What is AI?
I’ve been struggling with this rather basic question for several years now, so today I figured I’d write out loud about it, and see if anything coherent surfaces.
First, let’s define what I mean by “AI.” For the most part, I’m referring to the at-scale generative AI services offered by Google, OpenAI, and Anthropic. But while these three companies dominate the market at present, generative AI is filtering into just about every digital surface where there’s money to be made, and as it does, new leaders will emerge, just as they have in every major phase-change in technology history.
Tech tends to move in a consistent pattern of punctuated equilibrium. A burst of innovation drives new consumer adoption, which builds into a period of relative stasis as the new products and technologies are integrated into society and the economy. Then a new burst happens, and the pattern repeats.
Sometimes we understand the potential of these waves early, but it takes years for their impact to mature. I’d argue that was the case with the World Wide Web in the early 1990s and mobile in the mid 2000s. Other waves arrive relatively unheralded and take years to develop. The personal computer was first introduced in the 1970s, but it took a decade for it to land on the cover of Time as the “person of the year.” Still other innovations spread quickly and are immediately understood as breakthroughs – it took Microsoft just 18 months to pivot from DOS to Windows once the Macintosh demonstrated the power of the graphical user interface, or GUI.
So where does AI sit in this framework? Clearly its impact was immediately heralded, but we’re still trying to understand what exactly that impact will be. The technology evades easy classification. Is AI a software product or service? A platform, or possibly even an operating system? A utility, like compute or storage? An interface? A new medium?
The easy answer is that AI is arguably all of these things and more. Its ecdysial nature sheds definition like a snake shedding skin. Three years ago we all thought AI was a replacement for search – a product delivered as a service over the Internet. But as the technology evolved and innovators began to build on top of AI’s capabilities, it became clear that AI was more than one product – it was becoming both a utility and a platform. OpenAI announced an app store, Anthropic integrated with critical Internet infrastructure like cloud and hosting providers. Entrepreneurs and VCs pitched AI as a utility that would be built into…everything.
But as billions of us began what is likely a lifelong relationship with AI services, we began to understand that AI represents more than just a product, utility, or a platform – it’s also an entirely new model for how we interact with digital technology, an interface upgrade akin to the shift from DOS to the graphical user interface.
So AI is all these things – a product, a service, a utility, a platform, and an interface. And if AI is all those things, it’s arguably also a new operating system – which the Oxford dictionary defines as “the software that supports a computer’s basic functions, such as scheduling tasks, executing applications, and controlling peripherals.” But with AI, the “computer” is not limited to the machine on our desk or the phone in our hand. My son recently leveraged Claude’s “operating system” to create a novel CRM system for his company (it took him less than a day, sorry-not-sorry Hubspot). It runs in the cloud, is expressed as a web page, and is integrated with at least half a dozen distinct products, including Google Suite, Slack, and SMS. That certainly sounds like what an operating system is supposed to enable – coordinate the underlying capabilities of a computer and providing a platform for its expression. In the case of AI, it’s becoming an operating system for … pretty much everything that might be possible on the Internet.
So is this the metaphor that might stick to AI’s slippery skin? Is AI the new operating system of the Internet? As a hypothesis it feels accurate, but incomplete.
Perhaps AI is also a medium?
The famously inscrutable media theorist Marshall McLuhan defined media as “any extension of ourselves” and “any new technology” that extends our physical or nervous system into the world. That certainly sounds like AI, doesn’t it?
Thanks in large part to McLuhan, digital technologies were collectively dubbed “new media” when they first broke into the popular consciousness in the late 1980s. It was a catch-all phrase ripe with implicit bias: the “old media,” with its one-to-many broadcast model, was giving way to an interactive, electronically distributed model of “new media*.” But the term aged poorly as the web broke out. The web was more than just media, it was a novel platform in its own right. By the mid 2000s, Tim O’Reilly and I, among many others, coined the term “Web as Platform,” which we imagined as open, distributed, and generative in nature. The word “media” simply wasn’t broad enough to encompass the web’s potential.
While much of the web ended up dominated by closed business models of extractive capitalism, I still believe in both the idea of “web as platform” as well as a liberal interpretation of the phrase “new media.” I can imagine how AI might play a role in delivering on the excitement and potential evoked by those now historic phrases. But they are now relics, neither of them ring true to me as potential descriptors of what AI might become in our culture.
So, nearly 1,000 words into this rumination, I’m no closer to answering the question I posed at the top, but there is one word I’ve not yet surfaced – and it builds on the core definition of media: Language.
“Language” is defined as “the principal method of human communication, consisting of words used in a structured and conventional way and conveyed by speech, writing, or gesture.” Language is considered a unique product of human intelligence, and it is our language that provides the foundational corpus for generative AI as we know it today. McLuhan considered language a medium, but then again, he believed the same about the light bulb. Regardless, as I struggle to find a handle by which to pick up the concept of AI, there’s something about “language” that feels appropriate. Maybe AI is, in fact, equivalent to a new language, one we’re just learning to speak. That certainly rings true to me.
—
* This was the entire focus of my masters thesis at Berkeley in 1992. Sigh.