Predicting the future is not impossible — it's how we survive. For four thousand years, humanity has used arithmetic to do it better: to forecast the stars, navigate oceans, design bombs, and break codes. Arithmetic Is All You Need: The Human Story of Computer Intelligence uncovers the line from Babylonian astronomy to modern AI, to the moment when arithmetic became fast enough to predict language, and what emerged was something we recognized as an intelligence. It explains how computer intelligence works, where it came from, and what its continued development implies.
The book is told through the ideas and people that made each predictive leap possible: Tycho Brahe building the world's first data center, Napier's logarithms transforming science, codebreakers spurring the development of electronic computers. In its final chapters, the book looks to the future where the real danger of AI is not extinction but erosion. When the same optimization techniques that create computer intelligence are deployed to predict us, they narrow our choices, fragment our shared reality, and reduce our ability to reason. The book closes with an afterword on how AI agents may offer a path away from the attention-industrial complex — but the window is narrowing.
I am a Principal Research Scientist at Databricks, a former DeepMind researcher where I led the training of some of the largest language models ever built, and co-founder of Adept, an AI unicorn. My PhD is from Stanford, where my work was among the earliest on GPU computing. I am part of the tradition this book describes, and the book's central insight, that prediction quality improves predictably as computation increases, is a relationship I helped quantify.
The manuscript will be 60,000–75,000 words. Three and a half chapters and the front matter are complete. For readers of The Information, Chip War, and The Coming Wave.