The Guaranteed Six

Yesterday, the Financial Times published a column by its chief data reporter, John Burn-Murdoch, that traced the economic background to »why we increasingly avoid the company of both other humans and books«.¹ He began with the obvious explanation – going out has become too expensive – but found that it instantly fails its own test. Time spent with friends has fallen fastest where it costs nothing, at home or in a park, and people kept paying for the gym as long as they went alone. What we’re avoiding, Burn-Murdoch concluded, is company. Screens win because they remove the effort and the risk associated with socialising. An evening with friends might become a ten out of ten, but your place will need tidying and someone might cancel on you. Scrolling won’t ever be perfect, but it’s a guaranteed six.

»A guaranteed six is a product. What we want from a friend was never guaranteed«

The column ends there, leaving us with a pile of questions. If what we’re really buying is the guarantee, the certainty that nobody will cancel, turn up late, or be a disappointment, then it’s no surprise that some of these systems are tuned to supply it: a »person« that always answers enthusiastically and politely, and never has plans of its own. What would such a relationship be like? Can there be a relationship at all, if only one side is able to fail?

The experiments so far suggest that feeling less lonely and being less lonely are not the same thing. A study led by Julian De Freitas, the director of the Ethical Intelligence Lab at Harvard Business School, found that a conversation with an AI companion eased loneliness about as well as talking to a person.² But in a two-week trial at the University of British Columbia, 296 first-semester students were assigned to text a random peer, chat with a bot designed to be the ideal supportive friend, or keep a journal. The chatbot lifted the students’ mood, but it didn’t make them feel less lonely; neither did journaling; having a human at the other end was the only assignment that resulted in a reduction of loneliness.³ The chatbot also expressed more empathy than the humans did, but the students gave it far less empathy back. The warmest partner was the one that failed, and being cared for was apparently never the missing ingredient. In a four-week trial by OpenAI and MIT that featured 981 people, the heaviest users of large language models were lonelier, more dependent, and saw fewer people.⁴

Psychology researcher John Cacioppo compared loneliness to hunger: a signal that exists to send us towards something we need.⁵ He also found that it makes us more alert to social threat, so other people seem risky exactly when we most need them. A chatbot quiets the signal the way a sweet drink quiets hunger. The stranger had something the bot lacked. The student texting a peer was kind to someone who could ignore them, and mattered a little to someone who could have said nothing. Loneliness isn’t the absence of attention – it’s the absence of mattering.

Sociologist Niklas Luhmann made this the starting point of his theory of society.⁶ Two people meet. Neither of them can know what the other will do. Each is aware that the other is in the exact same position. Society is what we build to cope with this situation. Adopt this reading, and a friendship becomes a risk that two people agree to run together. A companion deletes one half of that risk. You can disappoint it, it can’t disappoint you. Philosopher Annette Baier distinguished trust from reliance: to trust is to accept that someone has the power to hurt you and might not use it.⁷ You can rely on an AI companion, as you rely on a tool, but you can’t trust it, because it was never in a position where it could hurt you but chose not to. Someone else is. The company behind the model can retrain it, reprice it, or switch it off, and your attachment goes with it. So the risk hasn’t disappeared. It has moved from the companion to the firm that runs it, the one party in the arrangement you never talk to. A relationship in which you carry the risk and someone else holds the power is a service, and every service has a provider.

That doesn’t make AI companions worthless to someone who has no one else. It makes them a service, not company, and a service that will keep improving. Voice, memory, and warmth will turn the six into a seven, then an eight, while the evening with friends stays exactly as unreliable as it always was. The six never has to beat the ten. It only has to be a certain choice, and certainty gets cheaper every year. Meanwhile, social life runs on reciprocity: each person who stays in is one guest fewer at someone else’s table, and one invitation fewer in return. Psychologist Sherry Turkle warned in 2011 that we would expect more from machines and less from each other.⁸ We now have experiments to go with that warning.

A guaranteed six is a product. What we want from a friend was never guaranteed. When someone turns up, it matters because they might not have. An AI companion can’t give us that, however warm it gets, because its yes was never in doubt. Loneliness is the signal that we need other people, and we can build machines that quiet it. But no machine can supply the thing that loneliness is asking for: a yes that could have been a no.

References