AI + Crypto’s First Creation Is the Multiverse

The first time I got the call was a little over a year ago. Isaac, Founder & Creative Director at Monster Truck Ninja, a longtime Saga Innovator: Maybe I can do a crossover collab with another Innovator? I think my Bacano Go characters would be cool in these games. Sure, that’s a great idea.

Two weeks later, Leo Simon at Ampy, another Saga Innovator: Can I contribute my music NFTs to the games and short films on Saga? Music is usually the hardest part of any production. That’s very true — and why not?

The ideas sounded like benign business proposals, until I realized how unusual they were. Saga was not a studio. We had never commissioned content, nor had we ever offered our gaming and entertainment Innovators more than passing notes to their rough cuts. And yet, here we were being asked to facilitate a creative process, much as if we were a full-stack production studio.

Around that time, ChatGPT-3 came out, and the requests and offers began to flow. Only weeks prior, few projects had heard of, let alone used, AI tools like stable diffusion and large language models (LLMs). Now, everyone had at least experimented with Generative AI, and many resourceful creators and developers were actively using it in their production process and to extend their content universes.

More importantly, and this is where Saga stepped in, our Innovators had the urge to contribute to each other’s universes, as projects and individual creators and storytellers. However, they could only produce so much with the technology available at the time. Now that AI was freely accessible, they had the bandwidth to scale their creative presence while maintaining ownership and control over content through web3 rails.

Whether the Saga Multiverse had just been born or existed all along — no one had time to think about it. We only cared that it was here, and the raw power contained within it would change everything.

AI + Crypto: Please Make This Happen

During the great bear market of 2022–2023, crypto was a distressed asset looking for a white knight. VC funds that had previously started to allocate heavily into crypto switched their focus, and massive valuations, from crypto projects to AI companies.

To regain relevance in the narrative, crypto tried to marry itself to AI — a largely unconvincing effort given the other party did not seem as enthusiastic. The story went something like this: AI is awesome but cannot be trusted, and the blockchain ledger is the only way to make AI trustworthy in the long run. In an interview with The Block, Nansen analyst Sandra Leow said, “Both AI and crypto are quickly maturing, and while the potential of artificial intelligence and blockchain integration remains largely uncertain today, the emerging use-cases and possibility enabled by both technologies is massive.”

Most of the use cases proposed in the crypto space center around AI agents, on-chain bots that operate on AI logic based on ongoing human input. As AI becomes more prevalent in daily life and business functions, these AI agents become more sophisticated and cheaper to run. To be able to accurately track their activity, adjust as necessary and ultimately hold them accountable, the logic that drives them must live on chain, and they must communicate with one another in a trustless decentralized system.

Did you catch the assumption there? In order for all of that to be true, one has to presuppose that non-blockchain rails cannot be trusted as a means of executing the actions of AI agents and allowing them to transact with each other. The primary situation in which that holds is one where the non-blockchain rails have a greater interest in preventing the communication between AI agents than becoming part of a network and leveraging network effects.

Let’s see how the suggested use cases stack up to that criteria so far: AI-Driven Smart Contracts, Secure and Decentralized Computing Power and Data Storage, Training GPTs on Blockchain Data, Transparency Through On-Chain Language Models, Authenticity Verification on the Blockchain to Combat Disinformation. All of these proposals are in the family of increasing the automation, transparency and safety of AI. While it may be true that the enormous challenge of integrating AI and crypto is still largely a technological one, there’s a reason why very few are clamoring to execute on these use cases, and why they are still ideas without proven product-market fit: the trust issues with AI remain at the algorithmic and machine learning level. There’s nothing blockchain would be able to do to meaningfully abate that. To focus the solution on blockchain means to a certain extent that we’ve done what we can to make the algorithms and the companies that develop them as unbiased and incorruptible as possible, and the only hope to keep them in check is the open ledger of the blockchain. We’re not there yet as a society.

The crypto industry remains united in its confidence we’ll get there. Jeremy Allaire, Co-Founder & CEO of Circle, says that AI “and blockchains are made for each other,” with the tech suited to “machine-generated and enforced contracts” and “machine-to-machine value exchange.” Brian Armstrong, Co-Founder & CEO at Coinbase, told Decrypt, “It’s hard to imagine a world where different AI agents are paying each other with Visa, or something like that.” They will likely use “a digitally native currency like crypto.”

We’re so early is a favorite copephrase in crypto. It is a meant as both an argument for crypto’s significance as much as encouragement for the teams working here that despite their anemic user numbers, there is something worth building. One thing crypto projects are never too early for is speculation, and that is where the numbers might be the most sobering. The token prices for AI-related projects as a whole lag far behind the market. While the technology may be a worthwhile project, the market is simply not there yet.

So why don’t we return to first principles: focus on the user. Who wants AI agents to communicate via blockchain? It would have to be the humans who deploy the agents on their behalf. And why would they want this? AI Agents to date have been primarily used as extensions of their human principals. Their mandate is to be as free and sentient as their principals would like to become through the deployment of agents. Sentience is achieved internally; freedom traverses. So we’re looking for users who want to take their agents beyond the walled systems from which they originate. But to bring this use case into being, they can’t simply want it: they have to be impatient for it. Who is building this without waiting for permission from any of the powerful institutions that currently have the wherewithal to deploy large AI-based systems? It is the same users who agitate against the constraints of these institutions to begin with.

Completely oblivious to any tech narratives, they have been simply creating the world they want to live in.

Enter the Multiverse, where people use technology to become their real selves.

The Metaverse: What You Consume, The Multiverse: What You Create

The Multiverse is an old concept, originating in physics and cosmology to describe the existence of multiple parallel universes beyond what any inhabitant of one universe can observe. Since then, it has inspired countless creative interpretations. To define it, I can’t do better than George FR Ellis, so I won’t try:

The word “multiverse” has different meanings. Astronomers are able to see out to a distance of about 42 billion light-years, our cosmic visual horizon. We have no reason to suspect the universe stops there. Beyond it could be many — even infinitely many — domains much like the one we see. Each has a different initial distribution of matter, but the same laws of physics operate in all. Nearly all cosmologists today (including me) accept this type of multiverse, which Max Tegmark calls “level 1.” Yet some go further. They suggest completely different kinds of universes, with different physics, different histories, maybe different numbers of spatial dimensions. Most will be sterile, although some will be teeming with life. A chief proponent of this “level 2” multiverse is Alexander Vilenkin, who paints a dramatic picture of an infinite set of universes with an infinite number of galaxies, an infinite number of planets and an infinite number of people with your name who are reading this article.

Over the last few years, the Multiverse has started to pick up steam in pop culture. Everything Everywhere All at Once celebrates it in all its chaotic glory, and Disney is prolonging the life of the Marvel franchise by making as many alternate storylines about our beloved superheroes as possible. Rather than cause confusion, these blatant agitations against the indivisible self are fully embraced by the viewing public.

We are at a point in time where one self is simply not enough. The staleness of the Metaverse, where our digital selves were originally supposed to live, makes this all the clearer. Over the last five years, it has only ever been populated by content from other people. Those who visited the Metaverse could only consume. The Multiverse requires imagination and therefore invites creation. Not only has it remained relevant for far longer than Metaverse, we’re only scratching the surface.

Web3 Gaming Multiverses

The primary kind of digital world that survived the great purge of metaverses a couple of years ago is the game, and the MMORPG (Massive Multiplayer Online Community Role Play Game) in particular. They have all self-enhanced with AI. Wilder World, an AI-based world reminiscent of Grand Theft Auto, has significant AI elements in its characters and gameplay. Abyss World was originally built to accommodate several Realms, and AI is used both to enhance the art direction and game design of its highly experienced game development team and to grow Realms into their next iterations. Angelic has many of the same elements but emphasizes the lore and storytelling around its characters more so than its peers. Mythic Spell Book GDD seeks to allow its gamers to escape to a world very much like the one they inhabit in physical reality, complete with urban economies, professions, distinct neighborhoods and a hierarchical community.

Building off an enormously successful card game, Parallel has revealed work on Parallel Colony, an extensive simulation game with significant Generative AI elements for its original characters. It has been a thought and design leader in the use of AI Agents in gameplay, owing to its near fanatical devotion to its community. Every technology design decision at Parallel has been made in service of handing as much control of the game as possible to the players, whether that is in design of game elements or monetization opportunities in the in-game economy.

It’s not simply a matter of ambition that so many major web3 games aim to be worlds of their own. In many ways, web3 architecture was made for the next generation of gaming.

Web3 solves three major problems in traditional gaming: lack of robust in-game economies that are transparent and interoperable, gaming guilds rampant with coordination and governance problems and limited extensibility of games given that mods are still beholden to the original publisher. Each of these problems only besiege games that have the breadth and community necessary to sustain a system of lore and huge catalog of in-game assets that gamers find valuable enough to trade. In time, even the simplest arcade games and Candy Crushes of the world can benefit from web3 gaming elements, but the early breakthrough titles are solidly their own universes.

There remains, however, one pain point that web3 games have not been able to address with existing infrastructure, and that is the problem of interoperability. Now, to be sure, interoperability is as much a technology question as it is a commercial arrangement. Games do not have to accept in-game assets from other game versions, let alone from other titles. However, there is a major push from gamers to truly own the assets they have played so hard to earn. If they are unable to take it to other ecosystems or marketplaces, then there is no true ownership: we may as well be in the traditional gaming model in which the publisher has full control over utility and extensibility of a game. We will explore this later, but suffice to say for now that it is serious enough of an issue such that web3 gaming is truly incomplete without it.

Not Every Creator Is a Gamer; But Every Gamer Is a Creator

Long gone are the days where gaming was for a niche, nerdy audience with very limited mainstream appeal. Last year, The Economist devoted an entire issue to the gaming industry, probably owing to the fact that, “Consumers are forecast to spend $185bn on games this year, five times what they will spend at the cinema and 70% more than they will allocate to streamers like Netflix. Once a children’s hobby, gaming has grown up. Console players in their 30s and 40s now outnumber those in their teens and 20s.”

The staggering growth is partially because the lines between traditional entertainment and gaming are blurring. Pop stars will perform concerts in games. The Walking Dead had its season coincide with a massive game activation on Facebook. The gaming industry came into the crosshairs of the SAG-AFRA strikes last year because the actors were concerned about the AI-based manipulation of their voices and likenesses in games. Many recent hit movies and TV shows began their life as games, including The Last of Us and The Super Mario Brothers Movie, further leading to renewed popularity of the original games themselves.

However, this only accounts for the big studio-backed productions. The far bigger growth trend is in UGC (user-generated content): skins, avatars, mods. Roblox is perhaps the most famous example of this, but mostly because it hit at the right time: customization and personalization had become mainstream, and children (with their parents’ credit cards) were able to spend. Before Roblox came Fortnite, Minecraft and League of Legends, just to name a few, where teenagers and adults have long cut their teeth creating custom content that could be sold to their fellow gamers. Modifications, or mods, made by gamers able to alter the source code, are now a standard way to make games extensible without the original studio expending any development or production resources.

Pandemic accelerated these trends. Roblox estimates that in 2020, 75% of American 9- to 12-year-olds were using the platform. In a significant indicator of where it believed its future growth would come from, Fortnite announced at GDC last year that it would send 40% of the creator-generated revenues back to the creators themselves.

And all of this is for in-game modifications. There is also the greater world of streamers and other content creators who post content on YouTube, Twitch and TikTok about games. On YouTube alone, creators currently earn about $600m each year. While not directly built for the games, this content can often yield storylines and ideas that eventually make it back into the canon.

If all this was possible through the efforts of gamers and creators willing to take on content creation as a profession, imagine the exponential growth when they can use AI to significantly decrease the cost of production, and amateur users can much more easily create content. “This younger generation doesn’t just want content thrown at them,” said Craig Donato, Roblox’s chief business officer, in an interview with The Economist. “They want to feel a sense of agency or co-ownership of the medium…[and] that they are not just consumers of content, but that they can also be participants in the creation of the content.”

Hollywood is wise to the trend. It would seem an obvious answer to its widespread fear of AI replacing its creatives, but community-based control of content is not within its DNA to the extent that it is in gaming. It is still through entirely haphazard or highly slow and structured channels that UGC makes it into film or TV franchises. That doesn’t mean the fans don’t form communities around their work, just that the studios have yet to recognize any direct revenue from their efforts.

So Goes Gen Z, So Goes the Cultural Zeitgeist

The “younger generation” Donato is referring to could encompass Alpha, Gen Z and millennials, but there’s a reason why the middle child garners such attention from companies, political parties and arbiters of culture. Politically active, militantly individualistic and accepting, nostalgic for a simpler pre-digital time they’ve never known, Gen Z sets our trends.

Their preeminent digital medium is TikTok. Unlike its social media predecessors, TikTok encourages raw, authentic content geared towards niche communities. Its influencers are especially trusted by their followers. The videos are largely geared towards entertaining challenges and user participation, further enhancing the feel of community. Its remarkably global base makes it forward-thinking and inclusive. That culture has propelled TikTok to a cool 1.1 billion MAU and one of the most profitable platforms a brand can be on.

Gen Z’s pride of place on TikTok is not because it makes up the lion share of users, but because it is the content engine. Millennials and older generations on TikTok tend to scroll — to be consumers of content. It is a form of media and entertainment for them. For Gen Z, the platform is a vital means of communication and self-expression. It is through TikTok that they find like-minded communities where they feel safe to be their authentic selves. Marketers everywhere have understood that, “It is almost like an extension of themselves, a virtual personality, a platform for self-expression, and a creative outlet. It checks all the boxes as a multi-utility platform for Gen Z.”

And production quality of the content is nowhere near as important as preserving the illusion that it is raw and unfiltered. Gen Z connects with other people: they have no allegiance to brands and are deeply suspicious of Big Tech (ironic given what has been well-documented about TikTok’s data practices).

Despite backlash against Gen Z’s very different values, the wider cultural moment cannot deny that technology is bringing it closer to their reality and the all-important endeavor of self-expression.

We Become Our Real Selves through AI

It is no surprise that one of the first people to be truly enthusiastic about the implications of AI for creativity was a legendary comic book writer. AI allows you to become your true self, Gareb Shamus said to a completely shocked and disbelieving room at Comicon two years ago. I immediately took what he said to heart, and not just because I’m a fan of his work.

Two years on, a hugely popular musician relayed to me the time Reed Hastings gave him 30–3 wisdom. Whenever I talk to an artist about AI, they are terrified for about 30 seconds. After 3 minutes, they are thrilled. The democratization of creative production extends to those who are at the top of their professions, too.

At ChatGPT-3’s debut, a Reddit comment dared to say what many writers did not want to hear but could perhaps only begrudgingly accept as having the ring of truth: “GPT-3 may show how unconscious some human activity is, including writing. How much of what I write is essentially autocomplete?”

In a hugely thought-provoking essay exploring whether AI has an unconscious, all the more so because if its publication in n+1, a bastion of artistic purity, Meghan O’Gieblyn extended this assertion to make the point that:

Mechanical metaphors for the unconscious would evolve alongside modern technologies. Freud spoke of the drives as hydraulic; Lacan envisioned the deepest level of the psyche as algorithmic. The unconscious was blind, mechanical, and repetitive, but it was also a vault of hermetic knowledge, a reservoir that contained the entirety of the patient’s past and could reveal the true meaning of actions that appeared, on the surface, to be meaningless.

In other words, if the process of forming a human unconscious — where creativity, intuition and talent reside — is much the same for people as it is for a machine meant to mimic human cognition, then we ought to be quite comfortable with the conclusion that a machine can have an unconscious as well.

For a user base hungry for more tools to assert its authentic self in the public forum, there could be no better partner — indeed, no better agent, than AI.

Freedom First, Trust Second

Time and again, we have shown as users that security is nowhere near as important as other values when choosing to engage with a new technology. Freedom, convenience, vanity all receive more priority, just to name a few.

The reason why most of us choose web3 is not because we feel unsafe in the world of Big Tech, but because we feel walled in — unseen and unrealized.

AI gives you the means to become your real self. Crypto will have a place in this story because it gives you the freedom to become your real self.

There would come a time when self-determination became such an uncontainable urge that current tech no longer satisfies it. Appearing in TikTok as your normal, unfiltered self and reaching millions? Not good enough — because that is not really me. I am so much more, and the only thing that will make people believe me is AI. It is my extended self.

That is the ethos and urgency that will drive innovation at the intersection of AI and crypto. Without putting it in so many words, the user has demanded that it get there. It is already here.

The System

AI systems require scale. Blockchain systems, even the most scalable, are not built with scale as the primary objective. The key to designing an integration between the two is to minimize the number of touchpoints and dependencies without sacrificing the permissionlessness and security that we’re striving to achieve in such a system.

To limit the scope, there are a few parameters that we can probably safely take as givens. First, not all AI data needs to live on chain. AI data can be divided into three major groups: the inputs that train the machine, the prompts from the human principals and the logic that processes both in order to produce outputs. To verify the quality and correctness of the outputs, many have argued that the inputs and prompts, at a minimum, must live on chain. Many would argue even that is unnecessary, which we can explore in a bit, but there is significant consensus that on-chain logic is simply not feasible. The logic necessary to run an LLM or even create art through Generative AI is too expansive to live on chain, and there are many ways to represent them on chain without adding them to state.

Second, simply because an asset is generated by AI or is directed by semi-autonomous AI does not mean that asset needs to live on chain. Even now, for instance, NFTs and other large cryptoassets do not live on chain. Rather, their metadata contains a URL that points to the actual file stored in IPFS or simply a cloud server like AWS S3. There is no clear reason why AI assets should be treated differently.

Third, the function of the chain is to have a decentralized record of the actions of AI agents and to hold them accountable for adjustment should they no longer accurately reflect their principals. A short while ago, there was still the hope that if we stored as much AI data on chain as possible, then anyone would be able to reproduce and therefore verify AI outputs. AI models have evolved since then such that given the same inputs and prompts, they will probably produce different results given that the machine updates its logic with each iteration.

With these three parameters in place, the task is starting to seem much more manageable, even familiar. There isn’t much daylight between how we would ask a chain to interact with AI agents and current non-AI assets that live on chain. Minimizing data in the state of a blockchain is considered good practice because it optimizes the cost and performance of the chain. Assets that can be stored off chain, should be. Their hashes live on chain, and either a human or an AI agent can call the program or contracts on chain to direct that data. And as with non-AI assets, storage using a centralized cloud service or IPFS are just as valid as before, though those who are especially rigorous about decentralization will choose the latter.

None of this detracts from the fact, however, that AI agents are able to execute a far higher volume of transactions on chain, and that is where improvements to scalability do matter. As we move to parallelized computing in web3 with an integrated stack like Saga, the ability to process large volumes of transactions will vastly increase.

However, web3 implements scaling through the addition of dedicated blockspace, much of which is not interoperable. That brings us to the major task ahead of web3 in order to realize the full potential of AI + crypto.

The open secret of blockchain is that while each chain gives you permissionless access, interoperability is still largely a pipe dream. That gave you freedom of movement within a community, but that’s not quite the same as freedom, is it?

The interoperability solution of choice in web3 is currently bridging, which requires that an intermediary (i.e. the bridge) effect a transfer of an asset from one ecosystem to another. Assuming that the bridge is safe, which many aren’t, bridges add support to chains through manual onboarding processes. The bridge has to perform due diligence on the chain to have confidence that the transactions sent by the chain are indeed final and valid. For dedicated blockspace that is not a chain, such as a rollup, there is a challenge period of about a week before a sending transaction can be considered final and valid.

That sort of delay is no longer tolerable given the proliferation of dedicated blockspace for scaling purposes. At Saga, because every chain that is instantiated is a fully decentralized proof-of-stake chain, bridges are able to consider the sending transaction final and can therefore implement the transfer to the receiving chain in a permissionless manner. That allows assets originating on our chains to benefit from fast bridging to other ecosystems. (Within Saga, all chains communicate through a messaging protocol called Cosmos Inter Blockchain Communication, or IBC, not a bridge.)

The result is true freedom of movement of assets, and therefore true control of intellectual property on the part of the creators and developers. Without this freedom, blockchain essentially remains a highly inefficient landscape of walled gardens, negating many of the reasons why this technology is implemented to begin with.

There you have it: the basis for the Multiverse is a scalable and interoperable web3 protocol.

Conclusion

We were so busy searching for a revolution that we never realized it was already here. It shouldn’t be a surprise that a technology so rooted in individualism and personhood would find its intersection with AI in the realm of self-realization.

AI gave us the means to become our true selves. Crypto gave us the freedom to become our true selves.

We live in their creation. Come meet me in the Saga Multiverse.