We've only just begun to figure out how we interact with AI

Michael Sayman created SocialAI, a 'social' network populated entirely by generative AI followers – it might be one key way we use AI in the future.

Today, the experience of ‘using’ AI for most of us involves going to a website or opening an app and entering text into an input field that resembles a chat messaging interface – one human, messaging one AI chatbot. Other paradigms are emerging, but the chatbot has a persistence and prominence of place that’s hard to deny, which isn’t too surprising because the ‘conversation’ is a human technology as old as language, one we’re extremely comfortable with, and one that has almost no barrier to entry.

Lifelong tinkerer and boundary bender Michael Sayman doesn’t necessarily think a 1:1 conversational model is the be-all and end-all for our interactions with AI, however. To prove that other approaches might have equal – if not more – merit and applicability, he built a ‘social’ network where the ‘social’ refers to the UX, but not any actual interconnectedness between human beings.

“I believe what I'm trying to do is introduce a new interface for interacting with language models, and the application for that new interface as there are dozens of applications,” Sayman told me in an interview – prior to his recent announcement that he’s going to join Meta (where he originally began his tech career). “I'd argue it has as many as ChatGPT, if not more.”

“If anything, I'd say that this interface is better for some things that the chat interface is not good for, and vice versa,” he said. “And so that's really what I'm focused on – introducing a new interface for interacting with language models.”

Sayman’s theory that a one-to-many interface model might be beneficial for certain use cases of generative AI isn’t based solely on a hunch – it’s rooted in the recognition that we’ve actually developed social networks and social media in general over the years to be extremely good at driving engagement and activity.

“There is opportunity in an interface that has been designed over the past decade to maximize human interaction with other humans,” he explained. “Because these language models are very similar in the approach that humans take, which is fuzzy by nature – not certain”

The non-deterministic nature of generative AI is a key ingredient of the technology, and one that has resulted in countless businesses looking to mitigate the associated risks. A fundamentally non-deterministic tech is especially hard to deploy at scale in highly regulated environments, and when predictable results are a key business requirement. But leaning more into non-deterministic use cases might be the real path towards unlocking more value from generative AI, and Sayman thinks that having the interface reflect the multi-path outcomes of gen AI responses might be key to more effective usage in specific cases.

“There is no certain answer from the language model, but it has a bunch of ideas and it's probabilistic,” he said. “So a chat interface where it gives you one answer, just to me seems like it's the wrong interface. Or rather, not the wrong interface, but just one of many interfaces.”

Sayman believes, in fact, that the “broadcast layer” represented by SocialAI will in fact “be one of the most popular ways to interact with language models in the future.”

Despite the relative recency of generative AI’s popularity and ascendance, the original idea for SocialAI came to Sayman quite a while ago. He says he first attempted to build an app like this seven years ago, but quickly realized that the available tech at the time just wasn’t up to the task of delivering a product with a reasonable level of quality.

This points to what he says is his “only real skill,” – the ability to look ahead and anticipate what’s coming next, even if the fundamental technology to make it happen hasn’t quite evolved there yet. That’s been the basis of his entire career, which began when he was a teenager hacking around Club Penguin to such impressive effect that it caught the attention of (his once-again employer) Mark Zuckerberg. After being called in for a private meeting with the Facebook founder, he was hired at the company out of high school – a stint he followed up with gigs at YouTube and Roblox.

Sayman’s impetus for creating SocialAI wasn’t originally about conceptualizing new interfaces for generative AI – the idea predates ChatGPT by many years, after all. Instead, it had more to do with his experience of social media, especially as a younger person growing up with social media as already a dominant and omnipresent surface for interacting with others.

“The biggest thing for me with it many years ago was that I just felt like a lot of these social apps had like fun dynamics, and the parts of it that kind of made it not so fun were the emotional baggage that came with it,” he said.

“My earliest thoughts weren't very deep about how to build this product, and now obviously it’s evolved,” Sayman continued. “But in the beginning, I was just thinking ‘These apps, they're fun, but, there are all these people interacting – sometimes I get a little stressed. And I thought to myself, it would be a lot of fun if we could have a social game that wasn't really a social network, but was instead emulating one.”

Emulating a social environment online seemed to him like a way to short-circuit some of the more toxic and problematic elements baked into real social networks, but also had the potential to provide a set of ‘social training wheels’ to people who might aspire to more healthy online interactions with real people down the road, too.

“I always thought it was fun because Club Penguin to me was like a simulation of the world, but it was for kids and it was safer,” Sayman said. “It allowed me to learn about how to interact with other people, and how to do the adult life of pretending to have my house and my clothes and my things – so it's almost like a play, pretend kind of thing for kids of like the adult world.”

Contemporary life doesn’t so much resemble the traditional kids’ habit of playing house, thanks to the degree to which we rely on internet connectivity and online communication. SocialAI was influenced heavily by the realization that much of our adult lives today – for better or for worse – revolves around using social media, responsible but productive use of which something we’re still not great at preparing kids for ahead of time.

During our chat, we even discussed the possibility that we realize in time that the internet actually isn’t all that good for healthy social interaction, and it becomes a place optimized for information retrieval and co-creation with virtual entities rather than a space for people to connect with one another. That might instead go back to being a primarily IRL construct – leaving generative AI to dominate the once ‘social’ spaces our online lives became largely fixated upon.

Sayman announced last month that he’d be joining Meta once again, a homecoming for an incredibly talented builder who started his career there at just 17. The move came after he gained a lot of attention for SocialAI, and per Sayman himself his future work will focus on pushing forward the state of the art of gen AI products at the company, as well as shaping the future of our social interaction.