All Your Friends are Fake: Synthetic Social Networks with Generative AI
Why the next consumer social networks will be generative
Friendships are the fundamental unit of human existence. Throughout the ages, our social relationships fed us information that kept us alive. Our ability to empathize and communicate with one another became the bedrock that built massive civilizations of the past and today.
It is little surprise that our friendships have evolved as we’ve evolved. We’ve seen dramatic shifts in just the past decade: Our friendship webs have become looser and wider as social media enabled us to overcome Dunbar’s number and maintain relationships even when we’re not physically present. Our core tribes have moved online and became more interest based rather than geographically distributed in our local neighborhoods. And our instinct for human connection is being answered by our instinct to go online — people are deepening friendships, meeting their partners, and interacting with virtual people in greater numbers than ever before.
Most recently, our identities have shifted away from our IRL selves. We’ve gained the ability to project multiple identities online and separate ourselves from our physical self with pseudonymous identities in gaming, crypto, and v-tubing. It’s become so second nature to represent ourselves as digital figures (Reddit alone has 200 million pseudonymous accounts) that we don’t blink when we switch between our real life Instagram accounts and start chatting as ‘electricwhale5’ on Discord or acting as a giant robot monkey on Overwatch.
These changes in consumer comfort, along with the emergence of artificial intelligence technology that makes it easier to generate online characters, will catalyze an exciting new category of gaming worlds and social networks. Artificially generated characters will emerge as more than background NPCs and feel authentic as personal companions. The next big shift in our digital relationships will be the shift to synthetic social networks, or networks with both real and generated characters.
Enter your new best friend
Digital friends are nothing new. The first chatbot, ELIZA, was created in 1966 by Joseph Weizenbaum. ELIZA was a simple program that could mimic a human conversation by responding to user input with predetermined responses. Originally designed as a psychotherapy tool, ELIZA was popularized by the media and became well-known for users becoming emotionally attached to the program, often forgetting that they were conversing with a computer.
Flash forward to 2004, and it suddenly became normal and cool to make friends and form entire social groups and hierarchies online. World of Warcraft and Club Penguin popularized a category of massively multiplayer online role-playing games (MMPORGs) that made us comfortable interacting with synthetic players, including computer-generated ones like NPCs and quest bosses. When everyone’s hidden behind an avatar, natural social hierarchies dissolve and it feels equally acceptable to battle a computer boss or a real player. The differences between humans and computers shrink, and there is an illusion of an equalized playing field.
Today, it’s often hard to tell the difference between what’s real and what’s computer-generated — there are call centers run by bots, Google’s AI voice reservations that sound convincingly (or comically) real, and virtual influencers like Brud’s Lil Miquela. People have even formed deep, multi-year friendships with AI companions like Replika.
For certain pseudonymous social networks, these AI friends could help seed relationships and subvert the cold start problem. Imagine a new MMPORG where the first 100 players you meet are an indistinguishable mix of human- and machine-driven avatars. Regardless of activity volume on the server, you’ll always be able to access a group of friends to complete quests and raids with you. Imagine a new-age Twitter where you always encounter a high volume of niche content that’s perfectly relevant to you. Content can be generated by both humans and machines, but ultimately curated by human taste — our likes and retweets help determine which content, regardless of creator type, makes it to the most feeds.
These synthetic network effects can set powerful norms and define user behavior early in a platform’s history. The next wave of social platforms will no longer require a critical mass of human users to function as AI users can generate enough activity to keep us engaged. This will make it possible for many first-time users to stick to new pseudonymous social platforms; the time it takes to realize value on a new social game or app will exponentially decrease. This loop becomes even more powerful as more people get used to trusting and making friends online.
The next generation of social media no longer acts as a marketplace matching people with people and interests, it becomes a creation machine that generates an experience that’s custom tailored to you. Sooner than we know it, most of our digital friends will be fake.
If Internet 2.0 opened the door for online companionship, generative AI will rip that door off its hinges. Our online consumption will no longer depend on someone to create it, it will be generated live.
As dystopian as these AI algorithms may sound, there are even greater positive elements. In the face of America’s worsening loneliness epidemic, specialized AIs can scale a sense of companionship for those who need it most. For those who spend a lot of time online and lack a strong external support network, highly trained AI companions can provide an outlet to chat through mental health issues and help flag those at risk to human providers. For those in under-resourced communities, synthetic AI tutors can be online 24/7 to help, complete with generating new problem sets and math problem explanations to help struggling students understand all angles. For busy professionals, intelligent AI personal assistants can learn and adapt to your habits as a work companion.
Of course, this newfound companionship and optimization doesn’t come without its risks — if anything, the social media of Internet 2.0 have proven that the highest engagement platforms often appeal to our worst instincts around the 7 deadly sins and the most viral content often correlates with the most polarizing. These inherent biases of human nature may be accentuated in the datasets that we use to train our AI, the people who fine tune the algorithms, and the consumer-curators of this new generative content. Conscientious developers could select training data to mitigate these biases, but ultimately any technology designed for humans will always come back to our human touch and our human flaws.
Internet 2.0 already normalized meeting friends online. If most of our interactions today are evolving towards a blur of chat messages and virtual characters, why should it matter if the person behind the screen is real flesh and blood? This future is coming faster than we think.
This piece was co-written with GPT-3. Building something or thinking about synthetic networks? Feel free to contact firstname.lastname@example.org and let’s chat!
We’re most excited about consumer networks where synthetic characters can help overcome the cold start problem and make it possible to kickstart content and relationships without other real users. This will help accelerate adoption of these networks as your worlds learn and grow from you. We’re also interested in any AI that can enhance natural human interactions, whether as a perpetual companion, intelligent sounding board, or personal assistant, among others.