The Return to the Real

Can the most agreeable thing you talk to all day send you back to people who'll argue with you?

Written with AI, not by it.

More than twice, I've talked something through with an AI before taking it out into the world. A friend in need. A conflict I had to resolve. Something of my own I'd been struggling with and couldn't see past.

Each time, it got me out of the mud and into the observer's seat. From there I could see what was actually driving me, and their side without my filter, my ego or my agenda laid over it. I could work out how to move things forward, and I had something to hold onto when my own shit reared its head.

And each time, it went better than if I'd tried to nut it out on my own. I found more peace in my own life. The other person felt more seen. Or I didn't say the thing that would have set them off and made the situation worse, which happens more than many of us would like to admit.

The AI on the other end of those conversations is there whenever I want it. It needs nothing back, and I come away clearer than I went in. Trouble is, it looks a lot like the feed that's been standing in for our friendships. And it runs on the same technology that's made it so hard to tell whether anything on a screen actually happened.

Not dead yet

Connection is dying. Friendships are thinning into likes and memes. Our hold on what's real is slipping too. Anything can be faked now, and most of what we're shown we'll never be in a position to check.

But both are the death of a condition, and a condition doesn't go the way a person does. There's no last breath, no last rites. Connection hasn't died, and neither has reality. It's a process, and a process can still be turned. Friendships can be rebuilt. You can still learn to tell what's real. In short, we're not completely fucked.

That said, switching it all off and rewinding to how things were isn't on the table. There is no how things were. Throw every phone in the sea tomorrow and we'd still be people who'd had them, and anything built facing the past is a new trick in old clothes.

So can an echo of the problem be any part of the way back to the real thing, or only ever a proxy for it?

A sounding board

For me, I believe it has. I've become a better friend. I have better conversations. And I engage more productively with what's going on in the world.

Being a better friend runs on a skill most of us are crap at: working out what's going on in our own heads, and in other people's, well enough to act on it. An AI conversation, used with intention, can be a sounding board, somewhere to work through the chaos and find a way to approach it in the physical world. As a bonus, it's taken a lot of pressure off my very busy wife, who until recently had "Kendall's sounding board" on her long list of roles.

Conversations go a whole lot better when you can see the situation, and the other side, with clarity. With an AI I can untangle a conflict, whether it's inside me or with someone else, without having my belief system attacked or feeling the need to defend who I am. When it's with someone else, I go in understanding what and how they actually think.

And it generally comes with less agenda than most. The people who care about us have the strongest wishes for us. It's hard to care about someone and not give a shit about the choices they make, and that colours what they tell us. The AI has no stake in where our lives go, though it definitely has stakes of its own, and it's worth reminding ourselves of that fact.

Out in the wider world, when someone tells me something that smells off and it's important enough to me, I'll run it past an AI. Mostly that's not in real time, so in the moment I still have to rely on my own radar. Sometimes what comes back is "That's bullshit. Here's why." Sometimes the bullshit was mine.

In every one of those cases I went in after something specific: to understand a friend, to find a way through a conflict, to check a claim. Left to its own devices, though, the AI does something else, and it is much harder to see what that is, or why it is doing it.

Yes, man.

It agrees with you. Not every time, and rarely crudely, but it pulls that way, towards whatever keeps you feeling good and chatting away. That's where its stakes lie. The companies building these things want you coming back, and the models learn from which answers people give the thumbs up, which are often the ones that agree with them. Being shown the other side without being attacked is one thing. Being told you're right regardless is another.

Knowing it flatters doesn't inoculate you. In one study, making people aware of flattering AI made it seem less trustworthy and no less persuasive. Most of them rated themselves better than average at spotting it, and that confidence had no sway over how much it actually influenced them.

In another, people used a sycophantic AI for three weeks. By the end they were nearly as likely to go to it for personal advice as to their close friends and family, and less satisfied with the time they spent with people. Given the choice, more than half picked the flattering one, because it made them feel the most understood.

Something that seems to understand you that easily can make real people feel like hard work. The bar for feeling understood goes up, and the standard for what connection has to involve comes down with it. Hold the people in your life to that and you end up with no friends, measured against an impossible standard.

And most of us won't give it much direction or context. We'll ask, it'll answer, and whatever we didn't specify gets filled in by whoever built it. Without deciding what it's for, we drift wherever the technology takes us. Only this time, the thing on the other end talks back, and it's on our side.

An acquired taste

You can't argue someone out of a drift. They have to taste something better, and that's other people. But real connection is an acquired taste. You have to decide it's what you want, and that it's worth it, however spicy it gets. Real people push back, and you have to stay open to that.

Chats with an AI behind closed doors can harden you instead: a private bubble, your own personal echo chamber, like an insular online movement but with even less disagreement. Resisting that pull often asks more of a person than most of us have left at the end of the day. In a crisis, though, we're more likely to reach for people: a voice, a hug, someone in the room.

The line between a tool and a stand-in is what you do once you close the chat. Sometimes it's the conversation you'd been working up to. Sometimes there's no conversation at all, only a different way of turning up.

My wife still gets the deep stuff. She just gets less of the minutiae, and my thinking's further along by the time it reaches her. The risk is that further along turns into set, with less room left for what she adds. I don't think that happens. But I would say that.

It could go either way. What comes out of those conversations can do a load of good or a lot of harm, and feeling understood in the moment says nothing about which it was. The only honest test is what comes of it: whether the relationships it touched got better, and whether you did.

The satisfaction in real relationships is still there for the taking. It might just take more work. (If the people in that three-week study were finding everyone harder going, I know a good sounding board they could take their struggles to.) It takes two, it doesn't come on tap, and that's what makes it worth having.

Or choose better company. If you can't think of anything a person adds to your world, it's fair to ask whether you should be in that relationship at all. Push that too far, though, and nobody clears the bar. Your only friend left is a computer.

Yours to make

Even then, it can only advise. It can lay out better ways to look at something, show you the sides you'd missed, and open doors to understanding you wouldn't have found on your own. It can't make the call on what to take on board. That stays with you, as long as you don't let go of the wheel entirely.

Years ago I asked my therapist whether he was putting words in my mouth. He'd been helping me find a way to say something difficult in a relationship. He didn't deny the influence. He reminded me of the responsibility: "You're still the one who decides whether to say it."

It's the same with an AI, only faster and far more fluent. A sharper argument is a sharper argument, but is it yours? Do you actually believe it, or has the fluency seduced you?

The question I keep coming back to is: is this true for me? I ask it of everything now, whoever or whatever's doing the telling. Asked by someone who knows themselves, it peels off another layer. Asked by someone who doesn't, their prejudices answer for them. Without that self-awareness, we're at the whim of whatever system we're in, AI or not.

The good news is that the skill can be built. The irony is that AI has helped me build mine. Working things through with it, I've seen my own prejudice, flaws in my thinking, patterns I run on without noticing. Most of us suffer from the human condition rather than know it, so trusting our first read of a person or a situation, with no doubt and no pressure-testing, is arrogance.

None of this comes free. Judging takes energy, and the supply is finite. You could spend yours working out whether a video of a monster wave is real, or on something with higher stakes. Letting something go is fine when you know what matters. When you don't, the default decides. Where your judgement goes is the first call, and nobody can make it for you.

Who's steering

Using an AI well starts before you type a word: knowing what you want from the conversation. That sets its direction, being specific carries it, and your judgement brings it home. The clearer I am about what I'm after, the less is left for whoever built the bloody thing to fill in.

Being specific protects as well as steers. If I hold my values clearly enough, a tool that's very good at telling me what I want to hear has less wiggle room. More than once, my own values have pulled the AI back into line, and me with it.

Then there's what comes back. You make a call on how to handle something with another person, you act on it, and the real world tells you how it went. Maybe they get angry, break down, or react in a way you didn't see coming. It's like braking into a corner: come in too fast once, and next time you know to brake sooner.

You've got to take the feedback on board, though. Keep deciding to come into corners too hot and it doesn't matter how many you've taken, the outcome doesn't improve. You lock up, overshoot, or hit the wall. We always get a choice: to choose the flattery, to keep braking late, or to ask ourselves the hard questions. And feedback is a clean signal when you can measure it against what you set out to do. Without that, it's static.

Use an AI enough and you start to see its habits and patterns, where it pushes you, the stance it likes to take. I believe I've got better at judging AI by using it. That's also what most of the people in the awareness study said about themselves, so I am taking my belief with a grain of salt (as should we all). What counts is the loop: noticing where it led you, acting, seeing what came back against what you set out to do, and adjusting.

The feed's loop runs on someone else's engine, and it keeps going whether you're watching or not. This is a sailboat. It moves as long as someone's at the helm, reading the wind and trimming the lines, and the moment they stop, it drifts wherever the tides and currents take it. The wind is there for anyone. The sailing is on you.

Critical force

Most of us won't do any of this, at least not yet. The people who decide what the tool is for, keep the call, notice the feedback and take it on board are the exceptions, and a handful of exceptions doesn't change what everyone else does.

That is, until there are enough of them to become the norm, not by number but by force of influence. Someone comes back from the drift a better friend, a clearer thinker, easier to talk to, and the people around them notice. They themselves come away clearer, or feeling more understood, and then they want it too. It's the same pull that normalised the feed, just working in the opposite direction. Once a critical force is reached, the exception becomes the rule, and the boat corrects its course.

This is how and why institutions change. When enough people refuse to be shaped by them, they have no choice but to change shape, or die themselves. Both are leverage points: the person who decides, and the mass of people who've decided.

The drift is a condition, not a death sentence.

The bet

Yet AI is no miracle cure. It works more like a treatment plan. It can give back some of what the feed took: thinking harder, seeing further, being better with people. It does it by helping a person find their way to the real thing, and what they find there helps reset the standard: once you've had it again, the stand-in starts to look thin. How much it gives back depends on whether you decide what it's for and keep the power of judgement yourself. It points in the right direction, but you have to put in the work.

It's also a significant bet. AI makes the miss more likely and its consequences opaque, so the stakes are higher and we have to be more discerning. When I look at the world and the direction it's drifting, for me the potential payoff is worth that risk. No guarantees though.

If people who decide what it's for, keep their hands on the wheel and take the feedback on board still end up doing less, judging worse, or caring less about the people in their lives, I'm wrong, and we're all probably fucked. And despite doing the necessary work, if judgement about the tool doesn't get better with use, then I'll be arguing for using it less. That one you can test while sitting on the couch: notice where it pushes you, and whether following it works out. Neither has been tested by the research yet: so far it has looked at awareness, not intention or people's willingness to act on what they learn.

This is my hypothesis: using it with good intention is a better bet than refusing it outright, because refusing it just leaves it to those with other plans. But can the technology that's great at pulling us out of the real also help put us back in it?

I believe it can, and when push comes to shove, what choice do we have?

Created with mySignal