How Maven’s AI-run ‘serendipity network’ can make social media interesting again

Everything in society can feel geared toward optimization – whether that’s standardized testing or artificial intelligence algorithms. We’re taught to know what outcome you want to achieve, and find the path towards getting there. 

Kenneth Stanley, a former OpenAI researcher and co-founder of a new social media platform called Maven, has been preaching for years that this method of thinking is counterproductive, if not outright harmful. Instead of prioritizing objectives, Stanley says we should be prioritizing serendipity. 

“Sometimes, in order to find those stepping stones that will lead to the things we care about, we have to get off the path of the objective and onto the path of the interesting,” Stanley told TechCrunch in a video interview. “Serendipity is the opposite of finding something through objectives.”

The idea of seeking novelty for its own sake started as an algorithmic concept that Stanley studies called open-endedness, a subfield of AI research about systems that “just keep producing interesting stuff forever.”

“Open-ended systems are like artificially creative systems,” said Stanley, noting that humans, evolution and civilization are all also open-ended systems that continue to build on themselves in unexpected ways. 

This algorithmic insight morphed into a life philosophy for Stanley. He even wrote a book about it in 2015 with his former PhD student Joel Lehman called Why Greatness Cannot Be Planned. The concept took off, making Stanley something of an international focal point for the brazen idea that, actually, you can just do things because they’re interesting, rather than because you need to complete some stated objective. 

But in 2022 while leading an open-endedness team at OpenAI, Stanley said he was “boiling over with discontent” and “had this epiphany” where he decided to stop talking about bringing open-endedness to wider audiences and instead start doing something about it. 

What if, he asked himself, he created a “serendipity network,” a system that’s set up to increase the probability of serendipity, for other people to enjoy?

So he quit his job and set about to create Maven, a social network built around an open-ended AI algorithm that evolves to seek novelty. When signing up, users select a series of topics to follow — from neuroscience to parenting — and the algorithm shows them posts that align with their interests. There are no likes, upvotes, retweets or follows, and there’s no way to amplify content to the masses. 

Instead, when a user posts something, the algorithm automatically reads the content and tags it with relevant interests so it shows up on those pages. Users can turn up the serendipity slider to branch out beyond their stated interests, and the algorithm running the platform connects users with related interests. So if, for example, you’re following conversations about urban planning, Maven might also suggest conversations about public transit. 

And while there’s no way to follow people on the platform, you can see and connect with other people who follow topics you’re interested in.

In a lot of ways, Maven feels like an antidote to today’s social media, where the “objective paradox is on full display” as people fall over themselves to create sensationalist content that will garner more attention and popularity. 

“The echo chambers and the toxicity, the narcissism amplification and personal branding has gone totally out of control so that people are losing their soul and turning into brands,” said Stanley.

The addictive qualities of social media, harm to mental health in adolescents and adults, and ability to polarize nations is well documented. These, Stanley says, are the unintended consequences of ambitious objectives, the outcome of making popularity a proxy for quality.

“And then you get all these other things because once you have popularity, you have perverse incentives,” he said.

Stanley noted that Maven users can flag inappropriate content or misinformation when it pops up, and its AI is actively monitoring for highly inflammatory, offensive “or worse” content. He said Maven can’t fix the nastiness in human nature, but by eliminating the incentives behind sharing such content, Stanley hopes it could change the “overall aggregate dynamic of how people are behaving.”

Some social media companies have attempted to combat such incentives in the past. The OG of pushing out serendipitous content was StumbleUpon, a browser extension and app created by entrepreneur Garrett Camp, years before he co-founded Uber. Instagram in 2019 then tested out hiding “likes” to curb comparisons and hurt feelings that come with attaching popularity to content. X, formerly Twitter, is preparing to make likes private, as well, but for less wholesome reasons. In a very Elon Musk-inspired line of thinking, X’s goal is to create more engagement by allowing people to privately like “edgy” content that they otherwise wouldn’t to protect their public image. 

Maven is less interested in connecting users with audiences, and more focused on connecting them with what’s interesting. 

The problem of monetization

Stanley and his co-founders – Blas Moros and Jimmy Secretan – soft-launched Maven in late January. The platform publicly debuted in May alongside a Wired feature that Stanley says gave Maven a top trending spot on Product Hunt and brought on thousands of sign ups.  

Those are still small numbers compared to other new entrants into the social media space. Bluesky, which launched in 2021, has had 5.6 million sign ups. As of January 2024, Mastodon had 1.8 million active users. Farcaster, a new crypto-based social protocol that just raised $150 million, has counted about 350,000 signups. All of these new networks will need to grow significantly if they’re to be considered successful.  

It’s still an open question over whether Maven will even be able to grow its user base without the very toxic qualities we love to hate, but which nonetheless drag us back to the cesspit that is social media.   

Maven raised $2 million in 2023 in a round led by Twitter co-founder Ev Williams, Stanley told TechCrunch. OpenAI CEO Sam Altman also participated in the round. Stanley said Williams and Altman invested because, like many of us who have become endeared by Maven’s almost too-sweet-for-this-world ethos, they think the world and the internet needs something like this. 

And indeed, Maven’s idealistic hope to connect people to interesting ideas is reminiscent of the early 2000s, when the internet was a place of connection and exploration. Sentiments from early users on the platform are mostly positive and optimistic, as many came to the platform for genuine and serendipitous interactions and the promised freedom from toxicity.

But will idealism be enough to bring on more institutional investors later when Maven wants to grow? 

“I think the challenge we face is that going forward, that becomes a harder and harder way to raise money,” said Stanley, noting that investors won’t be throwing down millions unless there’s a clear path to get a return on their investment.

“I just need to find the right investors going forward and quickly get to a sustainable business model,” he continued, musing over the idea of a subscription model that would allow Maven to keep its ideology intact.

There are, of course, other ways for Maven to bring in revenue. Advertising is one path, but one that appeals less to Stanley because of how tied up it is with virality and sensationalism. 

Down the line, Maven could also potentially sell its data to companies like OpenAI that are training their algorithms on reams of data. OpenAI earlier this month signed a deal with Reddit to train its AI on the social media company’s data. And Maven’s value proposition from an AI standpoint isn’t even just the content on the platform – it’s the open-ended algorithm running it. 

Stanley told TechCrunch he believes open-endedness is essential to artificial general intelligence (AGI), a type of AI that aims to match or surpass human capabilities across a range of cognitive tasks. Open-endedness is “such a salient aspect of being intelligent,” Stanley said. “It’s like this creative and also curiosity-driven aspect of being human.”

“The data is interesting from an AI perspective, because it’s data about what is interesting,” said Stanley, noting that current AI models are missing the intuitive understanding of what is interesting and what is not, and how that can change over time. However, even though the data has potential value to AI, Stanley said Maven has no deal with any company to grant access to that data. 

And while he said he hasn’t ruled that possibility out in the future, he would think very carefully about what the implications of sharing such data would be. 

“That’s not the point of this for me,” he said, noting that he’s not convinced that it would be a good thing for neural networks to be completely open-ended because that might make any creative endeavors by humans completely pointless. 

“I really wanted to create this worldwide serendipitous community,” he said. “It’s not like I have a side plan that we’re going to use Maven to create open-ended AI or something. I just wanted to create something for people because I started to feel like everybody’s gonna be talking to chatbots more and more and we’re gonna be less and less connected with other people. And I was contributing to that being an AI researcher.”

“Something about this idea of a serendipity network made me feel morally better, like I could actually contribute to people being more connected rather than less.”