I’ve been retelling an anecdote from IBM’s speech-to-text experiments recently, and couldn’t remember where I’d got it… and indeed, I couldn’t remember if it was even true, as happens when you retell teh same story again and again…
Searching for combinations of things like “speech-recognition”, “IBM”, “faked test” and so on wasn’t getting me anywhere. But I’ve finally found a source: Jeremy Clark’s Pretotyping@Work eBook.
I’m posting the main bit of the anecdote here for two reasons. Firstly, I think you might find it interesting, and perhaps useful. Secondly, now that I’ve put it on my own site, using the aforementioned search terms which are the ones I clearly use to look for it, I’ll find it easier to find in future, hopefully…
“In the 1980’s, IBM was in discussions with several important customers about a radical product idea: hardware and software that could turn spoken words into a text on a screen. The fundamentals of the technology were still years away, yet customers seemed very enthusiastic: many declared they would pay generously for such a solution.
Traditionally, IBM would have launched an R&D effort to develop the algorithms and electronics necessary to demonstrate a prototype. In the case of the Speech-To-Text idea, however, a team member had an intriguing alternative suggestion: they should pretend to have the solution, to see how customers actually reacted to the capability.
What the team did was to create a movie-set like testing lab, in the form of a typical office space of the day. Customer subjects would be briefed on the Speech-to-Text solution, then seated in the space. The subject would speak into a microphone, dictating a variety of office correspondence, and would almost immediately see their words appear on the screen on the desk in front of them. What the subjects didn’t know was that the electronic output was being produced by a typist in a nearby room, listening to the dictation through headphones.
What the IBM team learned was that, in practice, customers didn’t like the solution, not because of flaws in the product (the transcribed text) but because of a host of hitherto-unseen environmental challenges: speaking taxed the subject’s throat, there was concern for privacy surrounding confidential material that the speaker would not
wish to be overhead, and so on.
Actual exposure to the essence of the proposed solution completely reversed the earlier customer enthusiasm.”