Google to Use PC Microphones to Listen In?

seriv writes "The Register reports that Google plans to use PC microphones to collect statistics on a user's environment. Peter Norvig, who directs research at Google, told Technology Review that this software would start to show up in Google software 'sooner rather than later'. The software collects short sound clips and removes background noise. Google then targets its ads based on the statistics collected. With the current level of online privacy, this new level of invasion would seem to have frightening possibilities."

Comments

Let me be the first to say...

[Noryungi]

WTF happened to "Don't Be Evil", Google?

Of course, this may be just FUD, but I am pretty certain it qualifies as unlawful data collection and breach of privacy in my jurisdiction. Try to hijack my microphone, Google, and I will sue you to kingdom come. You have been warned.

A note to self: make sure the Google toolbar is uninstalled on every family computer ASAP.

Re:Let me be the first to say...

[Jahz]

don't be evil" (by the way, where the FUCK on any of Google's pages does it actually say that? I've looked fairly hard and not found it, nor 'do no evil'), but I think they've got all the best intentions.

"Our Philosophy" ... "6. You can make money without doing evil."
http://www.google.com/corporate/tenthings.html

Yeah, right...

[MythMoth]

For once I have not read the 'effing article.

The Register is not a reliable news source. Moreover, Andrew Orlowski has a bee in his bonnet about Google and constantly writes articles attacking them with very little merit - I would be astonished if this article is not by him, but even if it isn't, their association with him completely discredits them in my eyes.

Finally Peter Norvig is the author of the seminal Paradigms of Artificial Intelligence Programming (if you haven't read it, go and buy it right now) and is definitely not a complete idiot - I simply don't believe the story as summarised in the slashdot writeup regardless of whether it correctly reflects El Reg's article.

Case dismissed.

Old News

[AnswerIs42]

Two months late for this story...

http://www.techcrunch.com/2006/06/08/google-resear ch-prototypes-ambient-audio-contextual-content/

The original source

[clobbersaurus]

Here's the article cited by The Register.

How to counter data mining.

[Opportunist]

I'm a statistician. And if there is one thing that's worse for a statistician than getting no data, it's getting poisoned data. Data that has been deliberately or accidently been tampered with to generate results that deviate greatly from a result you would get with normal data.

If your want to counter data miners, give them what they want: data. You certainly can't give them more than they can handle, but you can give them false data. False data is worse than no data, because instead of getting no data from you, you are invalidating all data gathered.

It's not 4 bytes per 5 seconds! Calculations here

[rbarreira]

Actually it's not true that 4 bytes are generated each 5 seconds, it's much more than that. From the paper:

The audio-identification system starts by
decomposing each query snippet (e.g., five-seconds
of recorded audio) into overlapping frames spaced
roughly 12 ms apart. Each frame is converted into a
highly discriminative 32-bit descriptor, specifically
trained to overcome typical audio noise and
distortion. These identifying statistics are sent to a
server

So, even ignoring the fact that frames are overlapping, we have 32 bits per 12 miliseconds, which means more than 2600 bits per second! More than enough to code speech, even without speech recognition algorithms! The Speex codec (which is optimized for encoding speech) can code human speech at such low bitrates as 2.15 kbit/s...

Conclusion if you're not willing to trust what Google says, they could perfectly be sending your speech over the internet to their own servers.