Madame President, regular members, fellow probationaires, and honored guests,
Section 230 of the Communications Decency Act should be amended to not protect algorithmically recommended content. Just to set terms, section 230 protects online platforms from being legally responsible for the content users post to them. Intuitively, this regulates them as public squares, like library tackboards or sidewalks, as opposed to publishers, like newspapers or radio broadcasts. Implementing my vision would retain public square protection for content not recommended via algorithm, while making platforms liable for anything they recommend via algorithm.
My first contention is that the law is outdated in its vision of online platforms. In 1996 when the law passed, the Internet was dominated by personal websites and chat forums that displayed conversations in chronological order. Basically none of these sites ran ads, they all lost money, and they were moderated by one or two enthusiasts. Seeing this, legislators offered broad protections to these small website owners that were sacrificing money and time to maintain a public good for the internet. In the intervening 30 years we have developed incredibly powerful algorithms that have driven out all these webforums and enabled multibillion dollar megacorporations to dominate the Internet, but they still get regulated as if they're just a scrappy forum. Removing 230 protections from these algorithms would push platforms to behave more like the webforums those early legislators thought were the future of public squares.
This brings me to my second contention. Modern Internet platforms are essentially publishers via their algorithmic control of their information ecosystems. In his 1976 book, The Selfish Gene, Richard Dawkins argues that ideas, which he calls memes, adapt over time to fit their environment in the same way animals do. Just as the African savannah rewards giraffes with longer necks and thus selects for genes that lengthen those necks, different information ecosystems reward certain traits in their memes. To take a recent example, the Texas Secretary of State audited their voter rolls and found a few hundred people they couldn't immediately confirm to be citizens. A few people posted about this and the most attention-grabbing version of the meme caught on and went out to more people who made their own twists. By the time it got to me it read “Secretary of State admits that 1000s of illegals voted in 2020.” This is what happens when your information ecosystem selects for being attention-grabbing rather than being true.
Now this evolution is a part of any information ecosystem but algorithmic recommendation takes it to the point of being a publisher. Since the platforms get to choose the algorithm, they get to choose exactly what traits their ecosystem rewards. Yes, technically they're content neutral, anything can get posted, but in terms of what Facebook's audience actually sees, what memes are allowed to replicate and evolve, it is exclusively ideas which have been approved by Facebook. This is the exact same as if the library tackboard took down posters the librarians didn't like, at that point, it's not a neutral public square anymore, it's a publisher endorsing ideas, and it should be regulated as one. Either change your recommendation system to be content neutral or be liable for the posts you endorse.
My third contention is that platforms have abused these protections to destroy our politics. I haven't spoken yet about what kinds of traits algorithms select for. In principle, they could select for educational memes, but because we live in a capitalist system, they reward memes that maximize profit. That means time spent on platform and driving engagement, because both of those drive ad revenue. Which traits are those? Unfortunately, the research shows it promotes memes that inspire outrage. Just to state that clearly, we have built a machine that most Americans use over two hours a day, which uses the incredible force of evolution by natural selection to make the thoughts in our heads as divisive as possible.
This has caused our politics to be rewired around outrage. We cannot solve important political problems anymore because the information ecosystem is completely dominated by ideas that have been selected for precisely because they are divisive. Just as an example, there has been more reporting this year dedicated to whether one trans congresswoman is allowed to use the ladies room, than how we plan to fund social security when it runs out of money in 2034. These are terrible incentives for politicians. You don't get votes for solving problems no one is paying attention to, so you might as well dedicate yourself to the things they do. But the things people are paying attention to are unimportant problems that divide us. Section 230 protections have allowed this to fester by protecting information ecosystems from the consequences of their memetic selection. Repealing protections would push them towards cleaning up these ecosystems.
My fourth contention is that removing section 230 protections would incentivize platforms to return to nonalgorithmic recommendations. In principle, Facebook could try to heavily increase content moderation to make sure no one ever posts anything that could get them sued, but realistically they just have too much content being posted to moderate it all, and they'd get bankrupted by lawsuits within a year. More likely is that this will push platforms to cease use of algorithmic recommendations entirely, moving us back to systems like Reddit or early YouTube, where individuals choose for themselves what they want to see, or are recommended things in a nonpersonalized manner based on objective metrics of popularity. This would be a much cleaner information ecosystem. Only seeing things posted by your friends means there are social repercussions for socially harmful memes. Nonpersonalized recommendations mean you can't develop echo chambers where totem-pole memes of the other side go unchallenged.