Somewhat accidentally, I've written a short book on the history of AI. This wasn't on my plan for 2021-5, which was supposed to be all about finishing up a project on digitality and finally publishing some of my dissertation work, but I accepted an invitation to write on the historiography of AI for a special issue of Social Studies of Science. The result was loved by the special issue editors but desk rejected by the real editor who seemed unfamiliar with the concept of a historiographic essay. I resolved to put the material to work in a short book, but as the readers of short books don't typically want historiography either it developed into a narrative history of AI infused with ideas from the fast emerging literature on AI history.
There's a real challenge in trying to tell the story of AI, given that there's very little technical continuity in the technologies, applications, funding sources, research questions etc. to which the brand has been applied since the 1950s. I dealt with this by foreground the brand-like qualities of AI, documenting throughlines in the rhetoric of imminent revolutionary change that has been applied to AI since the 1950s to set against abrupt shifts in the technology that was supposed to bring about that transformation. This framing is the focus of a talk I've given several times, sometimes under the title "Artificial Intelligence: The Brand That Wouldn't Die."
Yet as someone trained in old-school AI during the early 1990s, I was strongly aware that AI was institutionalized from the 1960s to the 1980s as a prominent subfield of computer science, with the usual apparatus of courses, journals, conferences, textbooks and lab assignments. The reality of AI as what Kuhn would have called "normal science" and the rhetoric of AI as a technological revolution that always lay a decade away somehow coexisted.
My solution in the book is to focus on the institutional coevolution of AI and computer science over the course of eight chapters. Each chapter is split between a narrative section and a case study of a famous system which captures the aspirations, achievements and limitations of AI researchers during an era. The book is thus, in part, a history of Ai in eight systems, from the General Problem Solver of the 1950s to (inevitably) ChatGPT.
I wrote this with the MIT Press series Essential Knowledge in mind, but I've been told that because of a shift in editorial policy the press doesn't want anything with a strong point of view in that series. On the positive side, trying to squeeze all this into 40,000 words or less was proving challenging. In a slightly longer book I'll have some space to explain things better, contextualize more and bring some human details to the main characters.
In parallel, I've been trying out some of the material for the book in a series of articles for Communications of the ACM. What I initially planned as two contributions eventually became a series of five. So far four have been published. These are even more concise than the book will be, focusing more on the history itself than the "AI as brand" framing. They don't include substantial case studies, and given the venue assume more of a knowledge of computer science than the book version. But they do give a general sense of how I see the arc of AI over the past 70 years.
The final part, drawing some contrasts between 20th century AI and the current enthusiasm for generative chat systems, should be appearing soon.
Theme by Danetsoft and Danang Probo Sayekti inspired by Maksimer