Annotation
Licklider's 1960 paper is one of the founding documents of interactive computing, and nearly everything in it proved visionary — except the timelines. His five-year estimate for practical speech recognition was off by roughly half a century: IBM's 1962 Shoebox recognized sixteen words, Dragon NaturallySpeaking arrived in 1997 requiring pauses between each one, and genuinely fluent recognition waited for deep learning in the 2010s. The pattern — correct on the what, wrong on the when — became the defining signature of AI prediction.
What Actually Happened
Practical speech recognition did not arrive within five years. IBM's 1962 Shoebox recognized sixteen words. Dragon NaturallySpeaking launched in 1997 but required pauses between words. Genuinely fluent, large-vocabulary speech recognition did not become reliable until the deep-learning breakthroughs of the 2010s — roughly fifty years after Licklider's prediction, not five.