Could AI Recommendations Make Your World Smaller? [Raw Session]

00:00 — Recommendations are getting almost too good

Hey, welcome back to Slow Builds.

I've been thinking about recommendations lately. Not because—obviously recommendations aren't bad. You do get bad ones, but actually, again, like the complete opposite: how good recommendations are getting now. How the ones that are being served up to you, and what you're being shown, are actually fitting what you're looking for. Or unexpectedly, it's what—magically, it's almost like you were thinking about it and something heard you, and then it pops up in your Facebook feed or your X account or shows up in your search.

It feels random, but I know it's not. But the recommendations are getting so good that it feels eerie in a way.

YouTube is getting pretty good at figuring out what I like to watch. Apple Music, really good at figuring out my next listen, what my favourites would be. Or it's like, you know, play this artist and similar artists and similar songs type thing. And it works really well. I'm very impressed with it.

Amazon's pretty good at showing me, especially when Amazon Prime Days come around and stuff. It's really great at showing me recommendations of things I might be interested in, things related to what I've already bought. And it takes a lot of guesswork out of me.

It's bad. It's good in that situation. It's bad because it shrinks the wallet and it adds clutter to my life. It's good because it helps me learn control, but it also helps. A lot of times useful things come out of it.

01:43 — AI can understand why I like something

And now AI—and I know AI is in the back of that a little bit—but AI is taking it further, like much farther. Because instead of just looking at what I've clicked on before, I can actually have conversations with it. I can tell it what I like. I can tell it what I don't like. And it remembers everything about me and everything I did.

Eventually I could probably just say, "Find me something to watch tonight," or, "Give me a book I'd really like." Even, "I'm going away for five days. Plan me a trip."

And that's a pretty good experiment. It's an app idea I had many years ago. I have a three-day window or a five-day window, like a random window. It's this time of year. I'm flying from this location. I could drive if possible, and these are the people coming, and these are what everyone likes. Find me the best deal.

It was an idea I had a long time ago, but now with AI it pretty much seems like it's very possible.

02:50 — Could getting exactly what I want limit me?

But I start wondering if there's a downside to getting exactly what I want all the time. Because if something gets really good at predicting me, does it also start limiting me?

And that's a different way to look at it, because recommendations aren't new. We've had them forever. Netflix recommends the next watch. Spotify builds playlists. YouTube fills my homepage. And sometimes it's great. I've definitely found things that way.

But those systems have always felt a little dumb in a way. You know, very rigid. You watch one video about fixing something on my Jeep and suddenly my whole thing is filled with Jeeps. Jeep stuff, overlanding, [unclear] soft tops, whatever.

Searching for hotels somewhere, and now every ad is about that location. It could be about that hotel. It could be a trip. But mostly it's around, like, "Oh, you're planning a trip to San Antonio. Well, here's everything about San Antonio: what to do, where to go."

They see behaviour, but they don't necessarily understand why you did it. AI is changing it and has changed it a bit, I find, because now I can explain myself.

I can say, "I like this book, but I don't like this part," or, "I usually like these kinds of movies, but I'm tired of watching the same thing right now."

That's a very different recommendation system. It's not just watching what I do anymore. I'm helping it understand why.

And if these systems keep getting better, I think they're eventually going to know my preferences extremely well. Maybe better than I know them myself, because they'll remember those nuances, those little nitpick things that I've come up with that I might forget. They might forget that I don't like a certain type of sheet at a hotel or a pillow made a certain way, or I need two pillows rather than one. I don't need it. I'm just giving you an example here.

But it would remember those things, and it would almost feel like a tailored concierge in a way.

04:47 — The recommendation loop

And this is where I think it's getting interesting. Let's say AI figured out that I like a certain kind of music, so it recommends more of it. I listen to that. Now it has even more evidence that I like that kind of music, so it recommends more. I listen to more and the prediction gets stronger.

That's a loop, and there's nothing necessarily wrong with that. I'm getting music I enjoy. The system is doing exactly what I asked it to do.

But after a while, I wonder what happens outside that loop. Because the system isn't necessarily showing me the entire world. It's showing me the part of the world that has the highest probability of keeping me interested.

And every time I respond positively, I'm helping reinforce that version of myself. It starts becoming, "You liked this before, so here's more of it." Then tomorrow, "You liked more of this, so here's more."

Eventually, how much of what I'm seeing is actually me choosing? And how much of it is a system getting increasingly confident about what it thinks it knows about me?

05:45 — What happens when I change?

There's another part of this that bothers me a little bit. People change. My interests aren't the same as they were ten years ago. There are things I like now that the older version of me probably wouldn't have cared about at all. And there are things I used to spend a lot of time on that barely interest me anymore.

That's normal. But an AI is going to have history, potentially years of it. It might know what I've watched, what I've read, what I've asked about, what I've bought, and what I've ignored.

So what happens when I start changing? Does it notice quickly, or does the old version of me keep influencing what the new version gets shown?

I don't really know. I don't think any of us know what's going to happen. Maybe these systems will be incredibly good at detecting that.

But there's something strange about having your past preference constantly helping decide what your future options look like. Because sometimes I don't know that I'm interested in something yet. There's no history for that. There's nothing to predict that.

06:43 — When AI starts shaping what I try

And this gets bigger once AI moves beyond recommending music and videos and purchases. This is the part I'm finding extremely interesting here.

Because imagine I'm planning that trip. I could give AI everything it knows about me and say, "Build the perfect five days." It might know I like walking, all kinds of food. I like activities, things to avoid, and those kinds of hotels or Airbnbs, what my preferences would be.

And that's incredibly useful. I would use it. I'm not saying I wouldn't, and I 100 percent will use it, and I still want to build it.

But if I do that every time, I'm basically asking AI to keep recreating vacations, variations of trips I've already enjoyed.

Something like books or movies, hobbies, or even people. We've already seen technology recommend people to follow, communities to join, things to learn. Eventually AI might help decide where I go, what I try, what I read, and who I even interact with. Not because it's controlling me, because I'm asking it to help. And that's what makes this a little bit different here.

I don't think this requires some evil algorithm trying to manipulate everybody. It could probably—well, there's one company I believe would do that. It could happen with a system that's generally trying to help you.

If I ask, "What's the best option for me?" the AI should probably use what it knows about me. That's what makes the answer personal. But "best for me" usually means based on the person I already am. My existing preferences, habits, history, maybe even my biases.

And I don't always want those reinforced. Sometimes I probably need something that doesn't fit. Something I initially don't understand, might even dislike at first. Because that's part of how interests change. That's part of how growing and opinions change.

And I don't know how you optimize for that. How does an AI know when to give me what I'll probably like and when to give me something completely outside of that?

08:51 — A discovery mode for AI?

Maybe eventually this becomes a setting. That's an interesting point, because it is a system. It is code. It is ones and zeros. So a toggle, a setting, a threshold, a temperature, anything like that is possible.

So I don't know. Maybe you tell your AI, "Don't give me the obvious answer," or, "Show me something outside my normal interests." Maybe basically a discovery mode. That'd be fun.

Because I don't think personalization itself is the problem. I want personalization. If I'm asking AI for help, I don't want it pretending it knows nothing about me. That would throw away one of the most useful parts of technology.

But maybe there's a difference between "know me" and "keep giving me more of me." Those sound similar, but I don't think they are. One helps the system understand where I'm starting from. The other could potentially keep me there.

09:41 — From knowing me to influencing who I become

That's probably the strange thread running through these last few videos. In the first one, I was wondering whether AI can really know me from the data it has. Then I started thinking about how every AI actually knows a different version of me. And now I'm wondering about what happens when those versions become good enough to start guiding me.

Because eventually this isn't just about whether AI understands who I am. It might start affecting who I become. Not dramatically, probably not in the same science fiction way. Just slowly, one recommendation at a time. A choice made easier, options moved to the top, one thing I never see because something else was predicted to fit me better.

And maybe most of the time that's completely fine. Maybe it's better for me. I just don't know yet until we get there. We can't really think that one through until we see how society moves with this.

I use AI every day and I want it to know me better. That's part of what makes it useful. But I don't think I want every decision optimized around the person I already am, because I'm still changing. I'm still finding new interests. I'm still changing my mind about things. And sometimes I don't know what I'm looking for until I find it.

So maybe the question isn't, "Can AI predict what I'll like?" I think eventually it'll be extremely good at that. It already is. Maybe the harder question is, "Can it know me really well without trapping me inside its prediction of me?"

Can it recommend something that's technically wrong for me today, but might become part of who I am tomorrow?

I don't know. But I think that's a much more interesting version of personalization than just getting better recommendations.

11:26 — Unscripted closing thoughts

The more I read about—script is over. Thanks for watching. But this one, I almost didn't do this one because it was mixed with the other ones too much. But going through it now, and even ending on a shorter video, this one really feels like it hit home here for me.

Like, are we—I don't want this to move. I just finished reading 1984 also. But I don't want to see this one as being creepy, Big Brother, society's pushed to one direction through what we're fed type thing.

But I do believe we need to keep an open mind. We have to question everything. We have to find ways. A good AI, a good prediction system, a good recommendation system is going to pull on other users, hopefully, that have similar interests, but are ones that are outside of the norm that I'm normally used to.

And I hope there's ways that we can build on this. And I think a lot of really interesting things are going to happen. Again, exciting times. I can't wait to see where this thing—every day changes, and I'm so excited.

All right, thanks for watching. Bye.