At the end of last year, I attended an event hosted by Google to celebrate advancements in AI. The company’s domain in New York’s Chelsea neighborhood now stretches literally to the Hudson River, and about a hundred of us gathered in a pier-side exhibit space to watch scripted executive presentations and demos of the latest developments. Speaking from a distance from the West Coast, the company’s high priest, Jeff Dean, promised “a hopeful vision for the future.”
The theme of the day was ‘exploring the (im)possible’. We learned how Google’s AI was used to fight wildfires, predict floods and assess retinal disorders. But the stars of this show were what Google called “generative AI models.” These are the content machines, schooled on massive training sets of data, designed to create writings, images, and even computer code that once only humans could hope to produce.
Something strange is going on in the world of AI. In the early 2000s, the field erupted from a lethargy – known as an AI winter – with the innovation of deep learning led by three academics. This approach to AI has transformed the field and made many of our applications more useful, enabling language translation, search, Uber routing, and just about anything with “smart” as part of the name. We’ve spent a dozen years in this AI spring. But in the past year or so, there was a dramatic aftershock after that earthquake when a sudden plethora of mind-bending generative models appeared.
Most of the toys Google demonstrated on the pier in New York showed the fruits of generative models, such as the flagship large language model, called LaMDA. It can answer questions and work with creative writers to create stories. Other projects can produce 3D graphics from text prompts or even help produce videos by running storyboard-like suggestions on a scene-by-scene basis. But much of the program dealt with some of the ethical issues and potential dangers of unleashing robot content generators on the world. The company went to great lengths to emphasize how careful it was when using its powerful creations. The most telling statement came from Douglas Eck, a chief scientist at Google Research. “Generative AI models are powerful – there’s no question about that,” he said. “But we also need to recognize the real risks this technology can pose if we’re not careful, which is why we’ve been so slow to release it. And I am proud that we have been slow to release them.”
But Google’s competitors don’t seem to have “slow” in their vocabulary. While Google has given limited access to LaMDA in a protected Test Kitchen app, other companies are offering an all-you-can-eat smorgasbord with their own chatbots and image generators. Just a few weeks after the Google event came the most sweeping release yet: OpenAI’s latest iteration of its own powerful text generation technology, ChatGPT, a lightning-fast, logorrical gadfly that generates coherent essays, poems, plays, songs, and even spewing obituaries at the slightest hint of a prompt. Taking advantage of the chatbot’s wide availability, millions of people have tinkered with it and shared its amazing responses, to the point where it has become an international obsession, as well as a source of wonder and fear. Will ChatGPT kill the college essay? Destroy traditional web searches? Putting millions of copywriters, journalists, artists, songwriters and legal workers out of work?
Answers to those questions are not clear at this time. But one thing is. Granting open access to these models is the start of a wet hot AI summer that is fueling the tech sector even as today’s giants lay off parts of their workforce. Contrary to what Mark Zuckerberg thinks, the next big paradigm isn’t the metaverse – it’s this new wave of AI content engines, and it’s here now. The 1980s saw a gold rush of products that moved tasks from paper to PC applications. In the 1990s, you could make a quick fortune moving those desktop products to online. A decade later, the move was to mobile. In the 2020s, the big shift is towards building with generative AI. This year, thousands of startups will emerge with business plans based on tapping into those systems’ APIs. The cost of producing generic copies goes to zero. By the end of the decade, AI video generation systems could dominate TikTok and other apps. They may be nowhere near as good as the innovative creations of talented humans, but the robots will dominate quantitatively.
