How will AI change games? How is AI changing games?
In this TechFirst we chat with Morgan McGuire, Roblox’s Chief Scientist and a former Nvidia research scientist. He tells host John Koetsier how AI is not only enhancing game creation through generative AI but also revolutionizing multiplayer game safety with advanced AI moderation systems.
We chat about the explosive growth of Roblox and share insights into how AI is shaping the future of interactive, social, and immersive gaming experiences.
Ultimately, McGuire says, we might be creating games with our voices in the not-so-distant future …
(Subscribe to my YouTube channel)
Subscribe to the audio podcast
Apple Podcasts
Spotify
Google Podcasts
Overcast
Other podcasting platforms
Transcript: making games with your voice, with Roblox chief scientist Morgan McGuire
This is AI-generated; it is not perfect.
John Koetsier (00:01.72)
How will AI change games? Hello and welcome to Tech First. My name is John Kutzer. The scope of the question I just asked is impossibly broad. AI is being used and will be used in many cases, have been used for a long time in all aspects of game development. We’re talking gameplay, game management, smart NPCs, endless levels, 3D interactions, faster development, more insane art, global communications with billions of players, regardless of what language they speak.
and importantly, democratization of game making. There’s so much. Here to chat about some of the biggest changes is an OG of AI. He’s a professor of computer science, a former research scientist at Nvidia. Tell me you kept your stock options. He’s a consultant at Unity. He’s the senior architect, was the senior architect at Oculus technology, and he’s the current chief scientist at one of the hottest game platforms on the planet, Roblox. Welcome Morgan.
Morgan (01:00.924)
Thanks John, it’s great to be here with you.
John Koetsier (01:03.87)
It is great to have you again. Last time we chatted, we were on stage at Collision in Toronto. What’s new since then?
Morgan (01:11.868)
So lots is new. One of the most exciting things for me at Roblox was you and I had talked about our 4D AI initiative we had just announced publicly at Collision. And this is the idea of going beyond 1D text, 2D images and materials, even the kind of nascent state of the art of can you make 3D models with generative AI, which is very much an emerging area, not an established category.
What we announced at Collision was going beyond that to 4D and interaction is that fourth dimension. This is how it weighs into games. So not just can you make 1D, 2D, 3D objects with AI, but can you bring them to life? And the exciting thing for me is that we’ve since shipped several of our initial features in that space. The most exciting one, I think, is our avatar auto setup, because it goes directly to letting anybody make their own 3D persona.
from scratch now instead of having to just choose from an existing catalog. So we get the full catalog, all the clothing, all the makeup, all the accessories. And if you can’t find what you want, you can make it now with AI. So super exciting.
John Koetsier (02:23.234)
This is super exciting and we’re going to dive super deep into that because this is like next generation, next level stuff. mean, like we’re getting excited about, you know, image generation. That was maybe a year ago, half a year ago. We’re getting excited about video generation, AI video generation, right? And so, and that’s, that’s still, you know, you see some cool stuff and you see some, wow, that went weird places, right? And you’re talking about like real time interactive multiplayer.
you know, stuff that works. So that’s super hardcore. We’re to get into all that. I want to hit some stuff first. One thing that I saw VC Matthew Ball, he wrote a post and wow, it was pretty impressive. Some of the numbers on Roblox, three hundred eight million monthly average users. That’s twice steam. That’s three times PlayStation. That’s three times the annual users of Switch. That’s five X Minecraft. Wow.
and more than twice Fortnite. And he said it’s likely Roblox has more monthly users than the entire AAA gaming ecosystem combined. This is astonishing stuff. Six billion hours spent a month on Roblox, that’s double what people spend on Disney Plus. And also the hours per daily average user are growing. So it’s not just numbers growing, but the amount of time that people are spending is growing.
growing in APAC, rest of the world, not just Canada, the United States, other places like that. Interesting fact he mentioned was that Roblox never contracted after the pandemic. mean, most games did, right? Games exploded as a category during the pandemic at home. Lots of time. What are you going to do? Play games. Roblox never contracted. And he said, your run rate spending should hit $4 billion this year more than any other game. Like, wow, what’s going on?
Morgan (04:15.548)
So yeah, it’s a very exciting time, I think, for the space, for the platform, for Roblox. I’ve really valued my past interactions with Matthew. We’ve definitely met on stage and interacted a lot. And I think especially his early writing on the metaverse, which is a term that I think in its classic sense, the sort of snow crash metaverse social 3D UGC interaction.
Roblox is definitely one of the most credible platforms in terms of trying to realize that science fiction vision. And so think Matthew is very educated in this space and does a great job in it. Setting aside his exact article, the directionally, yes, the excitement is real. I just pulled up, I was curious, I was looking at our daily active user chart from 2018. So well pre -pandemic.
to Q2 2024, it is basically a straight line. Like you cannot see a pandemic lip on it. So when you say, has Roblox has not contracted since the end of pandemic, we exited 2021 with about 40 million daily active users. We just wrapped a quarter at 80 million. So not contracted is a funny way of saying doubled.
So I think the important message there is not the 80 million daily active users, 79 .5 to be precise, but it’s the fact that that trend has been so powerful. And what’s most exciting to us inside the company is not the raw number. Our goal is a billion people connected with positivity and optimism and 80 million, 100 million.
John Koetsier (05:58.062)
Mm
Morgan (06:08.272)
whatever, like we’re, we’re, we are going for a good chunk of the world, experiencing a civil online, 3d social space together. So where we’re at now is only the beginning of the growth. The exciting thing is what you referenced. So our growth in Japan, in Korea, the fact that we are growing more rapidly outside of our sort of home base in North America, if us and Canada, where Roblox started.
The fact that we’re growing more rapidly outside of our classic demographic of under 13 is now 17 to 24 is our biggest demographic. To me, that’s the story, not the absolute number, but the diversity of users, the diversity across countries of experiences, kinds of 3D interaction. That to me is really fulfilling. That’s where we’ve always targeted. And we have such great traction going in 2024 halfway through, and it looks like our best year ever.
John Koetsier (07:06.84)
Amazing. And that upscaling in the age demographic bodes well for monetization as well. That said, we didn’t come to talk about numbers or monetization or other stuff like that. We want to talk about AI in games. And you’re working on some very, very cool stuff, image generation, translation, safety, which is critical, obviously, when you still have a lot of kids on the platform. And this 4D generative AI, where do you want to start?
Morgan (07:33.306)
Let me start with safety because that’s as a company where we start.
John Koetsier (07:38.008)
Go ahead. what do you, let me frame it a second because we’ve seen a lot of attempts to use AI to manage safety. A lot of them with meta, right? Facebook doing certain things and it’s hard. False positives abound. Other stuff comes through. People find interesting ways of spelling words or referring to things in cryptic ways. And it’s.
This is a challenging problem.
Morgan (08:09.852)
Absolutely. So safety is a form of security from a technical perspective. And security is always a case of the bad actors are going to keep upping their game and you just have to stay ahead of them. It’s not a static thing where you can release a technology and say, we’ve checked the box on safety. You have to keep iterating constantly. You have to stay ahead. And as technology evolves, it has good uses and it has less good uses.
Our job is to make sure that the good uses are outweighing, especially on the safety side. So a great example of a new safety tech that we just released, and we moderate everything on the platform with AI now, 100%. So video, audio, images, text, whole 3D experiences, avatars themselves, clothing. So it’s a really monumental task. It’s all the kinds of media you can imagine. We are moderating everything, not just text.
John Koetsier (08:44.258)
Mm -hmm.
Morgan (09:09.748)
And there’s all kinds of cute things people try to do to circumvent them. And one of the most powerful features we have is that we’re now using AI throughout that whole process. It used to be a collection of classic safety technologies. It’s called traditional natural language processing. So parsing sentences for nouns and verbs, taking speech and turning it into text and then moderating the text. But then you miss the nuance. You miss the sarcasm.
One of the most exciting things to me is not just that we’ve rolled out AI across the platform for safety, but that we did it for voice and we did it for voice for all ages. And this has been a holy grail of safety tech. And, you know, for, I would say for better or for worse, but I think for worse, traditionally the reputation in 3D spaces of voice chat is pretty bad. So mostly industry has not been able to keep up with moderating. It hasn’t been able to handle slang.
John Koetsier (10:00.184)
Mm -hmm. Mm -hmm.
John Koetsier (10:06.85)
Mm -hmm.
Morgan (10:07.104)
and hasn’t been able to handle just the enormous technical challenge of the bandwidth of millions of audio streams simultaneously and in real time, trying to moderate those. So that’s been a real challenge for the field as a whole. And I think the gaming segment in particular has done a lot of work on that has tried hard, but has not been successful historically. And so this was a challenge. It was one of the things that we set out when we founded sort of our R and D lab at Roblox when it came back three and a half years ago.
John Koetsier (10:16.78)
Mm -hmm.
Morgan (10:36.166)
was we said we want to do real -time voice. Ultimately, we want to do voice translation so you can speak to people in different languages. We want to ultimately do voice fonts, it’s called. So basically voice role -playing kind of transition. So I’m speaking in my voice, but you’re hearing a dragon. So it fits my character, right? So the avatar is not just the virtual 3D persona, but it’s also your voice. And that’s also important for privacy. We’re not there in either of those technologies yet.
John Koetsier (10:55.914)
Yes.
Morgan (11:05.692)
on the translation and the voice transformation. However, we do have research papers. We’re getting pretty close. It’s mostly about closing the gap on making it cost effective in real time. What we have, I think, really nailed is the voice moderation. So we have the world’s first voice moderation system that will monitor 100 % of voice chat on Roblox. This is deployed. And it works for all kinds of voices. Really difficult.
Most traditional voice stuff, if you look at even projects like, you know, big things like Siri and the Google assistant and Alexa, they’re trained on adult voices. They’re trained on adult voices in a noise free environment. You have a 12 year old, a 15 year old on a playground holding their phone. The microphone’s getting wind noise. So, so this is a really difficult, this isn’t the easy you’re standing in a sound booth trying to filter.
So we’ve cracked that case. solved that. So for all of our 13 and up is what we enable voice chat for today as an extra safety precaution. And if a user has enabled voice chat, goes through a moderation system and we’re able to flag all kinds of bad action there. And then we have a really nice way of dealing with it. Cause again, you know, that might be a 13 year old, right? This isn’t, they repeated something they shouldn’t have said. They said something out of context, think instantly banning someone from a platform.
is basically you’re inflaming a situation that didn’t need to be inflamed. And so we want to turn down the flame usually. And so what we do is when we detect problems, depending on the severity, we give you a little prompt just explaining, you hey, you’re in time out. Look, you might not have noticed, but the thing you said was not appropriate in this context. Here’s why we’re turning off your mic for 30 seconds so you can cool down and reflect on what just happened. And we’re going to turn it back on.
And if this happens again, it’s going to escalate quickly. And again, it depends on the kind of infraction. But I think this is really key is that we wanted to use AI not in a way where we would instantly be punitive, but in a way where it actually contribute to steering the community in a more constructive direction. And it’s a really effective way of doing it because it means that then we also have an opportunity to have an appeals process, to have ultimately human moderator if we want.
John Koetsier (13:13.602)
Mm -hmm.
Morgan (13:28.026)
and take a look at the exact case, the transcript, the context, the 3D. Because we want to be fair to everyone. And our goal is not to sort of have winners and losers, but to have everybody be a winner. And so we’re trying to move players who are in that category of saying inappropriate things into the category of being constructive and staying on platform.
John Koetsier (13:38.239)
Mm
John Koetsier (13:48.302)
So you blew my mind when you said that 100 % of the content on Roblox is being checked by AI. We mentioned already 6 billion hours a month spent. There’s people talking. That’s real -time interactions. That’s text, which is obviously easier. But it’s the things like clothing that you mentioned. Unbelievable. Do you have any metrics as to how successful you are or how much you catch or?
or anything like that, how do you gauge your success?
Morgan (14:20.614)
Yeah, it’s a great question. So that we have a tremendous number of metrics for looking at all aspects of performance and safety. And as I said earlier, safety is ultimately this game where people are going to keep trying to thwart the system. Sometimes playfully, they don’t know better. Sometimes harmlessly maliciously. They’re trolls. And sometimes seriously maliciously. They’re trying to do something really bad on the platform.
John Koetsier (14:47.864)
Mm -hmm.
Morgan (14:49.688)
excellent at catching people who are trying to do something really bad on the platform, preventing the harm. We work really closely with law enforcement on that. Obviously the exact numbers around that aren’t something I can share, but it’s sort of a point of pride to me that we are, I think, one of the absolute best in the entire industry at making sure that we protect our users, especially our younger users, and in some cases, specific groups that have not found a home online.
that have found that Roblox is a safe space for them. To me, that’s just a really values line, really important thing. And it goes beyond business operations to core values and about the kind of world we want to live in. So I can’t solve all the world’s problems, but maybe I can help solve all the world’s problems when they come to Roblox and give them an ideal world there. On the moderation side for the assets and voice,
John Koetsier (15:29.408)
Mm -hmm.
Morgan (15:45.616)
Basically, we are currently at levels that we measure as superhuman. years ago, we had large numbers of human moderators, and that was never desirable. So it’s obviously expensive, but more than that, it’s a job nobody wants. Yeah.
John Koetsier (16:00.5)
It’s an awful job. People get so, so damaged by it.
Morgan (16:05.404)
Exactly. So the idea is like no human being should be exposed to that kind of content. You know, even if they’re an adult, even if it’s their job, even if they have training and it’s just, that’s something that is a great use for AI, right? It’s taking a task that no human wants to do, automating it and in doing so protecting privacy and all that. But we deploy when we hit superhuman levels. So we don’t deploy when we say, 80 % of human, but there’s a cost savings.
We actually target saying, is this the best way to do it? And when it is, we’ll switch over. And so for us, there is a cost savings. It’s been great for our operations. It’s been great for scalability. It used to be as more, right? That doubling in the last couple of years, we had to double our moderators if we were doing everything manually with safety. And by automating, means that, we get these economies of scale from our supercomputing clusters.
John Koetsier (16:53.73)
Yeah.
Morgan (17:00.93)
And so it costs us less than twice as much to do the moderation, but we can scale without limit there. We can just keep throwing more machines in there. We don’t have to hire more and more employees. So we’re able to have sort of a stable size of Roblox, the company. We’re able to automate some tasks that no human wants to do. And we’re providing best in breed safety to the whole world. So win -win.
John Koetsier (17:23.254)
Love it. Love it. You reminded me of the Florence Nightingale quote there. can’t do everything, but I can do one thing. So we handled safety. I do want to talk about translation briefly, but you kind of already talked about it, right? Because you’re enabled. If you need to know, if you’re going to moderate, you need to know what people are saying. You need to be able to translate that. I want to make sure we leave enough time for image generation, which is critical for game making and democratizing game making.
and your 4D generative AI. Maybe briefly hit translation. You’re trying to build a community, a safe place for a billion people. That’s a lot of different languages.
Morgan (18:04.924)
It is. We operate currently in about 180 countries around the world. there are about 20 languages that we support, sort of supported, but there’s about 45 languages people speak. And in some cases it’s a little tricky because what’s a language, right? There are dialects, are sort of within a language, there’s all kinds of slang. There’s, you know, there’s Portugal speaking Portuguese, but there’s Brazil speaking Portuguese. And it’s a slightly different variation. It’s a different.
John Koetsier (18:16.974)
Mm
John Koetsier (18:24.301)
Yes.
John Koetsier (18:32.45)
Yes, Yep.
Morgan (18:34.842)
accent, and both of which are big countries on Roblox. that was one of our first languages that we did beyond English was actually Portuguese. Most companies go to French as their next language. But we happen to have a huge population of Brazil that was embracing the platform. said, OK, we want to make sure that all the safety works for them as well. so it’s definitely absolutely a lot of languages.
John Koetsier (18:46.476)
Yep. Or Spanish.
Morgan (19:01.116)
We tend to train in a small set of representative languages first when we’re doing our testing on AI. And then the next thing we do, and we make sure to specifically, we don’t just start with English, which is sort of the dominant language within the company. We start with about five languages and then we quickly spread out and make sure that everything we’re doing will transfer. to be clear about the voice, which is very exciting and we’re now looking at this for images as well.
John Koetsier (19:07.521)
Mm -hmm.
John Koetsier (19:22.51)
Mm -hmm.
Morgan (19:31.098)
For voice, we don’t turn it into text for moderation. We’re directly moderating the voice. So that’s how we get all of the nuance. And that was the technology that was great collaboration of R &D and engineering partners across the company. I got to personally work on it because it was one of the things that I was really excited about in terms of that safety, internationalization, and sort of hit all of my personal goals. We released that model, by the way, as open source. So you’re an open model.
John Koetsier (19:37.058)
Mm -hmm.
Morgan (20:00.422)
We have a technical paper that just came out. went to the inter -speech conference. We told the world how we built the safety system. And we thought, this is so important, right? Like I had said, the rest of the industry has not done a great job on voice safety. I think we had some really good breakthroughs. Let’s share this. So we took a version of the model, not the exact one that runs on Roblox. But we took a version of the model and released that to the whole world so that other companies can go ahead and use that because we think it’s too important.
to that kind of technology to ourselves. We really want safety to be everywhere, not just on Roblox. So for the translation, yeah, we try to do everything effectively native in its language. We don’t translate from media like voice to text. We also don’t translate Indonesian to English and then moderate. We do natively in Indonesian.
John Koetsier (20:50.167)
Yeah, that was interesting. I read your blog post about that and you didn’t do like a translation module of a language to a language. You did them all at once in some sense. Super interesting, super cool.
Morgan (21:00.463)
Mm
Morgan (21:05.178)
And it’s, as a scientist, this goes beyond the scope of what we need for the product in the short term, but I think in the long term is the right kind of thinking. You have to be careful when you draw analogies from AI systems to the way that humans think that there isn’t a lot of credible evidence that those are necessarily particularly similar, even though it’s a biologically inspired model. It’s pretty heavy on the inspired rather than the directly modeling.
But it is the case that when you build these systems, it doesn’t just help you solve the problem in front of you. If you do it right, and this is the point of R &D versus engineering, gives you insight about the problem itself. And that will help you solve the next three or four challenges. You haven’t just knocked one thing off. You’re understanding the space better. So what is the space of human communication? What is the space of safety moderation? And so I think one of the early
speculative insights that we’ve gained from doing a lot of this kind of safety and moderation technology is that there seems to be some commonality across the language. The transfer is really good between even very dissimilar languages and they can get extremely dissimilar. You can have languages that are tonal. You can have languages where the verb is at the end of the sentence versus in between the direct languages that have gender, languages that don’t have gender.
So I think we are starting as a community to learn a bit about language, about what it means to have a constructive interaction with someone independent of language. There’s some things that seem to be sort of deeply embedded in human culture. And I think this probably, you you mentioned you’re recently in Honduras. I don’t know if you speak Spanish fluently or not, it’s perfect.
John Koetsier (22:56.302)
Poco Spanish, Poco Spanish.
Morgan (23:00.378)
So it’s definitely the case that even in a language that you cannot communicate in very effectively, I’ve traveled around the world, it’s one the best parts about being a scientist, that I find that your body language, your tone of voice, interaction, the politeness, the exact details of etiquette vary across countries. But then in general, there is sort of a pan -human culture of I’m friendly, I’m polite.
I need assistance, I’m offering assistance. And I think we’re starting to see beyond what sort of you know, sort of linguistics has gotten us. We’re starting to see that because AI can process huge amounts of data, right? This corpus of all communication that’s been happening on Roblox this year, that’s even just the texting, that’s a trillion messages, right? Nobody’s ever been able to analyze data at that scale and put it into a single system. So we’re starting to get some insights about
it seems like we can’t exactly put a box and say, this is what positivity and civility is as a mathematical equation, right? That’s not going to happen. But we can definitely say this interaction seems much closer to positive and civil than it does to a negative interaction. And that’s really powerful. So I think the future of AI moderation is going to go beyond, here’s the exact terms of service, here are lists of appropriate words, things like that.
And in the far future, I’m speculating that it’ll be much more about in the way that a human does, looking at an interaction, looking at the participants, looking at their relationship. this is a parent and child. When they’re saying go to bed, that’s okay. But there are other cases where it might not be okay for someone to say go to bed or I’m taking you to bed now. And having a system that can understand that level of nuance, I think that’s the potential of AI. And that’s one of the things that we’ve never seen before in safety systems.
John Koetsier (24:58.082)
Well, that just blew my mind. You’re building a universal translator with emotional intelligence and contextual relevance. super, super interesting. Okay. We got to hit 4d generative AI because I mean, that’s kind of the Holy grail, right? You run a world, you run a virtual world. And if you’re going to have generative AI, sure. You can have generative AI for objects and clothing and you have that and we will get to that as well.
but everything has to exist, not just in a 3d space, but in a moving 3d space, not just in a moving 3d space, but an interactive 3d space where things change. Talk about what you’re doing there.
Morgan (25:39.708)
So the key to Roblox is growth, right? We had this runaway growth that started a little before the pandemic, went straight through and keeps going. And I mentioned this sort of, you I’m very proud of this 80 million daily active users, right? That was a nice milestone, even though it’s a small step along the way to where we’re going. That growth is fueled by having a great platform, running on every possible device. So we got Android and VR headsets.
and set -top boxes and consoles. But the content, the content is why people are there, right? It’s social interactions with their friends, both real -world friends and online friends, and doing something inside of a 3D world, having a great 3D evidence. So all of that content is what keeps players coming back. That amazing content, Roblox produces zero of it. So part of our community, and hopefully someday 100 % of our community.
creates content. It’s a user -generated content platform, UGC. And there are people who make their entire livelihood selling content on Roblox. There are people for whom it’s a fun pastime. So just like any kind of creation, there are people who make clothing professionally, and there are people who are knitting at home for their family. So that’s where the content comes from. It’s from our wonderful community. We have millions of creators. There are five million
active experiences every day. So there’s an even longer tale of total experiences, but there’s five million of them that somebody went to every single day. And so that’s incredible amount of content. As the user base grows, we need the content to grow. It has to grow roughly proportionally. And the challenge there is that Roblox has created with Studio, with our importers, wonderful best of breed.
John Koetsier (27:15.608)
Mm -hmm.
Morgan (27:36.188)
previous generation tools. So they’re the easiest 3D generation tools to learn, but it’s still 3D generation. You still have to know a lot about 3D modeling, about programming. It doesn’t matter that we have the friendliest programming language on earth. It’s still programming, right? That’s a real thing. You got to learn. AI came along, generative AI came along, and we started embracing it, started playing with it at just the right time in our growth curve, where we said, we’re going to have this problem of how do we produce
John Koetsier (27:43.01)
Mm -hmm.
Morgan (28:06.684)
And going back for roughly the last 15 years, the games industry and the film industry had exactly this problem of they couldn’t produce content fast enough. They were victims of their own success. And it was one of the things I’ve worked on in previous roles in the industry and given a lot of talks about, you just draw these curves and you say, Hollywood won’t be able to make a movie in 10 years because the budget of the movie will be the entire planet.
John Koetsier (28:14.158)
Mm
John Koetsier (28:33.422)
10 billion.
Morgan (28:34.492)
So everybody on Earth is going to be like editing pixels that Lucasfilm or something. So Roblox is in the same space. It’s still 3D content. It’s a different level of fidelity. It’s a different set of tools, but still 3D content. That’s hard. AI is the answer. AI is what enables us to lower the barriers to creation so that people with great ideas, the strength of their ideas will shine through, and they don’t have to spend one year, three years, five years
learning how to program, learning how to do 3D modeling, learning how to do rigging, animation. So our avatar auto setup already does this for you for the avatar, so for your character. And it’s amazing. It’s one of the most technical aspects of 3D content creation is making characters. They have so much nuance and you have to stay out of the uncanny valley and all of that. So we’ve automated that. You can go back and edit it if you are a professional. We’ll save you a ton of time. But if you’re not a professional,
John Koetsier (29:21.25)
Mm -hmm.
Morgan (29:32.548)
It means that you can go zero to avatar by yourself for the first time. We have our studio assistant. It’s a conversational AI. You chat with it, literally chat, microphone chat, and it helps you to build things. So you’re inside of our studio tool, but it’s doing a lot of the work for you. And I think we’ll get to the point where in the future you could, if you wanted, make a game entirely with voice. So by talking to the system, working with it over the course of an hour,
John Koetsier (29:36.56)
Mm
Morgan (30:01.454)
And it’s just, it’s up leveling what you’re doing. Instead of typing every single thing using your mouse, we can do have AI take your intent and produce it for you.
John Koetsier (30:12.098)
make an entire game with your voice. That would be mind blowing. That is mind blowing. Absolutely incredible. And it’s interesting as well because you talked about, hey, we need to grow the amount of content as the number of users scale. Absolutely. But people are also getting different perceptions of what a game should have, what a game should look like, how a game should operate, how rich it should be, how good it should be.
how interact, all that stuff. And so you not only have to create more, you have to always be upping the ante for quality also, correct?
Morgan (30:47.376)
Yeah, and one of the, and I think quality is exactly the right word. And I want to dig into a little bit about what quality means to us. Quality, one part of it is visual fidelity, but it’s honestly, it’s the smallest part. And this is really important to understand. And I think to understand why is Roblox so popular? Why is Roblox growing so fast?
Because if you look at Roblox on the lowest end devices, we’re really proud of the fact that we scale all the way down to 32 -bit, 2 -gigabyte Android, really low -end devices by the industry standards. Orders of magnitude below what my enthusiast desktop gaming PC is. So obviously, the visuals you get, if you have the latest e -sports gaming rig with a 4090 GPU rig, the visuals you get are phenomenal there from a AAA game that probably took
about 5 ,000 person years to produce. Most of the games I worked on in the industry were about that. So a few thousand people working for a few years. That’s very different than what you expect from the visual fidelity from the typical Roblox game, which is three person months of development. So really fast iteration cycles on Roblox. And yet the players are saying, they’re voting with their feet, they’re voting with their wallets, and they’re saying, this is high quality. This is in some sense, higher quality.
John Koetsier (31:59.714)
Mm -hmm.
Morgan (32:11.11)
than some of the AAA games or movies, right? That’s where a different generation is spending their time now is on Roblox. And it’s because it’s quality is visual fidelity, but it’s also, it’s the interaction. It’s the community, right? It’s having that positive supportive community around you so that they’re building you up instead of pulling you down. And it’s about new kinds of interaction. So one of the experiences on Roblox that’s been really popular lately is…
John Koetsier (32:13.516)
Mm -hmm.
John Koetsier (32:29.25)
Mm -hmm.
Morgan (32:40.304)
Dressed to Impress. And I started playing this with my team. On Fridays, we take our lunch break and we all play different experiences on Roblox together. And we try to get away from our recommendations, right? So we’ll use someone else’s recommendation, because we want to find things that are different than what we personally are used to experiencing. And we hopped into Dressed to Impress. it’s this, you’re thinking, I’m going to play a video game or it’s a video game that the marketing images make it look as targeted at female identified players.
John Koetsier (32:48.918)
Nice. Smart.
John Koetsier (32:58.53)
Mm -hmm.
Morgan (33:10.428)
I don’t know how this is going to map to my experience. And it was amazing. We loved it. We were going to play it for 10 minutes, and we ended up playing it for the full lunch hour because it was our favorite game that day. And what Dressed To Impress is, is you’re sort of in a mall with your avatar. And your avatar is kind of like a model catwalk kind of character. And you make an outfit to a prompt. The prompt will be fall fashion or something like
John Koetsier (33:37.728)
Mm
Morgan (33:37.828)
and you go and you pick out your plaid skirt and your nice hat and on your handbag. And then there’s sort of a community rated, the timer goes off, everybody’s avatar walks down the catwalk and everybody communally rates each other. And so you have, it’s kind of like a game show. It’s kind of like a party game, right? It’s very different than what you might think of when someone says video game. This is an area that we categorize as social role playing.
Morgan (34:06.266)
Also very different from what role playing traditionally means for video games, where it’s sort of you’re walking through a movie, but you have some choices. Role playing on Roblox is literally, you’re embracing a role. It’s kind of improv, right? It’s theater like. Dressed to impress is huge. Charli XCX just announced a collection on it after she had done her concert on Roblox. I honestly only know who Charli XCX is because she happened to have a…
John Koetsier (34:10.093)
Mm -hmm.
Morgan (34:34.448)
music video crossover with Billie Eilish last week that my daughter was showing me. But I’m informed she’s cool, unlike me. So big things happening, right? Lots of crossover from different media, lots of different worlds coming together. So Dressed to Impress, great example of something that maybe it’s a game, maybe it’s not a game, it’s some other kind of thing. But that’s the kind of diversity, creativity that to me is getting at.
quality, right? No one was offering an experience like that in AAA gaming. That didn’t exist.
John Koetsier (35:07.296)
And it can only happen, it can only happen in a 3d world that is social with elements. Yeah.
Morgan (35:12.27)
Absolutely. And where it’s user generated, right? A million people are making new ideas on Roblox every day. And if one of them is a winner, then it’s great for all of us. We all get to experience that new thing.
John Koetsier (35:26.168)
So cool. Very, very cool. Absolutely love it. Our time is pretty much done. Maybe let’s tie a knot on this and say, hey, if there’s one thing you could wave a magic wand at and fix in terms of AI, generation, and gaming in the next three or six months, what would you do?
Morgan (35:54.328)
One thing I would fix in AI and gaming.
Morgan (36:01.248)
I’m just saying my list is like I have a thousand things. So I’m trying to figure out which are the most important or the most interesting of them. So I think, well, I’m go with most important, but I’m gonna try and spin it as most interesting. So I think one of the biggest challenges in AI today sounds completely uninteresting when you work in the field, which is realistically,
John Koetsier (36:03.47)
Now we’re in prioritization.
John Koetsier (36:11.628)
Let’s go interesting.
Morgan (36:29.403)
AI is primarily not limited by the quality of our data anymore. 10 years ago, we were, we’ve now gotten really good as a field and Roblox as a company at producing high quality data. Roblox has been really leading about making sure that creators control the rights to their data so that we’re training on stuff that is on platform. It’s appropriate, but for which we’ve got an explicit permission from our developers to use. So that used to be the limitation.
John Koetsier (36:48.579)
Mm
Morgan (36:57.83)
Today, the limitation is actually machine efficiency. And so when we look at programs, there’s a technical term called Magurov complexity, is sort of the, good could it be, right? Like if you could press the file down, how small can it get if you squeezed out all redundancy? Computationally, like how many of these virtual neurons should you need for a task? And when we look at most of our systems,
A large part of our R &D is spent not on improving quality. We’re kind of there in quality. It’s on making the system efficient enough at that quality level to be deployable, to be economically sustainable. And so why is this interesting? Well, I think the world is probably pretty interested in power efficiency, right? In green computing. So that’s an area where every time we make something more efficient, we make our business better, right? We monetize better, our profit.
John Koetsier (37:47.885)
Mm
Morgan (37:55.42)
margin gets better. But we’re also consuming fewer resources to do the same task. And so that’s kind of a, that’s a net positive that I think if you look at the scale of computing around the world, if you look at the scale of where AI is going, it’s really important that we do it in a power efficient green way. And so I think if there’s one problem I could, I could wave a magic wand at, we know that it’s possible to, as a field to do AI about 10 times more efficiently than we are today.
We don’t know how to do it. It’s the cutting edge of research. is, this is a, this is a trillion dollar problem if somebody could come up with a solution, but it’s not going to be a silver bullet. It’s all kinds of different things. And it’s about really deeply understanding from first principles, how modern AI systems, how these deep neural nets work, why they work so well, what causes them to work better or worse. And so if I could wave my magic wand, it would be.
10x improvement in AI processor efficiency. I’m putting my money where my mouth is. It’s something we work on every day. I think the rest of the field is. But I think that’s a key thing for everyone in the world to be tracking is how are we doing as a field? How well are individual companies doing in that space?
John Koetsier (39:10.432)
Absolutely love it. Absolutely love it. Thank you so much for this time, Morgan. I really enjoyed it as last time and really appreciate your time.
Morgan (39:19.76)
Thank you, John. It’s always a pleasure. I look forward to the next time and I hope I’ll see you on Roblox too soon. Take care.
TechFirst is about smart matter … drones, AI, robots, and other cutting-edge tech
Made it all the way down here? Wow!
The TechFirst with John Koetsier podcast is about tech that is changing the world, including wearable tech, and innovators who are shaping the future. Guests include former Apple CEO John Scully. The head of Facebook gaming. Amazon’s head of robotics. GitHub’s CTO. Twitter’s chief information security officer, and much more. Scientists inventing smart contact lenses. Startup entrepreneurs. Google executives. Former Microsoft CTO Nathan Myhrvold. And much, much more.