The Anti-Intellectualism of Social Media Design
Takahe has limited support for this type: See Original Page
I feel enlightened now that you called out the self-reinforcing nature of the algorithms. It makes sense that an RL agent solving the bandits problem would create its own bubbles out of laziness.
Maybe we can take advantage of that laziness to incept critical thinking back into social media, or at least have it eat itself.
Thanks!
I feel enlightened now that you called out the self-reinforcing nature of the algorithms. It makes sense that an RL agent solving the bandits problem would create its own bubbles out of laziness.
You’re totally right that it’s like a multi-armed bandit problem, but maybe with so many possibilities that searching is prohibitively expensive, since the space of options to search is much bigger than the rate that humans can consume content. In other ways, though, there’s a dissimilarity because the agent’s reward depends on its past choices (people watch more of what they’re recommended). It would be really interesting to know if anyone has modeled a multi-armed bandit problem with this kind of self-dependency. I bet that, in that case, the exploration behavior is pretty chaotic. @abucci this seems like something you might just know off the top of your head!
Maybe we can take advantage of that laziness to incept critical thinking back into social media, or at least have it eat itself.
If you have any ideas for how to turn social media against itself, I’d love to hear them. I worked on this post unusually long for a lot of reasons, but one of them was trying to think of a counter strategy. I came up with nothing though!
@theluddite@lemmy.ml @kersplomp@programming.dev I have a lot to say on this subject but unfortunately do not have the time right now to write out anything worth reading! I will return perhaps tomorrow.
@theluddite@lemmy.ml @kersplomp@programming.dev I didn’t fully follow the connection between the social media post and multi-armed bandit problems. Is the idea that a user has k options about what to view, chooses one, and experiences some kind of payoff from the choice? If so I’m not sure the situation is well-modeled by bandits, since the typical social media user is presented with a smallish set of options chosen for them by an algorithm, with each user choice resulting in an algorithm presenting them with another smallish set of options that might be of different size and comprise different options. That kind of situation might be better modeled as an extensive form game of user against “the algorithm” with a finite but variable set of choices for the player at each ply. It’s common in a turn-taking game for both player’s and opponent’s choice to affect the choices available to player next ply, which is why this feels like a better model to me than k-armed-bandits or the POMDP type setups usually explored in RL.
If what the algorithm does can be approximated that way (as a reward-maximizing player in a multi-ply game that chooses what category of content to show a user at each turn), then you can get partway towards understanding how it works functionally by understanding how the tradeoffs between monetization, data gathering, and maximizing surprisal (learning) in its reward function are struck. I suspect that splitting the bins/categories more and more finely sometimes makes the tradeoffs look better, which might explain why social media companies tend to do this (if you have one bin of stuff with red and blue objects, and people choose randomly from it, they’ll be less happy on average than if you have a bin of red objects and a bin of blue objects and are able to direct red-preferring and blue-preferring users to the appropriate bin better than a coin flip would).
People are not static utility maximizers, but these types of algorithms assume we are. So I think they tend to get stuck in corners both because of how they strike tradeoffs (you get manosphere content because that’s what’s most monetizable) and because people’s preferences aren’t expressed consistently in their actions and change through time (you keep getting shown scifi content because you looked at a few scifi videos in a row awhile ago when you were feeling nostalgic but you don’t usually prefer it).
That’s what I have for now. Sorry for length.
If what the algorithm does can be approximated that way (as a reward-maximizing player in a multi-ply game that chooses what category of content to show a user at each turn), then you can get partway towards understanding how it works functionally by understanding how the tradeoffs between monetization, data gathering, and maximizing surprisal (learning) in its reward function are struck. I suspect that splitting the bins/categories more and more finely sometimes makes the tradeoffs look better, which might explain why social media companies tend to do this (if you have one bin of stuff with red and blue objects, and people choose randomly from it, they’ll be less happy on average than if you have a bin of red objects and a bin of blue objects and are able to direct red-preferring and blue-preferring users to the appropriate bin better than a coin flip would).
People are not static utility maximizers, but these types of algorithms assume we are. So I think they tend to get stuck in corners both because of how they strike tradeoffs (you get manosphere content because that’s what’s most monetizable) and because people’s preferences aren’t expressed consistently in their actions and change through time (you keep getting shown scifi content because you looked at a few scifi videos in a row awhile ago when you were feeling nostalgic but you don’t usually prefer it).
That’s what I have for now. Sorry for length.
Edited 42d ago