AI Recommendation Algorithms and Human Freedom: Perspectives from Tech and Philosophy

James Brusseau speech

It's my sense that almost no work has been done in this area of discussion. So right in the present, what we are discussing is the intersection of recommendation algorithms, things like what movie Netflix suggests you might want to watch, and artificial intelligence. Now quite a bit of work has been done at this intersection for economic reasons. Companies like Netflix want you to stay on their site. So there is quite a bit going on there. But to my knowledge, very little or perhaps nothing has been done in the area of human freedom, AI, and recommendation algorithms. So the plan today is to begin mapping out what this area of investigation might look like, what kinds of questions arise, what dilemmas exist, what pathways forward might we find.

So I believe that everything is very, very contemporary. However, I also believe that the questions circulating through this area of investigation, the questions perhaps started more than 300 years ago with a philosopher named John Locke, who proposed the following thought experiment. What he asked us to imagine is a situation where we are in a room which is extremely pleasurable. That is, you can see here, we have wine and we have a wonderful view, have a nice something to lay on. You're in the room doing whatever you like. You are happy. You have no reason to leave. You don't want to leave. So you spend your time doing your little things in your room. But unbeknownst to you, someone has locked the door from the outside. So you cannot leave. And the question that Locke asks there is, well, are you free at that point as a human being? Do you have autonomy? You feel free. Of course, you're doing whatever you want to do. You don't even realize the door is closed, but are you truly or genuinely free? And here, what John Locke says is no. He says you are not free because you do not have the ability to change what you want. You're getting what you want in this great room, but you cannot change. You cannot change what you want. And in a certain sense, you cannot change who you are because you keep repeating these very pleasurable actions in this room.

Okay. So that's 300 years ago. What does this have to do with recommendation algorithms? Well, what we want to say is it seems like something like Netflix is doing almost the same thing. It's recommending movies, films, documentaries that satisfy us, that we want to watch, that we enjoy. But the way it is doing that is by generating a list of recommendations from our past interests, from the movies we have given five stars to, from the documentaries we have watched all the way through. So the result is, in a certain sense, we are trapped in the interests and preferences that we have already established. We are happy in front of our TV watching Netflix. We're getting what we want, but there's no way to change what we want because everything we are offered resembles the things we have seen in the past. So if you watched Indiana Jones and you loved it, then the next thing you get is Mission Impossible or something.

Now, as it happens, something like this was referenced in the New York Times just this weekend. It came up in this article in the magazine. Specifically, there's this experience that is referenced, which has actually happened to me and maybe happened to you guys too, where you go to an Airbnb and you turn on the TV and it's someone else's Netflix. And you look at these films that are suggested and you think, where did that come from? I did not even know those things existed. There are all these programs that are completely foreign to you. So I think that what that shows is how much of a silo there is in Netflix, how you can be—trapped is not the right word because you're happy—how you can be satisfied by this range of movies and films that are completely different than anyone else's. And that satisfaction keeps you in the silo because you keep getting what you want. So again, I think what we're replicating here with Netflix is the John Locke experiment.

Now very quickly, how does Netflix actually do this? What are their tools? Here are two things that Netflix really does. One is, and Spotify does this too, one is if you like a movie like Indiana Jones, they try to figure out what other movies are like that movie, maybe similar actors or a similar plot or something, and then they recommend you that, right? So if you liked one thing, you get something else which is just like what you've already enjoyed.

The other strategy they use, collaborative filtering, is they try to find a kind of twin for you out in the world. So if Marianne likes Indiana Jones and Mission Impossible, and I like Mission Impossible and Indiana Jones, and then she watches another Mission Impossible movie and enjoys it, then that movie will be recommended to me. Again, it's the same thing, whether it's my own past and my own interests or my interests as they're reflected off someone else, I'm caught in the preferences that I have already established.

So as a quick note, it's true that Netflix doesn't matter. I mean, who cares if you're imprisoned by Netflix? I mean, who cares if you don't have human freedom when you're on your sofa drinking a beer, watching a movie? Doesn't matter. But you can move up levels, right, from Netflix to Twitter. If you get to LinkedIn, then you're talking about your professional life. If you get to Tinder, then you're talking about your personal life, who you will marry. If you're searching—I was speaking with some of our panelists about where we live—one way you end up living somewhere is the real estate engine recommends apartments for you. Same thing. Maybe in the future, when we are choosing the genetic makeup of our children, we will get recommendations: do you want child A or B, right, suited to you and your own past, your own established preferences. So I think it's true of—well, of course it's true that Netflix is not a profound subject, but I believe that the dilemmas that are at play in Netflix recommendations are significant now and will be very significant in the future, significant for our human freedom.

Okay, so we're getting toward the end of the introduction, the last two points. The first one is, well, what can we do? If, if—and you don't have to accept this, maybe you don't believe John Locke's definition of freedom—but if you accept that that freedom means the ability to change who you are, then for recommendations what we want is not something we already want, because that just keeps us who we are, and not something we don't want, because that's not a recommendation, that's like random. We want a film, a documentary, something that we could want. Not something we want now, but something we could want, something that would correspond with different kinds of interests that we don't yet have, something that would indicate a sort of enjoyment that we have not yet experienced. That's what we would look for if we want to have human freedom in these recommendation algorithms, right? We want suggestions that help us change the kinds of things we want, that help us develop new preferences.

And then penultimate point, this is a really hard problem. Mary Lou will talk about this perhaps in a bit. We need to find things—in order to get this kind of recommendation, we need to find things that are unfamiliar, right? Something that's not like what we've seen before, but it also has to be engaging, relevant. If you stop for a second and think about that, that's extraordinarily difficult. How exactly do you find that kind of film to recommend for a user? Something that they've never seen before or thought of before. Kind of like when you walk into the house with someone else's Airbnb set, excuse me, someone else's Netflix set. So it's completely foreign, but it's also engaging. That's the difficulty.

Good. So that is the introduction. This is the problem. The problem is that recommendation algorithms provide convenience and satisfaction, but they diminish freedom by limiting our future prospects to our established preferences. We are imprisoned by who we are. So that is the introduction.