to be a pretty serious programmer to mod a game, right? And now, you know,
00:24:49
Speaker
Kids are doing it in Roblox, right? It's much more accessible, which allows people to be very creative, to create levels for their friends, to then have recommendations and find all the content because the lower the barrier to entry, the more people are creating, the more content there is.
00:25:09
Speaker
which means to find the content that's gonna really matter to you and resonate with you becomes a harder and harder job. Again, another area where AI can come in and help you and help guide you like Netflix does to the content that's most likely to appeal to you. So really at every point, I very much think of it like digital photography. I was in photography before it went digital and for a decade we called it digital photography. I don't think people call it digital photography anymore because all photography is digital.
00:25:38
Speaker
I think AI is going to do the same thing to game production. At some point, we're just going to stop talking about it because it'd be like talking about where can we use electricity in game production. You mentioned an interesting point about modding, right? We talked about that and I love the idea of like
00:25:54
Speaker
you know, playing Skyrim and typing in space theme Skyrim, which star field, right? But maybe people want to build it in Skyrim, right? And seeing what's created. But it brings up that question of ethics and what's responsible when it comes to using these tools and AI. And Steven also let me know that his daughter gives him a hard time because he's not 100% familiar with everything Taylor Swift does. But Taylor Swift kicked off this whole thing. What was it last week or late last week, right after the Chiefs win on
00:26:23
Speaker
digital photography, right? And being subject of AI creation and people are in outrage at it took Taylor Swift to get to Congress, but at least it's there now, right? Who cares how it got there. But like,
00:26:37
Speaker
There's going to be this fine line at some point on what can we use? What can't we use? New York Times is suing OpenAI for data that they're using. And you may not be able to answer this, but how do you moderate that? How do you take those precautions? How do you do that? I guess it's different from an indie company to a AAA company to a modder, right? But when do you have to start thinking about ethics and responsibility? How do you manage that, especially with these companies that you're working with, because you don't know
00:27:07
Speaker
where their sets are coming from? It's a challenge, for sure. When we're talking to companies, we're trying about a new technology. And this is not new to Gen AI, right? Some of the challenges are new. Some of the content creation challenges are maybe slightly novel or new in Gen AI. But the problem of bias in data sets of
00:27:35
Speaker
a decade ago, banks using AI to decide who gets a mortgage and who doesn't, and then figuring out that their data set is biased. It's favoring one demographic over another or one neighborhood over another. These things have been happening. It's one of the challenges of machine
00:27:57
Speaker
of all machine learning, including deep learning, Gen AI is biased.
Ethical Challenges in AI
00:28:02
Speaker
And we've seen that in our testing, right? And if anyone's played with these tools, you can see them, if you played with them a year ago, and you typed in, give me a picture of a doctor, a year ago, you would have got six white dudes with stethoscopes around their necks in white jackets, right? Probably all of them over 50.
00:28:23
Speaker
Because when you look at the totality of what's on the internet and you say, like, label all the ones that are doctors, like, all the doctors look like that on the internet, right? But the internet is not representative of the whole of the world. It's a very...
00:28:38
Speaker
Not not every doctor in the world has a profile photograph on LinkedIn, right? So the data set is biased naturally. And so these problems of bias and data transparency and aligning models with human values predate Gen AI. But Gen AI has thrust the whole thing into the public view and has also made those tools more accessible. And
00:29:01
Speaker
But I don't think, actually until we talked, the most recent Taylor Swift issue had actually passed me by. I missed it. There's so much news all the time. But as you say, this was in front of Congress. Sam Altman was in front of Congress six months ago talking about this and have been
00:29:24
Speaker
the worries about what it's going to do for not only our elections, but other people's elections, the ability to do deep fakes, you know.
00:29:32
Speaker
in a puffy jacket, right? You know, those things have been around. Now it's moved past images into text, into voice, you know, cloning and video. So as these technologies exist, it's the same old trolls coming back to us again, right? First of all, they were putting the Pope in a puffer jacket and there was a lot of use of these technologies. Now it's cloning people's voices.
00:29:59
Speaker
and or putting them in videos but it's the same challenges it's how do you get your data set to be clean how does the company providing this technology help the user police it and use it responsibly and I think that's
00:30:15
Speaker
That's the bit where both societal pressure and business fit is going to drive these tools. We're seeing that already. OpenAI, if you use Microsoft's tools to create an image, they'll actually watermark it. Not a visible watermark, we can see, but they're watermarking it so that
00:30:38
Speaker
they can tell that it's being created by a generative tool. And you're seeing more of that. You're seeing more transparency about where the data set came from. And then if you're talking to these companies from an enterprise position, what I find very heartening is that more and more they're talking about the work they're doing to retroactively clean up their data set or build new models on completely
00:31:06
Speaker
or on datasets that are trying to be cleaner, whether they've tried to license the entire dataset or they've tried to only source from the Creative Commons. I think Adobe led the way they're trying to build their Firefly engine with just images that they'd either licensed or were in the public space from an IP perspective. Others have followed suit.
00:31:34
Speaker
So there's a lot of work there, and that's because their users, both enterprise users and everyday users, are pressuring them to be better at this. So as I said, these are startup projects, startup, this is research that suddenly becomes startups. I mean, even the granddaddy in the industry, you know, OpenAI is like, is only being in the
00:31:59
Speaker
had a product for 18 months. So these are all startups. They're still figuring it out. But they're working hard to clean up their data sets and be more transparent about how these tools are getting built.
00:32:11
Speaker
Yeah, it seems to be this large gray area that most companies are playing in right now, because again, there's no rules. No, no, well, there's rules and laws, but perhaps how do you determine who's following the rules and laws? But something you said I found interesting was kind of the mortgage companies kind of using data and may have false positives, the same with that. But what other kind of cross industry or verticals are utilizing these types of tools today that people may not be aware of? AI is is especially
00:32:40
Speaker
especially if you're living in a digital space, if you're working with digital tools, if your day-to-day is in front of a computer or working with computers, you're interacting with AI all the time without probably even realizing it. I mean, one of the things I like to use as an example is photography used to be hard, right? Go back and look in your old family photo album
00:33:07
Speaker
would guess that unless you had an avid photographer in your family, my photo albums, family photo albums are full of out of focus, badly colored, you know, pictures of people with their eyes shut, right? And, you know, photography used to be hard. Now I can pick this up and I can
00:33:29
Speaker
you know, snap a picture of someone at twilight standing in front of a bright light and it will correctly focus on their eyes and it will set the right, you know, color balance and it will try and sort out the backlighting and I'll end up with a, you know, pretty damn good photo. And that's there's a lot of AI goes into that, right? It's not just clever lenses. There's a lot of AI figuring out where to focus and everything. And we're not aware of it. And there's so many touch points
00:33:57
Speaker
the things we think of as AI, like our, like our home assistants, things we maybe don't think of a lot as AI, you know, how the power grid is being optimized to power our homes or, or, you know, how traffic is being controlled in our town. You know, all around us, AI is, is positively impacting us. So we're about, we're a little more than halfway through this and I want to do a little fireball around and throw some curve balls at you. So you good to go with that? Yeah, sure.
00:34:27
Speaker
What'd you have for breakfast today? Uh, butter coffee. What? Actually butter tea. I stopped drinking. I switched, but people noticed. Done with caffeine, done with coffee. Uh, what, uh, what show are you binge watching? Oh, I've just watched a very funny show, uh, called, uh, deadlock.
00:34:49
Speaker
which is out of Tasmania and Netflix seems to be pointing me to a lot of Australian and Tasmanian programs right now. I found it to be a very amusing story about a pair of
00:35:09
Speaker
mismatched buddy cops trying to solve a rather prolific serial killer in their little town and in the back end of nowhere in Tasmania. And it had a lot of good Aussie humor in or Tassie humor. And it was it was it was great. I loved it. I've just finished that. Great. What is your what was the last book you read?
00:35:31
Speaker
Now this one I'm gonna have to look it up because I love listening. I'm listening to books all the time But I am absolutely dreadful at remembering the title So it was competing in the age of AI and I should call out before then the book I read before that I particularly enjoyed was reality is broken which is about the video game industry and it's a great read for anyone in the industry I'd recommend that one in case you don't get enough AI in your day job you read about it after work as well. So that's
00:36:00
Speaker
Yeah, Reality Bro is broken is about games and the use of games for the greater good of society. So that's less about AI. Speaking of games, and the last question is, what was the last game you played?
00:36:18
Speaker
The last game I played is almost always Doom. I love Doom. I play it a lot. The last game I actually tried, just to remind myself again, was Osmos Plus, which is a game about, it's kind of a
00:36:39
Speaker
It's a great little mobile game. I don't think it's that new, but you're kind of this cell and you have to eat the cells around you that, you know, you can't eat a cell that's bigger than you, but every time you eat a cell, you get bigger. Wasn't that a game from EA years ago? Wasn't there a game? It reminded me a lot of Asteroids, actually, because you're kind of firing yourself in a direction and then you're trying to
00:37:02
Speaker
I don't know, it had asteroid vibes to me. And I wasn't playing it for long. I think I played it for two or three days before I'd had enough of it. But that might technically be the last game I played. You're a Doom fan. I can see it because for those who are listening, but behind Steven is a Doom poster and behind me is also a Doom poster. So we both
00:37:22
Speaker
I had a Doom Slayer shotgun. Don't forget the Doom Slayer shotgun. It's a 3D printed shotgun that I finished painting up this holiday. I'm very, very proud of my Doom Slayer shotgun.
00:37:34
Speaker
Well, if anyone knows John Romero and he wants to be on the podcast, he could come in and join Steven and I and we could just nerd out about Doom and what Doom has done since then. So that's the end of our Fireball questions here. I want to go back to the gaming side of things because we like to talk about gaming because we're a gaming podcast. And you mentioned things like live ops being exciting to you in the world of AI as well as kind of matchmaking. Can you go into a little more detail on what makes that exciting?
00:38:02
Speaker
Yeah, so when I first started looking at applications of AI from when I was at AWS, one of the areas that stood out to me, partly because we had some technology in the space, but also just I felt it was an area, it just resonated with me as a father and a game player myself, was the idea of using AI to make healthier communities. Game developers,
Improving Community Health with AI
00:38:32
Speaker
build games to create it. They build experiences. They have this experience in their head, whether it's a single player experience or a multiplayer experience that they want to create for people. And that experience can be very negatively impacted or positively impacted by the community that is attracted and how you reward that community or police that community. And I felt this was an area where there was
00:38:58
Speaker
you know, great struggle, great need. You know, there just weren't enough community managers in the world to stay on top of it as games got bigger and bigger and they became more of like a social platform. And, you know, people used to have done for a while. They play games almost more to meet up with friends than to actually play the game. Right. And so especially during Covid, when so many people were online playing and so many people were
00:39:26
Speaker
Under stress, psychologically, I think it was a moment in time when there was a lot of people online and some of them weren't that happy for very understandable reasons. And it really felt like toxicity and community health was a big thing that AI could tackle. So that's where I focused a lot of my attention when I first got into AI.
00:39:48
Speaker
worked with partners like Spectrum, Modulate, but also with AI tools from within AWS. Comprehend recognition, these type of tools that can identify undesirable objects and images that have been created in user-generated content. Undesirable behaviors can be identified with anomaly detection, whether it's cheating or just unusual bad behavior.
00:40:14
Speaker
And so it was an area where I felt that the AI was already applicable, could already be a big help, and was very much needed. So that was really why I focused first. So community health and how AI can help there has always been important to me.
00:40:31
Speaker
Yeah, I love that. We spoke with Spectrum Labs on the episode of the podcast as well as Sharon Fisher from Keywords and Moduli as well on the podcast to learn about this. It's crazy. I mean, you have to be
00:40:47
Speaker
completely in the dark if you don't know about the stuff that goes on online, right? But you have to also understand there's people that are moderating that and seeing this content is not easy on either side, right? We want to protect people from it, but we also have to protect our moderators. And it's really cool that what we've learned from Sharon and Wells, what you're saying is like the tools that have come out there to really help protect both sides. We want to make sure it's
00:41:08
Speaker
a good community, a safe community for our kids and even adults and anyone of any gender to go on and enjoy it. And I mean, the more people that play games, the better and the more we can protect these people from trolls, right? Trolls end up everywhere. I guess that's the story of life is you can't stop trolls from coming everywhere. But the more you can do to protect them, I think it's fantastic. And I agree with you. I love that side of AI that I don't think people always really think about that. I think most people when they're thinking about AI these days is really just
00:41:34
Speaker
kind of the chat GPT and mid journey type of stuff and not really thinking about what tools might be existing in the back end. And I think understanding the difference between those and the importance of it in gaming is extremely important for people. Yeah. And I'd also say that, you know, people also, when they do think about those tools, they think about them purely as, as a punitive police, right? Identifying bad behaviors, you know, coming down on it, either reeducating people to the rules of the community or chucking them out. But there's,
00:42:04
Speaker
You know, being proven, there's a lot of progress can be made with identifying the heroes in your community, identifying whether you call them champions, heroes, mentors.
00:42:16
Speaker
You know, many communities are full of them. People who enjoy helping noobs as they enter the game, right? Enjoy helping, you know, those people getting their start in the game. Come over here. We've got a great group. Meet these people. You should go over there, right? And these technologies can also help identify those people, right? And you can put creative programs in place to reward them for that behavior.
00:42:43
Speaker
And that can be more beneficial in some ways than just going to play that eternal whack-a-mole game, encourage good behavior in the game. So that was one of the reasons why when I did the initiative, it was important to me that it was called community health and not content moderation because content moderation felt like it was part of the solution.
00:43:10
Speaker
And something I don't think we see enough is rewarding the positive in gameplay, right? What you're talking about is, right? We spend so much, I remember GTA 3 coming out, right? And everyone talking about violence in games and blah, blah, blah, blah, blah. And I feel like we always hear about the negative side of what's happening in games, but no one really touts the positivity in games and what it can make a difference. And I think if we put more focus on positivity, I think it spreads. I think it's a...
00:43:35
Speaker
contagious thing. And the more people you see getting rewarded for doing good, the stronger these communities will build up. I think it's all good.
00:43:43
Speaker
All good stuff. So in the world of gaming, we got a lot of cool buzzwords, right? We have, I mean, looking back even over time, some of them aren't buzzes. They actually become things like 3D graphics were a thing. Online multiplayer for kids didn't know that wasn't always a thing. VR is catching on right now. We just had Apple come out with their glass a couple of days ago. Now we have Gen AI and people are talking about that.
00:44:07
Speaker
And people are comparing it to NFTs, which NFTs might become something, right? We're not gonna say RR, but how do you look at generative AI when it comes to gaming? And does it have legs?
00:44:22
Speaker
I think it definitely does. Yes, I'm an AI person, so I'm potentially biased, but I've been an AI person long before Gen AI, so I believed in AI before it could generate an image, and it's impacting games. I think even the Gen AI use case is very powerful. When you look at
00:44:47
Speaker
When you look at what people, it's allowing people to do. And we talked previously about the modding community, about enabling, you know, creators of all skill levels to be able to create. I think these tools, whether it's helping you to write a script, whether it's helping you to create images, you know, right now,
00:45:10
Speaker
I don't think it's really high-quality content that's being output by these tools.
Generative AI for Creative Support
00:45:15
Speaker
We set such early days for these tools, but they do help in the creative process. If you've got talent in that area, they help more than if you don't. We're doing experiments across keywords, as you say.
00:45:33
Speaker
when I see what artists are able to get out of, say, mid-journey, right, compared to what I can get out of mid-journey, it's night and day, right? There's no competition. They get amazing results out of mid-journey. Writers get amazing results out of GPT that I couldn't do because my lack of artistic
00:46:04
Speaker
creativity.
00:46:07
Speaker
stops me from writing the describe. You have to first have the vision in your head to write the prompt to get the result. And an artist has that skill, and they know what they want. And then when the result comes out, they can see what's wrong with it. And I don't have either of those skills. I got some composition skills for my photography days. But I always shot in black and white because I have really crap feeling for colors, coal bind or anything.
00:46:36
Speaker
I'm not good at that, picking a good color palette for a picture or design or anything like that. So that's why all my photography was always black and white. It was just easy for me. But in the tools of truly talented people, these tools can do amazing things. And so I think there's definitely a use case. And also, even if all it was ever used for was user-generated content,
00:47:01
Speaker
That would be a use case in itself that would be incredibly beneficial to the games industry.
00:47:11
Speaker
these tools are getting better at figuring out the responsible AI side of it. I think of it almost like ethically sourced. We're very aware of where our coffee beans come from these days or how the minerals were sourced for our iPhones. Apple always makes a big deal of its production
00:47:37
Speaker
pipeline and how it sources the minerals inside of its cell phones, we're only just starting that journey in AI, right? And so yes, not everything is ethically sourced and not everything is, the entire pipeline hasn't been carefully, doesn't have the oversight to make sure that everything's being done as well as we would like it, but the industry is working hard at that.
00:48:05
Speaker
So if we presume we're going to get better at it, just like Apple has got better at building ethically sourced phones and we have better ethically sourced coffee, we will have better ethically sourced generative AI in my opinion. And that's going to be very exciting for creators at every skill level.
00:48:27
Speaker
I like what you said in the beginning because I've played around with these tools as well. I try and generate a piece of artwork in either mid-journey or Dali.
00:48:37
Speaker
see other people what they're doing. I'm just like, what am I doing wrong? And I think you've nailed it. It's like, you know, the right terminology, you know, the right types of brushes that you want to use, you know, the right pal. It's like, I can't tell you how many times I googled a hex code and be like, what colors go with this hex code? And like, I think as an artist or designer or a coder, right? There's this human touch that can't be replaced. And yes,
00:49:01
Speaker
I don't know if there ever will be a time when you really feel like this is perfect, this is it, but I think it's a great starting place.
00:49:20
Speaker
I love the concept of Gen AI in games in certain places. I still dream of days when I could talk with an NPC and maybe the NPC can take a look at what type of armor is he wearing? What type of weapons is he carrying? What's his name? What's all this stuff? What's all that stuff? And then almost you have this real time conversation. I don't know if we'll ever really get to a point like that, but I think there's a lot of cool uses for it that I'm excited for in gaming. And we'll see if something like that comes out. Who knows?
00:49:49
Speaker
I met a PhD student who was studying that exact problem. That's why I love my job. There's just a thousand really fantastic rabbit holes to go down. Even that MPC use case is fantastic. You woke up to the blacksmith. It's not your first time there. You've been there and upgraded your armor before.
00:50:10
Speaker
Then you went somewhere else and you got another piece of armor. Does the NPC ask you where you bought it? Is there some jealousy that you're shopping elsewhere? What do they know about the news? Oh, you slayed the dragon. Word has reached us. That was amazing. I can't believe it. Some of that obviously can be scripted and we've seen
00:50:32
Speaker
you know, again, artificially intelligent NPCs doing some of that. But to have a conversation, there's been some amazing demos of how that could work. Right now, the challenge is stopping it going off the rails, right? So you walk up to the blacksmith in the medieval village, and then you start asking it about which self
00:50:52
Speaker
phone coverage it has. And it tells you that it's recently switched from Sprint to Verizon because it's got an LLM back there somewhere that does know that these platforms exist and knows that people will switch back and forth. And so suddenly you've got this medieval blacksmith talking to you about why they switched their cell phone plan.
00:51:16
Speaker
That's a challenge that you have to overcome as you start to put those systems in place. Not to mention right now, it'd be pretty expensive. These models don't run for free. They're not small enough to run on your device yet. I'm getting there. There's some really interesting work in that space as well I'm very excited about on device AI. It helps with privacy. It helps with latency.
00:51:39
Speaker
So we could reach the point where there could be a very capable LLM running on your PlayStation 6 and it's able to give the NPC that level of
00:51:55
Speaker
you know, engagement. But then the other element of that is that there's still going to be an experienced designer who, you know, someone who's trying to craft a story and an experience for you. And so from their point of view, it's a little bit like having actors ad-libbing on stage, right? Maybe you want to allow it a little bit and some directors want to give their actors some freedom, but you want to be able to control how free they are when they're ad-libbing.
00:52:23
Speaker
There's
The Promise of On-Device AI
00:52:24
Speaker
a question then, right? The Samsung S24 Galaxy was just announced, and it announced it has an onboard AI chip. And I think a lot of people think that's gimmicky. And maybe at this stage, it is gimmicky. But do you think that in time, there's real-world applicability to that? I haven't looked into that chip. I think that's absolutely where we're going to be going. I think that's where we have to go.
00:52:52
Speaker
healthy, probably well-informed speculation about Apple secretly going to become a giant in the AI space, because from their point of view, they've been doing a lot of work to put a lot of the security, to put a lot of those engagement loops
00:53:11
Speaker
on the phone so that when you ask the personal assistant, it's never mentioned, but as you ask the personal assistant for something, they don't always have to go out to the cloud, transmit your private data to get you an answer. And when you think about a personal assistant, which for many people is like the ultimate AI aid, right? It's Microsoft's co-pilot. It's all about helping us do the things we wanna do.
00:53:40
Speaker
And for me, that private assistant that's helping me with my calendar and is helping me schedule meetings and do stuff and get through my day and remember to go pick up my daughter and those other three things I wanted to do today that I haven't got to.
00:53:55
Speaker
I'm overdue a doctor's appointment, whatever it is. That's all very private data to me. Some of that, at least, I'm not going to want in the cloud. And so it requires the actual model to be accessing private data that's only on my most private devices. And plus, from a latency point of view,
00:54:18
Speaker
from a privacy point of view, and who's going to pay to run all these super powerful personal assistants.
00:54:26
Speaker
if they're being run in the cloud, right? Now I have to pay a subscription fee. Whereas if it can run on here and I don't have to pay a subscription fee, that's also attractive. So I think on-device at-age AI is going to be a big thing in the next, I don't want to put time on it, but next couple of years, we're going to start to see some really powerful extensions of that AI right to the edge.
00:54:50
Speaker
Steve and I have two more simplest questions for you. One is a fairly straightforward question. With all this AI talk that we keep talking about, which AI assistant do you trust the most or do you use the most? I think I use Siri the most just because that's the ecosystem I'm in.
00:55:15
Speaker
my watch and my phone and my laptop, both all everything lit up there. If you dig into them, they can be very powerful. I dictate a lot of notes when I'm out on a walk or something and I'm listening to an audio book and I suddenly want to make a note of something. I can just tap the back of my phone to wake up the little script I've written, which then
00:55:42
Speaker
doesn't dictate, you know, I just say what I want to say, and it takes it transcribes it puts it in my note of the day. It's all automated, it all works. It's feels like magic every time I do it. And there's lots of little hacks like that. So I think just because it's the one that I carry with me on my wrist, that's the one I use the most. It's a good move. I invested in Alexa. And I think your previous comment where it costs, it's expensive to maintain a cloud AI.
00:56:09
Speaker
I think Amazon has kind of realized that, and Alexa's not quite as sharp, I feel like, as she used to be, and she's not keeping up with the competitors, like I feel like she should be. Last question, not really a question, but you will be at GDC. What are you gonna be doing at GDC? Well, I'm gonna be, I've got a couple of talks. Keywords is doing a couple of events that I'm gonna be at. I hope more,
00:56:39
Speaker
aiding conversations than talking. I'd like to hear from our customers and other people, but I'm not sure exactly how it's going to shape up, but yes, Keywords has a good presence at GDC this year. As always, I'm going to be there. I've been going for many, many years.
00:56:54
Speaker
For those of you that know me or those that don't, I very much like to take meetings by walking around the Yaba Garden, which is a lovely little park right above, kind of above behind the convention center. So as long as the weather isn't miserable, you'll find me doing walk and talks with customers and other folk there as well. So I look forward to seeing a lot of old friends and new ones at GDC as always.
00:57:23
Speaker
Awesome. And I heard a rumor that if you have an extra coffee and you give it to him, you can jump to the front of that line to get talked to him because we all need coffee every morning at GDC. But Steven, this has been a pleasure talking to you. We've learned so much. I am glad to be able to myself learn something here and share it with our audience. Before we go, is there anything else you'd like to share with everyone? Um, no, you know,
00:57:48
Speaker
Keep an eye on what we're doing here at Keywords. I think we do some exciting work and get out and play with some of those tools. There you go. They'll surprise you in many different ways. Awesome. We will have a link to information about Steven. You can check him at the PlayerEngage podcast website or check him out at GDC if you're going. Again, Steven, thank you so much for joining us today. I hope you have a great rest of your day. Thanks. This was great.