In the video, they say they have compressed translate, vision and speech into "just over 100kb."
That seems pretty magical. Perhaps it is all server-side?
That seems pretty magical. Perhaps it is all server-side?
Probably.
> translate, vision and speech
To me, these types of technology are both wonderful, and particularly dangerous.
With Google Translate, I was able to read technical posts on a Russian forum, which contained a piece of critical information to the problem I was trying to solve. And with Google's speech recognition on YouTube, highly accuracy text transcripts are now generated at an industrial scale, and even in real-time, that enabled me to see parts of unintelligible speech in some videos which are otherwise impossible to understand. In the future, it's expected that these types of technologies would see greater applications, for example, using automatic translation in international conferences or private conversations is not a "far future" Sci-Fi idea anymore, and may eventually be crucial tools for a specific propose, or even play a central role in the human society.
> The implicit message was clear: “Yes, we have all of your data, but the fact we have all of your data is a good thing, because it allows us to make your life easier.”
On the other hand, due to the very nature of these types of technology, it's mostly impractical to deploy them locally, the only natural way to implement these is running them on Google's computers. It means soon enough, Google, would have access to massive volumes of information, not only personal, but human information, such as private conversations and communications. And unlike those cheap IoT devices you could refuse to purchase, using these services may be essential (e.g. being able to understand the conversations) to live in a digital society. And even if it's not Google, the information collected would be vulnerable from another service provider as well, in addition to a government subpoena.
To some extents, the search engine already showed the existence of these problems - it's not possible to run a search engine on your computer (I know there are some FOSS Proof-of-Concept projects, but they are what they are...), the search engine has to be operated by a large centralized provider which would be able to see and track everything you've searched. Meanwhile, you must use an search engine to find information produced by the human civilization.
In the future, it's certain that Homomorphic Encryption and Differential Privacy may be important and must-have solutions, but only for some very specific applications, they allow some computations and queries to be performed by a 3rd-party without revealing their contents or nature. Theoretically, if there's a future breakthrough in applied cryptography, you would be able to execute an entire blackbox program on someone else's computer without revealing anything.
But I doubt if there's a solution in general.