“Tell me about your hobbies. Are you into model rocketry?”
“What?”
“Model rocketry.”
“No,” Greg said, “No, I’m not.” He sensed where this was going.
The man made a note, did some clicking. “You see, I ask because I see a heavy spike in ads for rocketry supplies showing up alongside your search results and Google mail.”
Greg felt a spasm in his guts. “You’re looking at my searches and e-mail?” He hadn’t touched a keyboard in a month, but he knew what he put into that search bar was likely more revealing than what he told his shrink.
“Sir, calm down, please. No, I’m not looking at your searches,” the man said in a mocking whine. “That would be unconstitutional. We see only the ads that show up when you read your mail and do your searching. I have a brochure explaining it. I’ll give it to you when we’re through here.”
“But the ads don’t mean anything,” Greg sputtered. “I get ads for Ann Coulter ring tones whenever I get e-mail from my friend in Coulter, Iowa!”
The man nodded. “I understand, sir. And that’s just why I’m here talking to you. Why do you suppose model rocket ads show up so frequently?”
Greg racked his brain. “Okay, just do this. Search for ‘coffee fanatics.'” He’d been very active in the group, helping them build out the site for their coffee-of-the-month subscription service. The blend they were going to launch with was called Jet Fuel. “Jet Fuel” and “Launch”—that would probably make Google barf up some model rocket ads.
They were in the home stretch when the carved man found the Halloween photos. They were buried three screens deep in the search results for “Greg Lupinski.”
“It was a Gulf War–themed party,” he said. “In the Castro.”
“And you’re dressed as…?”
“A suicide bomber,” he replied sheepishly. Just saying the words made him wince.
“Come with me, Mr. Lupinski,” the man said.
Read the story, its short and bittersweet. (Thanks Bill!)
The situation above raises an interesting question about AI, as I use the term. No person reads your GMail, but Google’s machines do read your email to serve up ads. I’ve argued before that this situation does not neatly align with our normal conception of privacy and generates conflicting intuitions about privacy violations. The argument presented by the TSA agent above suggests that seeing what ads gets served is not a violation of privacy: reading emails and searches is unconsitutional, but looking at the ads and other metadata is not. The argument, I suppose, is that because a machine read your emails, and not a person, your privacy has not been violated even when the results of that machine’s evaluation are used against you.
This presents two possible solutions to Cory’s dystopic vision:
Either we don’t allow anyone unauthorized access to even metadata, which seems to be technologically unfeasible given the structure of the internet
Or we consider the machines that read your email in the first place to constitute an invasion of privacy, contra Google’s own policy.
Of course, I have theoretical reasons for preferring the latter, but I don’t think this kind of argument proves anything. So I still am not sure how I fall on these sorts of privacy issues.
Thoughts?
An interesting puzzle indeed. Given my stance on machine intelligence (or lack thereof), I can’t say that allowing Google to read my emails is a violation of privacy. It isn’t. Google’s “observation” of my emails is no more invasive than an unmonitored video camera’s recording of my shower counts as voyeurism–if something has no consciousness, then it really isn’t observing at all. The problem arises, it seems to me, when someone looks at those tapes later (or reads Google’s output about my emails). I think the cases are almost exactly analogous, right down to the degradation of information that’s present in each of them.
Think about what it would be like for someone in the video camera case to make a similar defense: “Well, we weren’t _really_ watching you shower, we were just watching what the machine tells us about what it looks like when you shower. We can’t for instance, act as active participants in our observation in the way we can when we’re _really_ observing something; we’re stuck with the static two dimensional image that the camera provides. For another, the resolution of the camera is really terrible, and everything looks grainy and washed out. Real vision isn’t like that.”
The response to this, I think, would be to point out that even if the information from the camera isn’t perfect, it is enough for you to make at least a reasonably accurate estimation about what it looks like when I’m in the shower, and that’s an invasion of my privacy. Even though the observer isn’t directly observing me, the data he’s getting from the technology lets him act predict with a high degree of accuracy and certainty what he _would_ see if he were observing me.
The same seems true of Google and the use of Google’s metadata; if the metadata gives enough information for a third party to infer the contents of my emails with a reasonably high degree of accuracy, then that should also count as a privacy invasion, even though the only “agent” invading my privacy has no agentive function at all.