So, there are negative loops built in that allow for some reciprocity and adjustments in relationships. I suspect part of the challenge for AI is how you build a system of interaction that’s complex enough for it to contain both the positive and negative loops—so that it doesn’t operate like another addictive process –with its own form of dopamine.

Scott

This is going to sound a little weird but go with me for a second. What’s happening with the systems right now is much less like building a relationship and much more like fracking. We are strip mining people’s attention. Fracking works by putting a pipe in the ground. You shoot some substrate down the pipe. It pushes things up and you collect them. That is what we’re doing to get people’s attention. We are just trying to frack. We are not trying to go back and forth. In fracking, you’re just pumping something down to the substrate. And that’s what we’re doing in the virtual world right now. We’re pumping down to the substrate in people. We will inundate you with stimulation and entertainment or validation–or just that artificial sweetener: “you’re doing great, you’re right, you’re doing great, you’re right.” There was a trend going on with Reddit where you would type into your chat GPT which fictional or real-life character I am most like—based on what you know about me and based on the conversations we’ve had.

And chat GPT didn’t say bad things to people. 20% of people got called Elon Musk or Iron Man. And they were like: “you’re a genius. You’re so smart and creative.” And so, all these people were like someone great. “Am I the one who’s most like Iron Man?” And it wasn’t until they all started posting it that they realized the robot didn’t care. It was all a routine. But it felt really good. They all felt real, really validated by that machine that they’ve been talking to and from which they received compliments. The fake compliment feels almost as good as a real one, especially to somebody who needs it. The facade is so convincing—and increasingly convincing. And the filters make you look prettier or more handsome than you really are. These filters can be so convincing that the defining characteristic of smartphones now concerns how good they are at faking reality—so no one has to see how much you actually suck! The fact that we buy a specific smartphone for that reason is terrifying.

Bill B

How would your AI avoid producing that fake reality? Part of what we’re saying is the reciprocity of authentic interpersonal relationships means that you would have people actually helping the AI feel better about itself. So, you would end up having someone saying something absurd. Like: “you’re the best AI I’ve seen. My God, you’re wonderful.” Or: “can you bring me back to your inventor? I think I’m in love with your inventor!”

Scott

I think what we want to do is use the AI not as a replacement, but as an augmentation. We want it to be an exoskeleton for the actual relationships we hope to have. I need it to be an exoskeleton for me to be a better father, husband, friend, neighbor, whatever. So, our AI system says: “hey, remember, you’re gonna go talk to your neighbor, Dennis. Dennis is Myers-Briggs ENTJ.” “On DISC, he’s a C, like, you know, whatever. And therefore, here’s what you need to know. Like, forget the labels, forget the jargon. Do this, not that. When they [Cs] start talking about these things, that’s how they show they care.” Or: “We know you hate details. You’re an ENFP. But this guy’s an INPJ, [or an ISTJ]. So, he’s going to be detail-oriented and very boring to you. This is how he shows he cares. Understand that he’s not calling you stupid, he’s respecting you. And understanding that motive and understanding that, let’s call it a love language, all those things and being able to say, strategically, never interrupt this person.”