Kevin Kelly's The Inevitable: five AI predictions, ten years on
A blog post on taylor.town revisits Kevin Kelly's book The Inevitable a decade after it appeared. The post's author calls the book an insightful read that articulates a handful of macro technological trends and follows them to their logical conclusions, then reposts five AI-related excerpts with a few short remarks of their own.
The lead excerpt is Kelly's dated prediction: “By 2026, Google’s main product will not be search but AI.” Kelly recounts a conversation around 2002 at a private party for Google, before its IPO, when it was a small company focused only on search. He asked cofounder Larry Page where free web search would get the company. He admits his “unimaginative blindness”, noting this was before Google had ramped up its ad auction scheme and long before YouTube or other major acquisitions. Page replied: “Oh, we’re really making an AI.”
Kelly's reasoning, written at the time of the book: Google had bought 13 other AI and robotics companies in addition to DeepMind, and search made up 80 percent of its revenue. You might think the AI is there to improve search, but he argues it is backward: Google is using search to make its AI better. Every query, click on a result and link created on the web trains the AI, and each of the 3 billion queries Google handles a day tutors its deep-learning system. With another 10 years of steady algorithm improvements, plus a thousandfold more data and a hundred times more computing resources, he expects Google to have an unrivaled AI. He also cites Sundar Pichai, who in a fall 2015 earnings call called AI “a core transformative way by which we are rethinking everything we are doing.” The post inserts one line of its own here: in 2017, Google researchers published Attention Is All You Need, which introduced the transformer architecture.
The second excerpt is about concentration. Cloud AI, Kelly writes, obeys the law of increasing returns, or network effect: the more people use an AI, the smarter it gets, and the smarter it gets, the more people use it. A company that enters this cycle tends to grow so fast that it overwhelms upstart competitors, so the AI future is likely to be ruled by an oligarchy of two or three large, general-purpose cloud-based commercial intelligences. The post's author comments that Kelly predicted AI heavyweights while advocating a future of open weights, and quotes his line that additional intelligence should ideally be not just cheap but free, like the free commons of the web, feeding commerce and science and paying for itself.
Third, Kelly says the AI on the horizon looks less like HAL 9000 or a Singularitan superintelligence and more like Amazon Web Services: cheap, reliable, industrial-grade digital smartness running behind everything and almost invisible except when it blinks off. It would serve as much IQ as you want but no more than you need, plugged in like electricity.
Fourth, most commercial AI work will be done by programs that are not human-like. The bulk of AI will be special-purpose software brains: one translates any language into any other but does little else, another drives a car but cannot converse, another recalls every pixel of every YouTube video but cannot anticipate your work routines. Kelly's number: in the next 10 years, 99 percent of the AI you interact with, directly or indirectly, will be “nerdly narrow, supersmart specialists.” The post's author remarks here that not even Kelly foresaw the bitter lesson.
Fifth, conversational answers. Kelly expects that we will soon be able to ask the cloud, in conversational tones, any question at all, and if the question has a known answer the machine will explain it. His examples: who won the Rookie of the Year Award in 1974, why the sky is blue, whether the universe will keep expanding forever. The AI will learn to articulate what is known and not known, and may at first need a dialog to clarify ambiguities. As more books, movies and connected things are “cognified”, answers become ubiquitous, and Kelly predicts people will ask several hundred questions a day, mostly about themselves and their friends, such as where Jenny is or when the next bus comes. The cost of each answer will be nanocents, and search will become an essential universal commodity rather than a first-world luxury.
Key facts
- The post reposts five AI excerpts from Kevin Kelly's The Inevitable, headed by his prediction that by 2026 Google's main product will not be search but AI.
- Kelly's reasoning rested on 2016-era figures: 13 AI and robotics companies bought besides DeepMind, search at 80 percent of revenue, and 3 billion queries a day.
- He predicted an oligarchy of two or three large cloud-based commercial AIs, delivered like a utility in the style of Amazon Web Services.
- He also predicted that 99 percent of the AI people meet in the next 10 years would be narrow specialists, and that people would ask the cloud several hundred questions a day.
- The post adds only brief remarks, such as that Kelly advocated open weights and that not even he foresaw the bitter lesson.
Why it matters
Most technology predictions are vague. These carry a date and numbers: a stated 2026 deadline for Google's main product, a 99 percent share for narrow AI, two or three dominant clouds. Because the target year is the one the post is written in, the claims can be held against today's AI landscape. The post also surfaces the reasoning behind the headline claim: search as a training loop for AI, with Page's 2002 answer, “Oh, we’re really making an AI”, as the anecdote that anchors it.
Who it affects
Anyone who follows AI strategy and wants a decade-old, on-the-record baseline: readers tracking Google's direction, the large cloud AI providers Kelly had in mind, and people interested in the tension the post's author points to between concentrated cloud AI and open weights. It is a reading exercise for general technology audiences rather than guidance for a specific profession.
How to use it
Treat it as a checklist. Take the five excerpts, each with its claim, and compare them with what you see in current products: what Google's main product is, whether a few clouds dominate, how much AI you use is narrow versus conversational, how many questions you ask a day. For the full argument behind each excerpt, the book itself is the place to go; the post recommends it.
How solid is it
The excerpts are Kelly's own words from his book, and the post's author's remarks are short. The figures (13 acquired companies, 80 percent of revenue, 3 billion daily queries) are Kelly's numbers from the time of writing, not current ones. The post gives no verdict on whether Kelly's predictions came true; it only reposts excerpts with brief remarks. No current (2026) figures for Google's revenue, query volume or AI products are given.
Risks and caveats
This is a personal blog retrospective, not an evaluation: whether the oligarchy has materialised, or which companies would form the “two or three”, is not addressed. The remark about the bitter lesson is not explained. Kelly's own wording is probabilistic in places (the future is “likely” to be ruled by two or three clouds), and five excerpts are a selection, not the book's full argument, so it would be unsafe to score him from this post alone.
“Oh, we’re really making an AI.”
— Larry Page, as recalled by Kevin Kelly in The Inevitable (conversation around 2002)