Following our recent post on the dangers of overselling the capabilities of voice recognition software (see here), we commissioned a poll of over 2,000 Americans to see whether they had used Siri in iOS7, what they thought of Siri’s capabilities, and whether they had been “oversold” by Apple.
The results were a little shocking. Once we had pruned out those people who did not use Siri on iOS7 (being the most up-to-date version and now officially out of beta), we were able to ask two different groups of people to rate Siri, and to say whether they thought Siri had been oversold. Both groups came back with almost identical results.
The first group, asked if they felt that Apple had “oversold the voice recognition capabilities of Siri” came back with 46% agreeing, an amazingly high result when you consider this is a flagship application.
The second question looked more at how people rated accuracy. This question was asked of a different sample group, and they came back with a 44% “Hit or Miss” or worse score:
All told, not great news for Apple, but probably worse news for the voice recognition industry. We sell it like it’s magic, but as I’ve read elsewhere, it’s still an emerging technology that’s been emerging for 30 years. The sooner we are honest with people about what can really be achieved, the more chance we have of adoption. Back in the early days of OCR, we only got 50% accuracy, but that was 100% more than people had the day before. Perhaps we should reset people’s expectations as to what continuous speech applications can manage out of the box and in the wild.
We need to pick the applications carefully first where there can be real benefit (like audio data mining and indexing), and as the technology matures and we can get close to 100% accuracy in a variety of situation, allow the technology back into the mainstream. But that is going to require a quantum shift from how we do speech recognition now.
EDIT: Press Release posted here