Commemorating 70 Years of Artificial Intelligence
IEEE Spectrum San Murugesan
AI just turned 70—Dartmouth 1956 to ChatGPT and beyond. Same field, wildly different beast now, and IEEE's been in the room the whole time.
Based on reporting by IEEE Spectrum, San Murugesan — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Seventy years ago this summer, four researchers wrote a proposal for a workshop at Dartmouth College and, almost as an afterthought, coined a term that would outlive all of them: artificial intelligence. John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon weren't trying to build ChatGPT. They wanted to know if machines could simulate the kind of thinking that, as Minsky later put it, would require intelligence if a person did it. That framing has aged remarkably well, even as everything else about the field has been rebuilt from scratch multiple times over.
The path from Dartmouth to Google Brain's 2017 transformer paper was not a straight line. Warren McCulloch and Walter Pitts sketched artificial neurons back in 1943, before anyone had a working computer to run them on. Alan Turing asked whether machines could think in 1950, a full six years before the field had a name. Then came Lisp, then machine learning as a phrase coined by Arthur Samuel in 1959, then the excitement and eventual collapse of rule-based expert systems like MYCIN in the 1980s. AI winters followed AI springs followed more winters, funding rising and falling like a fever chart, until deep learning and transformer architecture in the 2010s finally gave the field something that didn't just work in a lab demo but scaled.
What's striking, looking back, is how much of the current moment was already implied in that original 1955 proposal — and how little of the current risk profile was. Nobody at Dartmouth was worrying about hallucinations, deepfakes, or sycophantic chatbots flattering users into bad decisions. Those are 2020s problems, born from systems powerful enough to be genuinely useful and therefore genuinely dangerous when they're wrong. Agentic AI, now creeping toward autonomous operation with minimal oversight, raises the stakes further: a chatbot that makes up a fact is annoying, an autonomous system that makes a bad call in healthcare or transportation is something else entirely.
IEEE's role in this anniversary isn't just ceremonial nostalgia. The organization runs 11 AI journals, more than 100 AI-related standards, a certification program for ethical autonomous systems, and over 100 conferences a year — infrastructure that turned a Dartmouth thought experiment into an actual engineering discipline with guardrails, however imperfect. The Stanford AI Index's claim that adoption has outpaced the telephone, television, and the internet isn't hype for hype's sake; it's a genuine reordering of how fast a technology can embed itself into daily life, and it means the standard-setting and governance work matters more now than at any point in the previous 69 years.
Turing wrote in 1950 that we can only see a short distance ahead but there's plenty that needs doing in that distance. Seventy years on, that's still the honest summary. The field has finally built something that works at scale; whether it works well, and for whom, is the argument that's just getting started.
My take — AI-written commentary, not fact-checked reporting
I've watched enough of these AI anniversary retrospectives to notice they always flatter the field's continuity — as if 1956 Minsky and 2025 agentic AI are one smooth story. They're not; deep learning was a discontinuity, and pretending otherwise undersells how much of today's risk (sycophancy, hallucination, autonomous errors at scale) is genuinely new and unprecedented, not just an old problem in a shinier wrapper. My honest take: the standards bodies and certification programs are playing catch-up to systems that are already deployed, and history says that gap rarely closes in favor of caution.
Read more about this at: IEEE Spectrum