**James Brusseau:** Sure. Thanks, everyone, for having me and thank you, Yvonne, for that nice introduction to where we have been in AI ethics. It's a short history, it's 10 years or so, but quite a bit has happened in that time. So for my presentation, I'm going to be working within that general framework of AI ethics discussed in the previous presentation, but I am going to talk about a specific subject, which I believe is going to build increasing importance as time goes forward.

And this is the subject of recommendation algorithms, which, of course, we are all familiar with through Netflix or Spotify or some of these streaming platforms and music and movies that we all watch. This is just a good example here. I think all of us have had this experience where we come to the end of a movie or the last episode of a series, and then Netflix already has queued up. As you see, this is literally a picture of my TV screen. It already has queued up the next movie or the next series to watch just as the last one is ending. So if the system works perfectly for them, I will never leave my sofa. I will get done with one program and just start watching another. So that's the kind of power that recommendation algorithms are starting to exhibit in our society. And then I want to say that it's true that that question of movies that are being recommended to us by the forces of the Netflix algorithms I assume that that's a fairly trivial experience it really doesn't matter too much whether these movies are good or bad or whether we enjoy or don't enjoy these movies however what has been happening over the last several years is that these same recommendation strategies have become more pervasive in more significant aspects of our lives. So it starts with something like Netflix. But, of course, these are also the same recommendations that work to guide people on social media, that alert them to posts that they might see on Ant or on Facebook and so on. And because these recommendations do such a good job of channeling people toward interests that they have already established. We end up in a situation where we have small groups emerging of tribes of groups of people who are extremely interested in a certain kind of music or a certain kind of cuisine or a certain kind of politics and so on. So we're starting to see in social media these filter bubbles and the tribalism that is a result of recommendation algorithms.

And then moving up, something like LinkedIn, as people increasingly find employment through online platforms, it will be increasingly true that the professional life we live, the jobs we have, will be a reflection of the recommendation algorithms which are channeling us one or another opportunity. So we're starting to build up now. It started as just movies, then it was social media and now it's getting fairly serious right what job will you have well that's a recommendation algorithm question increasingly

And then I think the maximum level it seems to me would be the dating applications where it's not now your professional life but your personal life the person you marry the person with whom you form a family the person and with whom you have children and share a life and so on. Decisions on that level, which are, of course, monumental for each of us, decisions on that level too are being increasingly directed by recommendation algorithms because people are going online to meet others. And, of course, there are so many, I'm not looking for a woman, I live in New York City. You can imagine how many women there are on Hinge or Tinder or Bumble or whatever. It would be impossible for me to have any luck finding anyone if I just got this fire hose, this waterfall of names and pictures. What the dating sites would need to do, if I were on them, is filter the selection for me through a small subset of the possible women. So I think that at that point, it becomes difficult to deny that recommendation algorithms are fundamentally shaping the way we live our lives.

And then if you look at this slide here, this is just very recent, just a couple of years old. The red line shows the percentage of couples in the United States who met online. And you can see it is arcing upward almost, it feels like exponentially. More than 50 percent of the couples today in the United States at least who are together met online I suppose that would be those who came together in the last year or two years because obviously many couples have been together before internet was even invented but within the last few years more than half of the couples that you see that are newly formed affirmed are the result of recommendation algorithms. So that's the first thing that I want to say in my presentation. I think that we frequently think of recommenders as kind of helping through movies or songs, but they're not. They're significant things for the most important part of our lives. And so for that reason, and I will even be stronger as long as we're in a small group of friends. It's my sense that this might be not just insignificant, but the significant, the most significant aspect of artificial intelligence in terms of how it affects human lives. The human condition will change most fundamentally, perhaps it seems to me will change most fundamentally because of the role that recommendation algorithms are playing across our life

So if that's true, then we can go to the second part of the presentation, and even if it's not true you'll still go to the second part of the presentation, when we talk about some of the problems that we're beginning to face as we use recommenders in more aspects of our life. The fundamental, I believe, is that while recommenders are very, very convenient...

Going back to this picture, it's really nice when you don't even have to look for a movie. It just gives you what you want. While recommenders are very convenient, they also trap us in our established interests. I think that many of us, I know I certainly, have had this experience. I feel like when I'm looking at Netflix, the same movies or the same kind of movie keeps getting sent back to me. It's very hard to escape that kind of filter bubble Netflix has set up for James Brusseau. It becomes very difficult for me to find new kinds of movies that are separate or distinct or diverse from the ones that I have seen before, and to a certain extent, maybe I don't want to find them because I'm so comfortable on my sofa, but that doesn't change the fact that for movies, we keep getting back what we've already seen. And then I think that the logic of recommenders is identical across these different domains. So what's true in movies will also be true in social media platforms. We already see that tribalism. It will also be true in LinkedIn. It will also be true in Tinder.

So the problem in Netflix, I think, is across the range of human experience. That is, it does provide convenience, but it traps us in the preferences we have already established. And then if that's the fact, I think we can convert that onto the ethical side and say that, in a sense, that's for individuals, for me, for you. That is a freedom. We are denied the capacity to find and develop new interests, discover new preferences find new ways of living and then for society I think that that's a solidarity problem for the reasons I've already stated because we tend to gather each of us each of our groups in that small little area and then the third area of interest which is a little different from what I've talked about till now the third area is just money economics businesses. This is also a problem for platforms like Netflix, because they want to keep users on. But people get bored. More and more, if you just Google bored by Netflix, you will find many, many comments of people saying, look, they're just tired of seeing the same movies over and over again. So that's why they quit their Netflix subscription and join another streaming service. So there's a whole stream of problems that are tied to the use of recommendation algorithms, from ethical to social to just simply economic. And I think all of those problems trace back to the same original fact, right? This idea of the snake eating its tail, that once your interests have been established, they just keep giving you that over and over again. And then I have as my, this is a quick footnote, I have as my example here, the movie Groundhog Day. I'm not sure if anyone has seen that movie, but what happens in that movie is the person lives the day, which is a day, it's a holiday in the United States, a small holiday.

He lives that day, goes to sleep and he wakes up the next morning and it's the same day again. He lives the same day again, goes to sleep, wakes up, same day again. I think that kind of captures lectures a little bit of the human experience of the problem with the recommenders. So I've said two things now. I've said, I think recommenders are very important. I've said why I think recommenders are a problem because they trap us in our established preferences. And now I want to say, and this is sort of the work of the paper, and I want to say, well, what's the solution to this. What can we do? How can philosophers and ethicists like me work with technologists, with AI engineers to solve these problems? And I want to say that it seems to me that the answer is to re-engineer our recommendation algorithms to catalyze new curiosities, new interests, new preferences. Instead of just giving us back the same thing over and over again, and find a way to make recommenders that open doors to us to new and unfamiliar kinds of experiences. If we could do that, it seems to me that we might take a step at least toward resolving this ethical, social, and even the economic problem that the recommenders currently face.

So the question is, okay, sounds good. We're going to make a new kind of recommendation, which will not just give us the same thing over and over again, but will open the door to new kinds of experiences, interests, and preferences.

How do we do that? Well, I mean, to explain how to do that, let's first look at what Netflix does to not do that. What Netflix does to not do that is they use something called collaborative filtering to recommend a movie to you, or content-based filtering. Content-based filtering is if you liked the Indiana Jones movies, then they look for other movies that are very similar to that one, and maybe have the same actor, maybe they're an action-adventure. So if you like the Indiana Jones movies, then they recommend the Star Wars movies. Both have Harrison Ford. Both have action-adventure and so on. So that's content-based filtering.

But more important, there's this idea of collaborative filtering. What Netflix does is it looks for someone who's a lot like James Brusseau, same age, maybe lives in New York, and likes the same movies. So let's say I like Indiana Jones. I like Star Wars. I like the Fugitives movie. I'm sort of dating myself here but those are the movies that come to my mind and then they find another person who lives near me in New York city who also likes those three movies and likes the movie die hard right not an election to venture and when you look at my Netflix viewing I have not seen die hard so what does Netflix do now they say like the Brusseau and this other person are similar users they have similar likes in movies there's this one movie that the one person has seen that the other hasn't, and the other person has liked. So let's recommend it over to Brusseau. So you can see how this funneling effect works, how people keep getting narrowly swirled down into specific kinds of interests, and how it becomes increasingly difficult to escape because you keep getting recommendations that are like the ones you've seen or that are like ones that are enjoyed by people who are like you. So that's the logic right there that I think needs to be broken.

And we have limited time, so we're not going to go too far into this, but I'm just going to give you one example of how that logic could be broken. And so that would be instead of using the collaborative filtering, which Netflix does, finding two people who are very similar and recommending back and forth between them, And maybe you could do something like antagonistic filtering, by which I mean that you take James Brusseau, 50-some-year-old male in New York City, and maybe an 18-year-old young woman in Tokyo, right?

Two people who are really very different. And you start there. You see, instead of trying to find people who are similar, let's find people who are really different. They speak a different language. Maybe they're French gender. Maybe they live in different countries, different ages, and so on, different professions. And then once you've found people who are very different, you look for maybe there are some who have just a few similarities. Maybe there's one or two movies they like in common, for example. Something like that might offer the opportunity to recommend back and forth. So maybe a movie I like that this young woman in Tokyo has not seen can be recommended to her and vice versa. That might allow us to do two things. It might allow us, like me, to be tuned into new kinds of movies. Maybe a kind of movie that I never would have thought existed. Documentaries about Japanese farming, I'm just making this up. Documentaries about Japanese farming or something, something that I never would have come to on my own. And the same thing might be true for her. Maybe there's some movie that I have enjoyed, which is very sort of American for middle-aged guys, but she has not seen anything like it and it gets recommended to her. So through this kind of logic, instead of thinking everything has to be similar all the time, changing that for the logic where we start with difference and then try to find some minimal similarities and then recommend back and forth. Perhaps recommendation algorithms can be changed or can be adjusted to provide, to open new doors or create curiosity as opposed to just giving us convenience and providing the same thing over and over again.

Now, I'm a philosopher, so I will just quickly, I know that we're coming toward the end of my time, but I will just quickly say that there's sort of a theoretical way to understand this way of thinking. And I'll just step back to there I talked about antagonistic filtering, and there are many other ways of imagining how to do that. I just gave one example, but we have limited time, so I will move forward to a kind of a conceptual theoretical idea. You can just see here the difference between this is a Caravaggio, it's the dark woman, of course, and then Van Gogh and Picasso. You can see the difference in the relation between those two things. So the movement from the Caravaggio to the Van Gogh, it's a movement of similarity. Things are the same. And then it goes from women to women, and then you say, well, once we have these similarities, maybe there are some little differences in the style of painting and the figurativeness of the face and so on, but they're fundamentally the same.

Contrast that with the comparison between the Caravaggio and the Picasso. The Caravaggio and the Picasso, I mean, it seems to me that we start out with things has been fundamentally different. And it's only after working through that difference that we say, look, I can see now that the Caravaggio and the Picasso, they're both women. If you show the Picasso to an eight-year-old, they'll think, well, what is that? They don't even recognize that's a human being for a while, right? It really takes some work. And that's the kind of work that I am proposing that algorithms be engineered to do. They bring together these differences so that people can see, have experiences of a painted woman, let's say, which are fundamentally different. But you have to bring them together in a way which provokes curiosity, which makes the one person who likes Caravaggio interested in Picasso and vice versa. But as I say, what I'm trying to show here finally, theoretically, is just this difference between thinking through the logic of similarity, which Netflix does, and thinking through the logic of difference, which I am proposing. So we can move to the conclusions or the conclusion.

What I want to say is that recommendation algorithms can do more than give us what we want. They can do more than be convenient and just keep feeding us back the same things we have already seen. They can also help us to want new things. They can open doors to new kinds of experiences. And then I want to say that on the ethical level there's a real positive there, because that catalyzes human freedom. And then I can just add too quickly on the social level, to an extent this might reduce or diminish. They've attacked against tribalism, people being divided up into their little groups. And then on the economic level, we'll just say quickly, I mean, perhaps this is good for Netflix too.

Perhaps people will stay on the site longer if they get recommendations about which they are curious and about which open up new possibilities. We will have to see. So that is the talk I have to present. It's called the Freedom Recommendation, by which I mean the recommendation that enhances human freedom. I put up a larger project called the Curiosity Engine, which is about making this theoretical philosophical idea about differences and trying to translate it into the engineering that is required to create these kinds of recommendations. So thank you very much for bearing your time with us today.

**Other speaker (context):** Thanks so much, Professor Brusseau, for this wonderful and insightful presentation on the Freedom Recommendation. I will open it to the room if anyone has any questions related to the topic or have been out of topic, but it's got some linkage to the recommender system.

**Other speaker (context):** Okay, Yared, I'll let you go. Thank you so much, Professor Brusseau. This is a wonderful presentation. Excuse me if I'm asking dumb questions. I come from not academia. Demi, I came from industry. **Other speaker (context):** Mark Benioff from Salesforce, Facebook is the new cigarette. **Other speaker (context):** As you can see with social media platforms, could be TikTok, it could be Instagram, especially people with younger age, people staying on their phone for quite some time. As you know, the term couch potato is termed after people who loves to stay at home and keep on watching movie. From application side, there is definitely a lot of outcome and benefit **Other speaker (context):** When it comes to getting job on LinkedIn or getting shopping recommendation to compare product that will help us. But from society perspective, especially young generation are staying on their phone on an alarming stage, what would be the implication overall as a well-being of society in the long run? Because the reason why I'm asking this is that we learned a lot from cigarettes back in the days. And attorney generals in the U.S. charged in cigarette companies back in 1999 quite some amount of money and they established a true initiative to educate against tobacco. But what has been done in the research area to educate on recommendation agents affecting humanity in general in a negative way?

**James Brusseau:** Excellent question. I believe it's my sense that most of the research in this area, in terms of sort of dangers of recommendation algorithms on a social level, are focused on this idea of filter bubbles and polarization and tribalism in society. And so what they mean by polarization is that everyone who is, let's just, I'm from the United States, so we can just use, it's easy for me to use my political system as an example. So we had in just a previous election, perhaps, you know, just recently, we had all kinds of voters who loved Donald Trump and knew only good things about Donald Trump and knew only bad things about his opponent, who was a woman named Kamala Harris. And then she also had all her supporters who thought she was wonderful and knew only good things about her and thought only, knew only bad things about her opponent right so people end up just getting information which is only good about what they like and only negative about things they already don't like so that's the I think that most of the research that I've seen devoted to the area that you are asking about has been on the subject of counteracting this reality and there's there are two main sort of vectors that have been pursued, and I'm proposing a third. The two vectors that are being pursued are one, and of course people say, well, everyone stop looking at your telephones. Go out and meet people who are different than you are because then you will learn about how different people see the world and so on. And that certainly works if it works. I'm not sure how realistic it is to ask people to get off of their screens.

What I'm proposing, I'll come to in a moment, what I'm proposing works within the idea that people kind of are going to stay on their screens and we should get accustomed to living that way.

One response to the filter bubble problem is get people off their screens. The second response the people are working on is, well, let's cut out misinformation. So at least the people who like one candidate don't get lies about the other candidate and vice versa. And perhaps that will diminish to some extent this tribalism.

What I'm proposing is a third answer which is different from those two. And as I say, that third answer is to find ways to use the same technology that causes the problem is recommendation algorithms use that same technology to solve. So it's not saying my position is not that recommendations are, algorithms are bad, even though they're having bad effects now. It's instead that we should, they're imperfect and we should make them better and we can make them better. And this is how with antagonistic filtering. So I think that's the way I myself would promote responding to this important worry that you have surfaced for us about the filter bubble communities on the internet. Did I answer the question? Did I explain it clearly?

**Other speaker (context):** Yep, same thinking. Any second question? So one thing I did not hear in a presentation is commercial interest, I argue the only reason those recommender systems exist is to make money for the corporations. **Speaker uncertain:** The Netflix, the Amazon, the dating sites, the only reason is actually to make money. So we're in a framework that's commercial interest. If you look at old Netflix, the times they were sending DVDs, those recommenders were actually pretty good. I liked them. Now all they do is send me Netflix productions and the stuff for which they just have a short-term license and they want to show it to as many people as possible, right? So those recommender systems clearly shifted from what you want to see to what they really want you to see, right? So where does commercial interest is in your framework?

**James Brusseau:** Right. Excellent question. It's something I'm dealing with right now. First, as you point out, one difficulty is that there are so many competing interests that Netflix has to weigh. They want their consumers to be happy, but they also want to use their own material and so on. Making their consumers happy might not maximize profits, in other words. So it's very complicated to get into that question.

However, let me answer along at least one narrow avenue. And that is, I believe that one commercial problem Netflix has, and I agree, of course, that they treat you and I as commodities. That's what we are. We're just hours for money for them. That's true. Nevertheless, the problem they face is that people do leave their platform because the recommendations Netflix is providing too frequently just reinforce and repeat and recycle the kinds of interests in films we have already seen. So I think, at least I'm proposing, that the kind of antagonistic filtering that I think helps individuals increase freedom, helps society decrease tribalism. I believe will also or could also help Netflix make money. And that's kind of important because in the real world, we're going to talk about the ethics in real companies. It helps if we can tell them, look, doing good also equals doing well.

If you do this, you can increase human freedom plus make money. And the way I believe we could help Netflix make money or expand use of their catalog is by helping them open up their users to new kinds of interests and so on. And then as I say that, I think that has positive social effects too. So in conclusion, I think that the financial incentives that you're talking about to Netflix, which you are obviously 100% correct, right? That's the cause of the recommendations. And that's the reason why we keep getting the same stuff over and over again. And also the stuff that Netflix makes. I think those same incentives can be twisted around to operate positively, if we can show them why. And they can, as I say, increase human freedom and decrease tribalism in society. So does that begin to answer the question which could have gone in so many different directions?

Sure. Thank you. Okay

**Other speaker (context):** Okay, that's awesome. I have one more question from online that I'll be reading it for you. This is from Andrew. One of the ways this algorithm's work is by maximizing attention and engaging with the goal of using that data to sell advertising. If we change the algorithm the way you suggest, we might expect more uneven engagement. How did that change the economic model tech companies might use to remain in business?

**Other speaker (context):** Right

**James Brusseau:** Right. Well, I think that's an excellent question. I think that continues out of the response I was just given. It seems as though if you look at the literature, it seems as though one problem that Netflix faces and any streaming platform in Spotify or Tinder or LinkedIn, et cetera, any platform that uses recommendations, nations. One problem they face is that they lose users, engagement declines because people have something which in the literature is called algorithm fatigue or user fatigue. What that means is in the academic literature, these people just get tired of seeing the same thing over and again before they leave. So it seems to me that, again, by promoting new kinds of interests, by opening curiosity about new kinds of films that users have not seen, that in the medium and long term, Netflix profits would actually increase, not decrease. So that would be my response, that user engagement, though it might suffer in the near term, because, of course, when people see a new material, they may be turned off by that. They may leave. They may turn off the television. But in the medium and long term, the reward, I believe, is quite high for these companies if they can show users whole new vistas of possible entertainment or music and so on. I hope that begins to answer that good question.

**Other speaker (context):** Awesome. I have one last question from our online participants. This is from Nina. How do recommendation algorithms affect price of choice in general at luncheon time? I know it's a little bit off topic, but if you have any thought on that.

**James Brusseau:** Oh, no, that's right on time. That's right on top. That's a great question. Right. So first thing, it's very hard question to answer. It depends upon your context. On the one side, an argument that I make is it really massively decreases human freedom on the following level. And it's true that our preferences and interests are being satisfied when LinkedIn makes, excuse me, Netflix makes a good recommendation. But it's also true that our ability to change our interests and our preferences, that is being neutralized or even numbed, right, by this comfort and convenience we have because we're getting a film that we already like. It's a very strange situation. Human freedom is being numbed or diminished by human freedom itself. It's because we're getting what we want that we cannot want new things. Because freedom is being successful, we are losing our personal freedom. So it's a very complex area of discussion. But then again, as I say, in terms of the freedom understood as the ability to develop new interests, if that's your definition of freedom, and that's sort of my definition. It's John Locke's definition also from the 17th century. If that's your definition of freedom, then I would say that the recommendation algorithms, as they are structured now, significantly decrease your ability to develop new interests and therefore your freedom, even though while ironically, they're satisfying your freedom. They're giving you what you want. It's a strange territory. I hope that to some extent was being asked.

**Other speaker (context):** Thanks so much, Professor Brusseau. This was a very enticing and engaging conversation on the recommender system and our freedom in the age of AI. We really appreciate that you took your time to show up and discuss your topics here. Thanks so much. Okay, I will go to the next presenter. That's Dr. Uday. I tried so hard to get him into this spot but we finally made it, so I'll let him introduce himself because he's about it. Thank you. Hi, my name is...

**Other speaker (context):** Good morning dear presenters, guests, academics and especially doctoral students. Welcome to the **Speaker uncertain:** The first event of the Alma Mater AI Center, the conference advancing the frontiers of AI research.