Voice input: people don't talk to their desktop or laptop computers, because they are sat in front of them with more detailed, immediate and nuanced input devices available; they talk to their phones and other small appliances that are "ambient".
> "Fundamentally, the concept that your computer can actually look at your screen and is context aware is going to become an important modality for us going forward."
I hate to break it to them but they are the people who make the OS that renders all the content on the screen. They could make it more efficiently context-aware without massive privacy risks, simply by using the information they already have at that moment, and introducing APIs that applications could use to communicate context, under their users' control.
They don't want to do this because they don't want applications or users to have that kind of fine-grained control.
Do they even talk to their phones? The most prolific use of speech I can see in users around me are voice memos being sent as messages, which I can kinda get. It's like a better, asynchronous phone call. I prefer text to speech, but for people who prefer speech over text, this seems very useful. But that's not really talking to your phone in any meaningful sense.
The people when have Alexa and the like in their house tell me they only use it to set timers and play music (and even then, it's 'play my playlist' or something similar. They know it's going to trip on 'play Völlig losgelöst', or even just 'play Peter Schilling' is not unliktly to fail).
Computers have more potential, but if they can't reliably recognise what you say, we've failed before we've even started. And I have little faith that they can reliably act on what you said, never mind what one might have meant.
> "Fundamentally, the concept that your computer can actually look at your screen and is context aware is going to become an important modality for us going forward."
I hate to break it to them but they are the people who make the OS that renders all the content on the screen. They could make it more efficiently context-aware without massive privacy risks, simply by using the information they already have at that moment, and introducing APIs that applications could use to communicate context, under their users' control.
They don't want to do this because they don't want applications or users to have that kind of fine-grained control.