“I agree in principle that there is liability, but I don’t think we have found the right set of terms to describe it,” said Jonathan Stray, a visiting scholar at the Berkeley Human Center. processes we care about. – AI compatible research recommendation algorithms. “What is amplification, what is innovation, what is personalization, what is recommendation?”
For example, New Jersey Democrat Frank Pallone’s Malicious Algorithm Justice Act would revoke immunity when a platform “knew or ought to have known” that it was making a “proposal.” personalized” for the user. But what is considered personalized? According to the bill, it uses “information that is specific to an individual” to enhance the prominence of certain documents from others. That’s not a bad definition. On the face of it, though, it seems to say that any platform that doesn’t show everyone the exact same things loses Section 230 protections. Even showing anyone’s posts by the people they follow are also believed to be based on information that is specific to that person.
Malinowski’s bill, The Act to Protect Americans from Dangerous Algorithms. rating, ordering, promoting, recommending, amplifying or similar alters the distribution or display of information.” It does, however, contain exceptions for algorithms that are “clear, understandable and transparent to reasonable users” and lists several examples that fit the bill, including a reverse chronological feed and ratings by popularity or user reviews.
There’s a lot of meaning to that. One problem with engagement-based algorithms is their ambiguity: Users have little insight into how their personal data is being used to target them with content. with which the platform predicts they will interact. But Stray points out that distinguishing between good and bad algorithms isn’t so easy. For example, rating by user reviews or voting up/vote down is difficult by itself. You wouldn’t want a post with a single vote or five-star review to get to the top of the list. Stray explains that a standard way to get around that is to calculate the statistical margin of error for a given piece of content and rank it by the bottom part of the distribution. That technique — which Stray took a few minutes to explain to me — was clear and transparent? What about something as basic as a spam filter?
“It’s not clear to me whether the aim of excluding sufficiently ‘simple’ systems will in fact rule out any truly practical systems,” Stray said. “My suspicion, probably not.”
In other words, a bill stripping Section 230 of the algorithmic recommendation immunity could look like an immediate repeal, at least as far as the social media platforms are concerned. . Jeff Kosseff, author of the definitive book on Part 230, Twenty-six words that created the Internet, points out that internet companies have many legal defenses to combat, including First Amendment, even without the protections of the law. If the statute contains enough exceptions and exceptions to exceptions, those companies may decide there are easier ways to defend themselves in court.
