The big trick will be whether or not it ends up being surmounted fast enough to ever amount to much. It’s clearly possible to build an expert system that lives locally and isn’t effectively an emissary of a dubiously friendly power(yes, Amazon and Google and friends have much bigger computers; but they still couldn’t afford to make it ‘free’ if serving an individual’s requests were too computationally expensive, so while a personal version would be underutilized much of the time, the hardware required can’t be too intimidating); but doing it the creepy way has certain advantages: lots and lots of data for training your voice recognition, lots of handy social network data to mine to help draw inferences about ‘uncle’ and similar potentially confusing natural language usages, plus most of the big players (for consumer applications: vendors of call center hell IVRs and the like have similar tech but deeper pocketed clients with more interest in keeping data in-house) appear to be more optimistic about keeping it ‘in the cloud’ and using the service for some mixture of data mining, enhancing the appeal of their ecosystems; and making buying stuff from them easier.
If the helpful-spybot versions are available better and sooner(and it certainly looks like that is the case) it is entirely possible that local versions may be relegated to a niche of offline systems, entities with legal or trade secret reasons to keep things in house, and hardcore free software nerds even if they become adequately mature.
It has happened with various other things going ‘cloud’, even ones where the local software was available and mature for years before the eternally-at-the-vendor’s-whim version was out of beta; and in this case it’s the cloud version that is maturing faster.
