I was fortunate enough to be able to write software for the first version of the CM (the SIMD one). Although I was keenly interested in AI even back then, I viewed the CM as a great system for parallel processing, with good I/O throughput, and not for AI research.