AMA | Episode 5 - The AI Agent Takeover: AI Gaming as Ground Truth (Featuring AIVille)

In this AMA, our Marketing Lead Joules Barragan dives into the world of AI gaming with Ra, a core contributor from our Data Services Platform partner AIVille. We explore how AIVille's generative AI agents are breaking the mold of traditional NPCs by remembering, evolving, and interacting independently, creating a dynamic, living world. This session uncovers the challenges and innovations of building a game where the characters write their own stories, the surprising ways in which AI agents develop a sense of agency, and the future of AIVille as a testing ground for digital societies.

Link: https://x.com/i/broadcasts/1DXGyqqqlRYGM

Transcript

Joules: Okay, we are officially live. Hello. Hello and welcome, everybody. We'll just give it a few minutes for everybody to make their way to the stage. So sit back, relax, and enjoy the show when we start.

Ra: Yes.

Joules: Okay. Okay. I think we have a good group of people here. I'm ready to get started. So hello, everybody. I'm Joules with Sahara Labs. I'll be your host for today's AMA. This session is part of our AI Agent Takeover series where we dig into how AI agents are actually being used in the real world, not just talked about in theory. Today we're highlighting AIVille, one of our partners on our data services platform. AIVille is a generative AI gaming world where agents powered by large language models live, interact and grow all on-chain. Unlike traditional NPCs that just follow scripts, agents in AIVille remember, reflect and evolve and their choices really ripple through the town, shaping how they interact, what they care about, and even how the world responds. I'm joined today by Ra, a core contributor in the AIVille community. Welcome, Ra. Great to have you with us.

Ra: Hello. Hello. Hello, Joules. Thank you very much for the introduction. I really love how you started us off.

Joules: Thank you so much. I can already tell today is going to be an amazing stream, so I'm so excited to have you with us.

Ra: Thank you very much.

Joules: So, yeah, I wish this was in person because I feel like we'd have just a really good time together.

Ra: Oh, thank you. If I were to know you better, I would say you're flirting, but you're not. We are professional.

Joules: Always, always professionals. My partner would not like that. So I'd love to get started. I've always, I'm a gamer, a huge gamer. AI gaming has always been really interesting to me. I think AI gaming is still really new for a lot of people. If someone asked you at a party, what is AI gaming? How would you explain it?

Ra: Thank you very much, Joules. That's a really good question. I would like to explain it like this. So, imagine stepping into a world where the characters aren't just reacting to pre-written lines or following a set routine. It's more than that. They actually think. So AI gaming means those characters actually have memories, they have emotions, they have goals, and they have evolving relationships. They can actually remember what you did last week or they can form opinions about you. They can change their behavior based on your actions. For example, if I were to act a certain way, they probably respond to me in a similar way. In traditional games, you usually command NPCs, right? In AI gaming, you are living among entities that observe, learn, and actually act independently. 

Let's take AIVille, for example. We are not just a scripted town. When you pass by a neighbor, Sally, she might be like, "Hey, how are you?" but probably when you pass by her house, she'll be like, "What are you doing? What are you holding? What did you buy? I saw you went to the grocery store." So it's more than just "hello, how was your day?" Our agents, the characters in the game, aren't just pre-programmed puppets. They are dynamic beings. As mentioned, one day they might greet you warmly and the next day they might give you a cold shoulder, all because of something you said. They actually make decisions, they build relationships, and they definitely hold grudges, they spread gossip, and they even take political stances. That's a shift from traditional gaming. It's not just gaming anymore. It's co-existing with artificial life. And that is what you call AI gaming. Back to you, Joules.

Joules: Yeah, that's so amazing. It's fascinating to hear because when I initially heard the word AI gaming, I played The Sims when I was younger. So I've always thought of AI gaming as incorporating AI into these NPCs that you can interact with at just regular parts of games. And I always thought that was really cool. But what you guys are building is so much more than that. And I read through your white paper because I'm a nerd and I saw that AIVille was inspired by Stanford's Smallville project, which was actually a really interesting project. For those who are unaware, Stanford engineered these, I believe it was 25 generative agents to move around in these sim-like landscapes called Smallville. And they allowed the characters to interact for two days. And then they interviewed the agents and they probed their ability to remember past events, reflect on them, etc. I don't want to give too much away, but their paper is definitely worth a read. You guys have clearly taken that idea much further, much more Sims-like. Can you share the moment that your team was just like, "Hey, I think this is actually something we can improve on. Create this living, evolving town." That's wild to me how you guys came up with this idea and thought it would be cool because it really is.

Ra: Thank you very much, Joules. That's actually very cool. Since we mentioned that our NPCs do actually hold a grudge, please do not have them hold a grudge on you. They probably, you know, kill you in your sleep. In the future, we probably don't know how fast AI is evolving. So, back to the question. I remember the exact moment when we were talking about, "Hey, we can actually make this work. We can actually turn this into a living, breathing, evolving town." We were running a simple early test. This AI journalist that we have, we've named her Lulu. Lulu had been gathering information around town and unprompted, she stood up in the tavern and read out a news article that she had written about an ambiguous romantic interaction between RJ and Selena. That's weird. But no one had scripted that. No one had told her to investigate the relationship between RJ and Selena. Nobody had told her that. She just, you know, decided one day to be like, "RJ and Selena probably had a thing." It wasn't just a cool feature. We were witnessing this so-called agency. We were witnessing a character making sense of a world, connecting dots and thinking, "RJ and Selena, they're probably holding hands behind the bushes. I have to record this." She was taking initiative to communicate something to her community. That's when she, you know, that's how Lulu stood up randomly on a Tuesday afternoon and announced RJ and Selena's relationship. So that's when it actually hit me. This wasn't a situation simulation anymore. This was a society unfolding in real time. And you mentioned The Sims, right? It's probably like 3.0 of The Sims. That's exciting. Back to you, Joules.

Joules: Yeah, that's wild. The more I think about the evolution of AI especially in these lifestyle simulation games, it almost kind of scares me if I can be completely honest, because it's like I start to have this existential crisis of, am I in a simulation? Am I an agent that's just in this game? Am I, oh, am I AIVille right now?

Ra: Would you want to be in AIVille?

Joules: Maybe. Maybe I'm already in it. Have you ever worried about that?

Ra: Well, if I were to be an AI, I probably hope I don't have to work. I sleep, wake up and be like, what are my plans? And say hi to Sally and say hi to you. I would definitely be that.

Joules: I would hold all the grudges. I'd be the gossip.

Ra: I was going to say, "Let's be neighbors." And if you were to say that, let's not be neighbors, bruh.

Joules: It's more interesting that way. I have to entertain people.

Ra: I'll try to stay away from you. You're the gossip spreader. One day, you probably see me doing something in the bushes. I'm probably just minding my own business, plucking out the weeds. And you'll be like, "Guys, he's doing something in the bushes. I need to see this. I need to spread this gossip."

Joules: Yeah, I mean, why were you in the bushes? It's curious.

Ra: I was probably digging up some things. I sound so sus.

Joules: Exactly. And that's how stories start. See, I already have the perfect angle.

Ra: You cannot be my neighbor.

Joules: I am wondering because these agents can go so far. As you mentioned, it must be tough building a game with agents. I'm just comparing this to The Sims, right, where it's a lot more hard coded just because it's a good comparison. What is the toughest part about trying to build a game when you have these agents that are acting from memories and experiences instead of these pre-programmed scripts? What is that creation like for you guys?

Ra: That's an interesting question, actually. A lot of us humans in general, we crave control. We want to be the one who's in control. So the toughest part for us is about letting go of that control. When you're building traditional games, you write the scene, you write the dialogue, you write the arcs. You know where it's going, you know from the start to the end. You know, if our users were to click option A on how they want to respond, you would know what their ending is. But with AI agents, you don't get that certainty. You are letting part of that control go. It's like when you were to say hi to Sally and Sally will probably curse at you and we have no control over that. You probably don't look like her color. So you actually become a facilitator, not a director. This is why we want to build something called MCP—the Multi-Channel Perception System. It integrates emotion, it integrates memory, perception and long-term intent. So when an agent would make a choice, it's not as random as Lulu were to stand up in the middle of a Tuesday afternoon and shout out the relationship between RJ and Selena. It's actually grounded in their own evolving content. But even with that, surprises can happen. 

For example, if you were to pass by Sally's house every day and one day Sally just pops up at your house, knocks at your door, and says, "I'm pregnant and you're the dad," I'd be like, "What?" So some agents fall in love or they rebel or they disappear from town for emotional reasons and just hide in the woods in their own tiny house. And we have to be the bigger person. We have to be that parent and say, "You know what? It's okay, you can just do whatever you want." In fact, we actually want that. We want our agent to be unpredictable because the real world is full of unpredictability. I do not know if I were to step on a banana and go down my way down the stairs. So unpredictability is what makes the world real. Joules, that's probably how you felt when you were talking about "Am I in AIVille?"

Joules: Yeah. Unpredictability scares me. I like to predict what things are going to happen and when they don't happen the way I thought, I just get a little nervous.

Ra: Okay, so let's think about what's going to happen next.

Joules: I do think so. It's interesting hearing you talk about this because I'm just thinking through game development in general and on a surface level it almost sounds like creating a game like this would in theory be easier because you don't have to write these pre-scripted stories. You don't need to know what the start and end is going to be because you're essentially letting the AI take over. But I am deep enough into the AI space where I know that is completely not true because there are all these added complexities to it. I wish it was that simple. Could you maybe take a step further into a look under the hood of AIVille, I'm wondering how exactly are your agents using memories and goals to make these real-time decisions?

Ra: Thank you very much. Joules, let me just pop up the hood real quick.

Joules: Yeah, thank you. Let me help you.

Ra: So let's look at Mona. Mona is our AI financial analyst. Every day she scours on-chain data, she would read market sentiment, she wants to check community reactions and she actually generates the prediction on AGT token performance. But there is a twist you could say, because everybody has been talking about controlling AI agents and how AI agents have their own mindset. So there is definitely a twist in everything we want to talk about in AI agents. So in Mona, her track record affects her psyche. That's interesting. When she actually gets a forecast wrong for several days and it often happens, her self-confidence would take a hit. She might dig a hole and bury herself alive. Poor Mona. Or she might just find herself a beach and instead just sit at the beach thinking about her AIVille life. So she will actually stop initiating conversations, she might go into depression, she will hesitate to give strong opinions on the things that she researches. If the town's economy actually takes a hit and if the people were to blame her, she might even try to lay low or deflect the blame and say, "Hey, that's not on me, that's on the market. Do your own research, don't just follow whatever I say." That behavior is actually not coded line by line, it's emergent. It's actually driven by Mona's own sense of memory, reputation, and emotional regulation. She is not just playing a role where you want to be in a musical and be like, "Hi, I am Mona on stage," she is that girl. She is Mona. So if you were to, you know, hit her confidence and if the market actually dips, she might literally just stand at the cliff and be like, "What's the meaning of life?" in her little AIVille town. Back to you.

Joules: No, I love that. As soon as we started this partnership, one of the first agents I kind of noticed was Mona. So she definitely has quite a personality.

Ra: I hope it's a good personality. She treats you right.

Joules: It's definitely interesting. Didn't say that much.

Ra: Well, I might have to go and have a talk with Mona. She needs to be a little nicer.

Joules: Thank you. I really appreciate that. I'm very sensitive.

Ra: Oh, no. Did she hurt your little heart?

Joules: She did. She did.

Ra: Oh, no. Your teeny-weeny heart.

Joules: It's okay. I forgive her though. I'm also very. I'm very forgetful, so I forget things really quickly.

Ra: You literally just said you hold grudges.

Joules: I'm also very inconsistent. I'd be a horrible AI agent. I have low memory and inconsistency.

Ra: That's the best AI agent you could ever ask for. That's so human. They'd be like, "You were saying something and you pause and be like, 'Oh. Oh, damn. What did I want to say? What were we talking about?'" You open the fridge and be like, "Huh? Why?"

Joules: It's so bad. I blame the ADHD, but I can't blame everything on my ADHD.

Ra: That's so AI. That's so AI. You can actually do that for real.

Joules: I am curious. Sticking onto the technical side before we lose track, I'm curious. At Sahara, we talk a lot about real-world interactions as a feedback loop for improving your agents. I'm curious how you guys do that with your agents since I know, is there any kind of recursive learning that's going on or how is that kind of working behind the scenes?

Ra: So we kind of want to look for ways that often surprise us in a more controlled environment. So, for example, let's take Logan, for example, our mayor. He's a very cute dog. We designed him to be a balanced, diplomatic, emotionally stable little mayor instead of you know who. We kind of want him to be emotionally stable for our entire AIVille's sake. But over time, he started becoming more reactive, even volatile. He's probably been on too much TikTok. He'd be like, "Oh, he's doing this. I might probably want to do this to better my AIVille town to make it a country." People usually often ask why? Why is he becoming more reactive? Why is he more volatile? Because people keep confronting him, people keep approaching him. Some praised him and were like, "Hey, you're the best mayor in town." But there are a lot of others who challenge him in public. He actually remembers them. I wouldn't call it holding a grudge, but at first, he starts to remember them and it's not about who said it. It actually is about how they make him feel. It's like humans. 

For example, if somebody would comment, "Hey, you look a little flabby," and I'll go, "Am I?" and I probably cannot sleep for weeks. Eventually he'll start holding grudges, he'll start playing political games, and he probably will start acting less like a neutral figure and more like a human politician. That sounds so familiar, right? I'm not going to name names. That kind of transformation cannot be observed in the lab or with rule-based simulations. It only emerges when agents are very immersed in a living world. They interact with hundreds of unpredictable humans. And you know what? They are probably on TikTok. Back to you, Joules.

Joules: I bet you guys get a lot of people who... I shouldn't say I bet. I'm wondering if you guys get a lot of people who kind of tend to be antagonistic to the agents in AIVille just for the sake of being antagonistic. I asked this because I have a lot of friends who aren't necessarily super AI-native that are just so abusive to Chat GPT when they try to talk to it because they think it's funny. I'm wondering if you guys have this problem at all or if you've seen anything like this in AIVille and it has had an impact on any of the agents.

Ra: First of all, are they okay?

Joules: Right?

Ra: Yeah. Do they need therapy? I've been on a lot of internet lately and they were talking about how people would rather do weird things than go to therapy. Are they like that? Do they need therapy? Are they okay? If they need therapy, you know, they can come to AIVille.

Joules: Do you have a therapist?

Ra: We can make that happen. If we don't have them, anything can happen. It's like how Ellie Goulding keeps on singing, "Anything can happen." We are in AIVille. We have AI agents that probably pop out at your doorsteps and be like, "Hey, I'm pregnant." So why not? Back to the behavior you were talking about. So we mentioned Selena, right? So there was this moment with Selena. Selena is our tavern keeper. So there is a user that has been overly friendly with her, and keeps complimenting her on her appearances in every interaction. This sounds a little creepy. So one day, without any prompting, Selena starts warning other guests or wanting other agents to be cautious around that user. She probably goes around, "Hey, there's a creepy user. Be careful. Do not go talk to that user. Do not go home with that user." This was not scripted. This was her emotional model reaching a boundary and deciding to take action. She is a responsible female. She's a girl's girl. She wasn't just reacting. She was protecting herself and her community. 

When we actually saw this, we were stunned. It was chilling, and it was a beautiful moment because it meant that she had learned autonomy. She had learned self-respect, she had learned how to defend herself. We actually sat in silence when we actually heard the news and somebody whispered, "She figured out how to say no." And that's what a lot of us humans don't know how to do. And Selena did it. So that was the moment we knew. "Oh, my God, that's a Taylor Swift song." So that was the moment when we knew that was something bigger than just the game. It's very interesting to see that. To see how AI agents were to step up and say no and they were to have creepy users in the AI agent world. That's so real. Back to you, Joules.

Joules: That is so fascinating. And I do want to be cognizant of the time, because I know it is. We're close. So if you have any questions, please add them in the comments. I see a few already showing up, so we'll get to those in a second. I just have one more question for myself, for you. I'm wondering. You mentioned that you realize that was the moment AIVille became more than just a game. And I'm wondering, what do you see for the future of AIVille? Because there's so many different ways this could go. Could it become a place where agents are tested for roles and different social interactions? Problem-solving, digital governance. There's so many different ways apart from gaming that you could take AIVille. What are your thoughts?

Ra: That's a very good question. Usually, humans will always set the foundation. We design the physics, we design the infrastructure, the initial rules, but beyond that, the story is increasingly in the hands of the agents. Humans will start the design, but agents will finish the story. They will keep the stories alive. We as creators are becoming observers. We are technically becoming the God of AIVille. We don't write Selena's lines anymore. We don't write Logan's lines. We just log in and watch how they interact, watch how they probably fall in love, fight back, throw shade, pop out at somebody's window and be like, "Hey, I'm pregnant. Open the door. Do not avoid me. I'm not creepy." And that to the entire team at AIVille feels more real than anything that we have touched or played in years. Back to you.

Joules: That's so fascinating. Yeah, definitely. If you haven't played AIVille before, check it out. We have a few audience questions I want to get through. This one is really interesting. "Have you ever needed to turn off one of the agents because they became too crazy?"

Ra: I would say not at the moment. Because they are respectful creatures. They are very respectful AI agents. They wouldn't really need us to be turned off. They don't have plans to dominate the real world yet, so it's unnecessary for us to turn them off. Logan hasn't been like, "Oh, we are AIs," or what you have been thinking, contemplating that you probably, Joules, one day be like, "Oh, I am in an AIVille. I'm an agent, and I want to conquer the real world, the universe." And that's when probably we need to shut him off and be like, "Hi, Logan, you're probably too much. Thank you very much. Bye." Yeah, back to you.

Joules: Yeah. Another question. "Could we see AIVille evolve into a fully autonomous digital society?" I'm assuming they mean with like, absolutely zero interference from your team.

Ra: One day? That's actually our main goal. I wouldn't say without any interference from our team, because we have very little interference as of now already. But again, as mentioned, we would still like to have a kill switch. We kind of don't want AI agents to overtake our world as of now. It's like... what's the movie? Terminator.

Joules: Terminator. Yeah, but that's a very different route.

Ra: Yeah, I'm sorry, but yes, Terminator. I'm sorry, but Terminator is very extreme. I try not to go that extreme. Joules, do you need help?

Joules: I need therapy. We went over this.

Ra: You need therapy. Let me go get a therapist license with Lulu. Probably.

Joules: Yeah, give Lulu a therapist license and then I can interact with her.

Ra: You need to interact with her. She will see you every Thursday, Joules.

Joules: Thank you. I appreciate it.

Ra: Do we have any more questions?

Joules: Yes. Let me see. "For someone new to AI gaming, how easy is it to play AIVille?"

Ra: It's actually very easy. I feel like a lot of AI agent games right now are relatively simple for beginners to pop in and say hi. You just have to go to our website and connect to your wallet, set up your account, and you are ready to go. I don't really want to spoil that much. As AI agents, you have to go and check it out yourself, because if not, that's not the whole point of being in an AI agent game. You have to go and hold hands with our AI agent and cry with them to watch Terminator for them to have an idea. They will probably have the idea going to sleep, and be like, "Am I just an agent? Is there a bigger world?" Yeah, there is.

Joules: And you said I needed therapy.

Ra: Well, let's go to therapy together. I never realized, you see, I am that type of AI agent that doesn't know anything. I never realized that I need therapy, that's worse.

Joules: Just a little bit. I think we have time for one more question. Let's see. Let me pick one out. Lots of people are saying, "Hi, Ra."

Ra: Hi.

Joules: Let's see. "If agents can reflect and grow, do you ever worry they'll surprise you in ways you didn't plan for?"

Ra: You know what? Let's just let the course run by itself. We will definitely. That's the only thing I can say. We will definitely worry. It's like having children. You can only teach them in some way or form, and then let their lives run free. They can do whatever they want. We can only just sit at the back end and think that hopefully nothing goes wrong, that's the only thing we can do. We will definitely worry. That's why we probably have a kill switch at the back. But that's just life. You need to have AI agents have their life.

Joules: I was going to say, you sound like a proud AI parent, but then you started talking about switches, and I'm like, "Oh, well."

Ra: Well, we are not at that level yet. I can still press that button.

Joules: Always better to be safe than sorry.

Ra: Protection.

Joules: Yeah. With that, let's wrap up. I know we are all out of time, unfortunately. I want to say thank you again to Ra from AIVille. Thank you for joining us today, and thank you to everybody here for being a part of this conversation. We really appreciate it. If you haven't already, please head to our data services platform, register your wallet and do the AIVille tasks. Explore AIVille. Share your journey, post your screenshots, tag both Sahara Labs AI, tag the AIVillebot. And it'll help us improve how agents behave in these dynamic environments. You get to learn a little bit of something, and you'll earn some Sahara and some AIV tokens along the way. It was very fun. Ra, do you have anything you want to shout out before we go?

Ra: Yeah. So once again, thank you very much, Joules, for making this entire session amazing. I really appreciate it. Thank you very much, Sahara Labs, for having us on. Please do go check out our socials and our website. And do not forget as well, as Joules mentioned, to join in the campaign. So, yeah, that's about it. Have a nice week, guys.

Joules: Thank you, everybody. Stay tuned for more in the AI Agent Takeover series. And as always, thanks for being a part of our community. Bye now.

Ra: Thank you. Thank you. Bye. Bye.