Google’s Gemini AI looks spectacular on the new Pixel 9s. MKBHD is impressed:
And so is WSJ’s Joanna Stern (I’m starting to appreciate her “watchful mom” kind of attitude), though she’s a bit harder to convince. She sat down with Google’s Rick Osterloh, head of Android (and apparently Chrome as well), and got him to say that Google does not use its user’s personal data on Android to train its AI:
“Are you using personal data to train Gemini and other Google LLMs?”
“No, we’re not using personal data to train LLms. We are using information that users have given us the ability to train on, or that is from the public web.”
“To confirm: my phone calls, my emails, my calendar appointments, my text messages, those are not being used to train (Gemini)”
“No, these are not being used, that’s right.”
But what exactly is “information that users have given us”? Is it attained through yet another long Terms of Service agreement deep inside Google’s in-house apps, like Google Maps or Gmail, that users accept without ever reading? Can Google get this information from Reddit, which suggests you sign in to your Reddit account with your Google account by default?
I appreciate Osterloh’s optimism about Gemini. I believe Google wants to use AI to enhance the productivity of Android users and not to spread more propaganda on Twittex.
Still, most Android users will probably share personal and private information with Google without their knowledge and consent, even if it’s out of ignorance. At the same time, offering a tool that can potentially make spreading misinformation easier means that some folks will use it specifically for that.