Here are some things I found out last night from John McCarthy:
When he famously created Lisp (which became the standard programming language in AI), he had gotten the idea of list processing from Allen Newell and Herbert Simon at the Darmouth conference on AI in 1956.
When he “stole” the lambda notation from Alonzo Church and used it in creating Lisp (giving credit to Church, of course), he didn’t know that the lambda calculus already was an universal formalism for computation, because he had bought Church’s book but didn’t read it all the way through. Had he known that the lambda calculus is a universal computing formalism, he might have tried to create a language based entirely on the lambda calculus, as people did 20 years later. But that, he says, would not have been as good a language as Lisp.
When he and Marvin Minsky decided to start a research project on AI at MIT, he intentionally avoided the opinion of (then influential MIT professors) Warren McCulloch and Norbert Wiener, because he thought they would have strong views on how he and Minsky should proceed and would try to influence their research.