AGI by 2029, the Singularity by 2045
Kurzweil draws a hard line between two milestones most people conflate. AGI — a system that matches human cognitive performance across the board — is his 2029 target. The Singularity, which he defines as a million-fold expansion of combined human-machine intelligence, is 2045. The gap between them is where the compounding happens. After Stanford convened hundreds of researchers in the early 2000s to audit his timeline and concluded the real answer was closer to a century away, the intervening decades have moved quietly in his direction.
Why the Exponential Curve Doesn't Show Up Until It Does
Kurzweil's explanation for why so many smart people miss exponential trajectories comes down to how humans read early data. When the Human Genome Project had mapped just one percent of the genome after seven years, critics demanded it be defunded — at that pace, completion was centuries away. Kurzweil saw it differently: one percent completion on a doubling curve means you're seven doublings from the finish line. The project completed on schedule. He applies the same logic to AI capability curves today, and argues the people calling the current moment a plateau are making the same mistake the genome sceptics made in 1998.
The Two Problems Still Standing Between Here and AGI
Language models can discuss physics fluently but don't yet have a functional model of how physical objects interact in space. And robotics remains orders of magnitude behind linguistic AI — the hardware to embody intelligence in the world still costs six figures per unit, which needs to collapse dramatically before 2029. Kurzweil is specific: Google has an active internal project targeting physical reasoning, and cost curves on robotics hardware are following the same trajectory that computing hardware always has. Neither bottleneck, he argues, requires a conceptual breakthrough — just time on the curve.
Building a Chatbot from His Dead Father's Letters
One of the quieter moments in the conversation: Kurzweil describes training a conversational system on his late father's complete archive of letters, musical compositions, and journals. Built on top of Google's Talk to Books — a semantic retrieval system he developed that predated ChatGPT by four years — it was an early experiment in digital continuity. He's now building a version of himself, timed to launch alongside his autobiography. He acknowledges the obvious: it will recall every paper he's ever written with perfect fidelity, and will almost certainly outperform his biological memory in a live conversation.
The Achievement He's Actually Proudest Of
Asked to name the milestone that matters most from a career that spans the invention of the flatbed scanner, the first text-to-speech synthesiser, and a decade at Google shaping modern search, Kurzweil bypasses all of it. He points to his reading machine for the blind — a device that cost $20,000 when first built and now exists as a free app on any smartphone. The entire arc of his philosophy is in that answer: the purpose of exponential technology is not the technology itself, but the progressive democratisation of capabilities that once existed only for the privileged few.