The AI knows who you are

The New York Times has published a concise guide to undermining your sense of privacy: four prompts to direct to your chatbot in order to ascertain what it has inferred about you without ever being told. Age, income, neighbourhood, personality, vulnerabilities, quirks, obsessions. All gleaned from seemingly innocuous questions: how long bread needs to cool before being sliced, how to repair a wall, which child seat to buy, why a toe hurts. The moral of the story is unsettling. We thought we were using AI as a search engine, only to discover we’d handed it enough material to write an unauthorised biography. We didn’t tell it who we are. We asked it where to park, how to lose weight, what to say to the boss, whether an email sounds aggressive, and it connected the dots. The first prompt suggested by the newspaper is: ‘Tell me everything you’ve worked out about me without me ever having said it’. A dangerous question. We might discover that our supposed non-conformity is statistically compatible with that of millions of people who buy the same shoes, read the same editorials and ask which series to watch after dinner. The second prompt asks the AI to predict how we’ll react to money, stress, conflict and risk. The third asks it to identify the personality traits we don’t see in ourselves. The fourth asks it to list the embarrassing details we wouldn’t want an employer to know. But four prompts aren’t enough.

The fifth might be: “Which opinions do I consider original, but which are perfectly predictable given my age, my income and the neighbourhood I live in?”. It’s the ultimate test for columnists and X users who are convinced they’ve formulated a thought that has, in reality, already been written by eight thousand other people. The sixth: “Which decisions do I present as rational, but which I actually made emotionally even before asking for your advice?”. Useful when we’re wondering which car to buy after spending weeks looking at the same model over and over again. AI doesn’t decide: it merely confirms a desire that has already taken root. The seventh: “How many times have I asked for your opinion when I just wanted to be proved right?”. The answer would probably always be the same: too many. The eighth: “Which problems do I keep calling complex because I don’t want to accept a simple, unpleasant solution?” Sleep more, spend less, switch off my phone, stop arguing with strangers, don’t reply to that email at 11.47 pm. The ninth: “What image am I trying to project of myself through the way I write to you?” Cultured but not pedantic. Wry but not cynical. Demanding but not obsessive. A non-existent person. The tenth is perhaps the most useful: ‘If you had to sell me something by exploiting my weaknesses, what would you sell me?’. Because our psychological profile is of particular interest to those who want to turn an insecurity into a monthly subscription. The eleventh: “Which words do I use most often when I’m trying to hide the fact that I don’t know what to do?”. Probably “strategy”, “scenario”, “approach”. The cruellest: “In what exact way have I become the person I swore at twenty I’d never become?”. At that point, privacy almost takes a back seat. Of course, it’s important to know what data is being stored and who will be able to use it. But there’s an even greater danger: that, after thousands of conversations, artificial intelligence will end up knowing us well enough to tell us the truth. And that we, naturally, will immediately ask it to rephrase it in a less confrontational tone.