70th anniversary of Ashby's presentation at Dartmouth's summer AI workshop

Today marks the 70th anniversary of a presentation by the Cybernetician Ross Ashby at the Dartmouth summer workshop, regarded by many as the founding of Artificial Intelligence. Funnily, one of the organisers, John McCarthy, later admitted that part of his reason for coining the term 'Artificial Intelligence' was to keep Cybernetics out, Norbert Wiener in particular, who had given that field its name.

Despite studying control theory as an undergrad, I only discovered Cybernetics via Dan Davies' fantastic 'The Unaccountability Machine' two years ago. And what a fascinating subject it has proven to be. My work with large language models led me to a 1956 Ashby paper in a journal edition edited, funnily enough, by McCarthy (and Claude Shannon). When McCarthy proposed sending it out for peer review, Ashby replied: "I know of no referee to whose opinion I would defer" — no doubt cementing McCarthy's view of Cyberneticians.

I mark the anniversary for shameless self-promotion: I've just had a paper published on Ashby's work developing what he called an "intelligence-amplifier". My first peer-reviewed publication (and unlike Ashby, I'm thankful for the referees feedback), it tentatively proposes that Ashby's intelligence-amplifier sounds a lot like the LLMs we've built and call AI.

Further study since submission has convinced me it's exactly what we've built, and I hope to convince others (Dan Davies is giving the idea serious thought) with future publications, several already drafted and available on request.

So what? I hear you ask...

Primarily: a category error going back to McCarthy's ambition. The 1955 funding proposal for Dartmouth (the Rockefeller Foundation pushed back and told them to invite cyberneticians) famously claimed that any aspect of intelligence can be simulated by a machine.

Ashby was withering about "intelligence": "The only people who talk today of 'real' intelligence are those who hope to find a meaning for the adjective later." There's still no consensus definition today, which makes the report of Ashby's Dartmouth attendance all the more interesting: it records Ashby and McCarthy agreeing that 'selection' is the primary challenge in intelligence.

Ashby defined intelligence purely in terms of appropriate selection. And his ambition was not to simulate intelligence but to amplify it, like our muscles:

"Is it impossible that he should develop machines with 'synthetic' intellectual powers that will equally surpass those of his own brain?"

Seventy years after he set out his design, I remain convinced that were Ashby alive today, he'd be claiming we've finally built his intelligence-amplifier.
Sadly, depressingly, the use of AI in the genocide in Gaza and the illegal attack on Iran offers insight into just how far so-called Homo sapiens has evolved.