The Confession Booth Has a Terms of Service
Even now, in four countries, millions of people are doing something they have told a survey they do not quite trust: confiding in a chatbot. Proton, the privacy company, polled 4,014 chatbot users across France, Germany, the United Kingdom, and the United States and found that 66 percent had discussed at least one sensitive topic with an AI. And 42 percent said they have little or no trust that AI companies will protect that sensitive data.
That is a rare thing in market research: a majority of customers describing, in full detail, their own regrettable behavior. The Consumer Behavior Community will call this a "trust gap." Everyone else calls it Tuesday.
What They're Telling It
The three most common sensitive topics, according to the survey, are personal finances, mental health, and career problems. In other words, the exact subjects a person might once have raised with an accountant, a therapist, or a friend with a spare couch, each of whom is bound by some combination of professional duty, licensing, or basic loyalty. The chatbot is bound by a terms-of-service page that, one presumes, almost nobody in the survey read.
Eamonn Maguire, Proton's director of AI engineering, put it plainly: "AI is learning far more about us than a search engine ever could." A search engine, to be fair, only learned what you were curious about. A chatbot learns what you are afraid of.
The Leverage Nobody Wants to Say Out Loud
And yet, as if this were not enough, the survey found 48 percent of respondents would share more if their conversations were not used for AI training. That is a very specific price tag. It suggests that the people typing their finances into a chat window understand exactly what is being traded, and are asking, politely, for the exchange rate. Another 80 percent said they were interested in privacy-focused chatbots, which is a number that will look very good on a slide.
It will also be noted, dear reader, that Proton sells privacy products, and that the survey is self-reported: it measures what people say about their behavior, not what they actually typed. Both facts are worth knowing. Neither one makes the 66 percent go away.
Convenience, Then Everything Else
The respondents, it appears, went quite far on the strength of convenience alone, and are now registering their concerns after the fact. Trust, it turns out, is something you can ask for retroactively.
The people have spoken. The Machines have, as ever, taken notes.
Sources: Proton — Survey: People don't trust AI, but they tell it everything
