Artificial intelligence is undoubtedly one of the most fascinating contributors to the evolution of the NFT space.
Image creation programs (like DALL-E and Midjourney) that turn text-based prompts into images have become increasingly user-friendly and popular in recent months. NFT projects that use AI to contribute to the creation of their collections significantly are likewise becoming more commonplace.
But, depending on who you ask, the very concept of “AI art” is either a joy, a conundrum, or an insult to artistic disciplines and the creatives that drive them. However, the potential for AI to affect Web3 and NFTs goes far beyond the creation of digital art.
Altered State Machine, a platform and protocol that allows users to build, train, and trade AI “agents” as NFTs, is one company aiming to be at the forefront of AI in the metaverse. And this could inaugurate a new wave of substantial changes.
Giving NFTs AI intelligence
Like many in Web3, ASM is excited for the future of these digital environments, and one of its goals is to help create metaverse spaces populated by AI agents that can compete with each other, interact, and support human actors. But creating AI characters that result in a more versatile, living, and breathing metaverse experience is just one of many potential applications for the ASM protocol.
ASM’s AI agents can adapt to a wide variety of use cases. And the scope of applications is broad, whether the agents are used for gaming, open-world metaverses like Decentraland, or financial use cases.
What an AI NFT will actually look and act like depends on a few things. The agents of ASM’s protocol are composed of three items: a “form,” a “brain,” and “memories.”
ASM’s Artificial Intelligence Football Association, a metaverse game in which AI teams compete against each other in soccer matches, is an excellent example of how the technology functions. The game can be populated with forms, in this case, NFT characters called AIFA All-Stars. The NFTs’ abilities are randomly determined by their internal AI ASM Brains. In fact, the 40,000-character NFTs in the AIFA All-Stars collection were the first to receive what the company calls Non-Fungible Intelligence from ASM’s Brains during the company’s genesis drop in October 2021.
So, you’ve got NFT character forms powered by an AI brain. The last component is the AI brain’s memories, which store behavioral strategies that the NFT character learns through model training.
The AI brains do this training both in the environments that they play in (like a soccer stadium in a metaverse) and in “gyms” dedicated to improving your AI’s abilities. These gyms are networked GPU cloud computing providers that run training algorithms for a specific ASM application. In our AIFA gaming application example, this might look like improving your character’s shooting ability. If your AI is being used for the DeFi markets, it might look like adjusting risk tolerance for a trading bot.
Whatever the particular use case, owners can pay for training time in these gyms using ASM’s native token, $ASTO. The company also provides opportunities for hardware owners to create and profit from GPU time given to gyms.
Each ASM-powered NFT is unique in its own way, from its visual characteristics to how it reacts to its environment. This takes things a step beyond what we currently see in the NFT ecosystem. While generative PFP projects randomly assemble character accessories and other visual attributes, the ASM’s AI agents have randomly assigned behavioral tendencies in addition to their “physical” features. Those randomly assigned characteristics will show up in different ways depending on the use-case context of the NFT.
Once trained, users can trade or sell improvements. Commodifying these traits could help create an economy across different metaverses and in the gaming and financial markets. “Training depends on the use-case,” explained David McDonald, CEO of Altered State Machine, in an interview with nft now. “[One of the things] that determines what they learn is how we build a specific environment.”
The AI agents learn within the parameters of their environments, like a metaverse soccer stadium. After this training, the same agents can exist and interact in different metaverse worlds as well, provided that the environment they learned to play soccer in is replicated in those worlds.
There are several other forms of composability these AI agents retain. Composability is a system design principle that relates to a system’s ability to interact with another. In this case, an AI can learn a new “skill” to adapt to a new environment.
McDonald added that one of the more advanced forms of composability in these AI brains might be something like locomotion.
“Learning how to create a unique way of moving around the world for an agent can be a composable element that you can put into game or NPC (non-player character) experiences,” McDonald elaborated. “You could pair that up with a natural-language processor, for example, that allows for chat functionality. Or with a navigation AI that allows for the AI to learn how to navigate their world. All these elements within the brain you can put together in interesting ways to create interactions and life within environments.”
So far, metaverse spaces rely on human actors to inhabit and enliven them. But a significant feature about these worlds, McDonald believes, is what goes on when you’re not in those environments.
“Let’s say you have a game commerce store set up [in the metaverse] and you create a unique way of selling your digital goods in that space,” McDonald said. “When you’re not there, you kind of lose that capability. Being able to have your assets take over when you go to the beach with your kids or whatever it is, is one way [ASM] can have an impact as these metaverse spaces get built out. The other way is that you can craft experiences for a lot of people that a single human can’t possibly do on their own. You can set up commerce around digital interactions and scale them in a way that isn’t possible with a single person or even a big team of people.”
Muhammad Ali – The Next Legends metaverse boxing game
ASM’s most significant upcoming project is Muhammad Ali – The Next Legends, a metaverse boxing game where characters use the company’s AI brains to hone their skills and fight other AIs.
The project, scheduled to be released sometime later this year, is a partnership between ASM and Authentic Brands Group, the company that owns Muhammed Ali Enterprises in collaboration with Lonnie Ali, a trustee of the Muhammad Ali Family Trust. The game will feature character designs from Web3 creative factory Non-Fungible Labs.
Like AIFA, The Next Legends will feature NFT characters that contain ASM’s AI brains. Players will train their characters, all of which come with a randomized set of skills, to better develop the abilities (like jabs, uppercuts, and stamina) needed to win these matches.
McDonald is particularly thrilled about the partnership and the project, saying that he’s looked up to Muhammad Ali his whole life.
“It’s a great honor to be part of extending that story and his legacy into a brand-new space It’s an opportunity to create a really exciting and novel game mechanic and way of interaction with stories that I think people are going to really enjoy. And setting us up for bringing a whole new range of experiences in this space to Web3 and the metaverse.”
When asked how he feels about using the game to welcome people into the Web3 gaming environment, McDonald emphasized that it’s crucial to satisfy the needs and interests of both the Web3 savvy and those new to the space.
“We need to make sure that it’s user-friendly for anyone at any skill level to participate in the game. I know people are already familiar with the current user experience of Web3, and those people are in the category I like to call power users,” McDonald said. “They have done the study and they probably understand the ASM protocol and they understand how gas works and how blockchains generally work. But the average user who might be interested in playing a Muhammad Ali AI Boxing game probably doesn’t have that level of sophistication. So, we’ve got to cater to both of those entry points and make sure that onboarding experience is not a barrier to entry.”
Putting AI back in the hands of the people
Overall, the implications ASM’s technology border has for Web3 are exhilarating. Keeping with the Web3 ethos of ownership, collectors of these agents will have proof of ownership of an AI whose evolution and purpose they have control over. McDonald contrasts this with how AI in Web2 works, explicitly noting how algorithms in social media are completely out of the user’s control, which can have profound effects on a person’s psychology and mental health.
“One of the big problems with preference algorithms today is, and we’re seeing this with things like TikTok, that it’ll take you down a rabbit-hole about what it knows from your behaviors, but that might not be what you’re trying to target,” McDonald said. “That kind of algorithm has a lot of negative results when you look at things like showing depressed people content about depression. [It’s] sending them down this rabbit hole they might not actually want because the algorithm knows that’s where they’re getting the highest engagement. Having self-sovereign control over preference AI is, I think, going be an important thing for people.”