What The Hell America!

There are really two OpenAIs. One is the creator of world-bending machines—the start-up that unleashed ChatGPT and in turn the generative-AI boom, surging toward an unrecognizable future with the rest of the tech industry in tow. This is the OpenAI that promises to eventually bring about “superintelligent” programs that exceed humanity’s capabilities.

The other OpenAI is simply a business. This is the company that is reportedly working on a social network and considering an expansion into hardware; it is the company that offers user-experience updates to ChatGPT, such as an “image library” feature announced last week and the new ability to “reference” past chats to provide personalized responses. You could think of this OpenAI as yet another tech company following in the footsteps of Meta, Apple, and Google—eager not just to inspire users with new discoveries, but to keep them locked into a lineup of endlessly iterating products.

Read: The curse of ChatGPT

The most powerful tech companies succeed not simply by the virtues of their individual software and gadgets, but by building ecosystems of connected services. Having an iPhone and a MacBook makes it very convenient to use iCloud storage and iMessage and Apple Pay, and very annoying if a family member has a Samsung smartphone or if you ever decide to switch to a Windows PC. Google Search, Drive, Chrome, and Android devices form a similar walled garden, so much so that federal attorneys have asked a court to force the company to sell Chrome as a remedy to an antitrust violation. But compared with computers or even web browsers, chatbots are very easy to switch among—just open a new tab and type in a different URL. That makes the challenge somewhat greater for AI start-ups. Google and Apple already have product ecosystems to slide AI into; OpenAI does not.

OpenAI CEO Sam Altman recently claimed that his company’s products have some 800 million weekly users—approximately a tenth of the world’s population. But even if OpenAI had only half that number of users, that would be a lot of people to risk losing to Anthropic, Google, and the unending torrent of new AI start-ups. As other tech companies have demonstrated, collecting data from users—images, conversations, purchases, friendships—and building products around that information is a good way to keep them locked in. Even if a competing chatbot is “smarter,” the ability to draw on previous conversations could make parting ways with ChatGPT much harder. This also helps explain why OpenAI is giving college students two months of free access to a premium tier of ChatGPT, seeding the ground for long-term loyalty. (This follows a familiar pattern for tech companies: Hulu used to be free, Gmail used to regularly increase its free storage, and eons ago, YouTube didn’t serve ads.) Notably, OpenAI has recently hired executives from Meta, Twitter, Uber, and NextDoor to advance its commercial operations.

OpenAI’s two identities—groundbreaking AI lab and archetypal tech firm—do not necessarily conflict. The company has said that commercialization benefits AI development, and that offering AI models as consumer products is an important way to get people accustomed to the technology, test its limitations in the real world, and encourage deliberation over how it should and shouldn’t be used. Presenting AI in an intuitive, conversational form, rather than promoting a major leap in an algorithm’s “intelligence” or capabilities, is precisely what made ChatGPT a hit. If the idea is to make AI that “benefits all of humanity,” as OpenAI professes in its charter, then sharing these purported benefits now both makes sense and creates an economic incentive to train better and more reliable AI models. Increased revenue, in turn, can sustain the development of those future, improved models.

Then again, OpenAI has gradually transitioned from a nonprofit to a more and more profit-oriented corporate structure: Using generative-AI technology to magically discover new drugs is a nice idea, but eventually the company will need to start making money from everyday users to keep the lights on. (OpenAI lost well over $1 billion last year.) A spokesperson for OpenAI, which has a corporate partnership with The Atlantic, wrote over email that “competition is good for users and US innovation. Anyone can use ChatGPT from any browser,” and that “developers remain free to switch to competing models whenever they choose.”

Read: The Gen Z lifestyle subsidy

Anthropic and Meta have both taken alternative approaches to bringing their models to internet users. The former recently offered the ability to integrate its chatbot Claude into Gmail, Google Docs, and Google Calendar—gaining a foothold in an existing tech ecosystem rather than building anew. (OpenAI seemed to be testing this strategy last year by partnering with Apple to incorporate ChatGPT directly into Apple Intelligence, but this requires a bit of setup on the user’s part—and Apple’s AI efforts have been broadly perceived as disappointing.) Meta, meanwhile, has made its Llama AI models free to download and modify—angling to make Llama a standard for software engineers. Altman has said OpenAI will release a similarly open model later this year; apparently the start-up wants to both wall off its garden and make its AI models the foundation for everyone else, too.

From this vantage, generative AI appears less revolutionary and more like all the previous websites, platforms, and gadgets fighting to grab your attention and never let it go. The mountains of data collected through chatbot interactions may fuel more personalized and precisely targeted services and advertisements. Dependence on smartphones and smartwatches could breed dependence on AI, and vice versa. And there is other shared DNA. Social-media platforms relied on poorly compensated content-moderation work to screen out harmful and abusive posts, exposing workers to horrendous media in order for the products to be palatable to the widest audience possible. OpenAI and other AI companies have relied on the same type of labor to develop their training data sets. Should OpenAI really launch a social-media website or hardware device, this lineage will become explicit. That there are two OpenAIs is now clear. But it remains uncertain which is the alter ego.