Yesterday, the Attorney General held a workshop on Section 230 of the Communications Decency Act. The question was whether the law can be improved. Section 230 does need work, though there's plenty of room for debate about exactly how to fix it. These are my mostly tentative and entirely personal thoughts on the question the Attorney General has asked.
Section 230 gives digital platforms two immunities – one for publishing users' speech and one for censoring users' speech. the second is the bigger problem.
- Immunity for what users say and do online
When section 230 was adopted, the impossibility of AOL, say, monitoring its users in a wholly effective way was obvious. It couldn't afford to hire tens of thousands of humans to police what was said in its chatrooms, and the easy digital connection it offered was so magical that no one wanted it to be saddled with such costs. Section 230 was an easy sell.
A lot has changed since 1996. Facebook and other have in fact already hired tens of thousands of humans to police what is said on their platforms. Combined with artificial intelligence, content fingerprinting, and more, these monitors work with considerable success to stamp out certain kinds of speech. And although none of these efforts are foolproof, preventing the worst online abuses has become part of what we expect from social media. The sweeping immunity Congress granted in Section 230 is as dated as the Macarena, another hit from 1996 whose appeal seems inexplicable today. Today, jurisdictions as similar to ours as the United Kingdom and the European Union have abandoned such broad grants of immunity, making it clear that they will severely punish any platform that fails to censor its users promptly.
That doesn't mean the US should follow the same path. We don't need a special, harsher form of liability for big tech companies. But why are we still giving them a blanket immunity from ordinary tort liability for the acts of third parties? In particular, why should they be immune from liability for utterly predictable criminal use of warrant-proof encryption? I've written on this recently and won't repeat what I said there, except to make one fundamental point.
Immunity from tort liability is a subsidy, one we often give to nascent industries that capture the nation's imagination. But once they've grown big, and the harm they can cause has grown as well, that immunity has to be justified anew. In the case of warrant-proof encryption, the justifications are thin. Section 230 allows tech companies to capture all the profits to be made from encrypting their services while exempting them from the costs they are imposing on underfunded police forces and victims of crime.
That is not how our tort law usually works. Usually, courts impose liability on the party that is in the best position to minimize the harm a new product can cause. Here, that's the company that designs and markets an encryption system with predictable impact on victims of crime. Many believe that the security value of unbreakable encryption outweighs the cost to crime victims and law enforcement. Maybe so. But why leave the weighing of those costs to the blunt force and posturing of political debate? Why not decentralize and privatize that debate by putting the costs of encryption on the same company that is reaping its benefits? If the benefits outweigh the costs, the company can use its profits to insure itself and the victims of crime against the costs. Or it can seek creative technical solutions that maximize security without protecting criminals – solutions that will never emerge from a political debate. Either way it's a private decision with few externalities, and the company that does the best job will end up with the most net revenue. That's the way tort law usually works, and it's hard to see why we shouldn't take the same tack for encryption.
2. Immunity for censoring users Detecting bias.
The harder and more urgent Section 230 problem is what to do about Silicon Valley's newfound enthusiasm for censoring users whose views it disapproves of. I confess to being a conservative, whatever that means these days, and I have little doubt that social media content mediation rules are biased against conservative speech. This is hard to prove, of course, in part because social media has a host of ways to disadvantage speakers who are unpopular in the Valley. Their posts can be quarantined, so that only the speaker and a few persistent followers ever see them but none knows of that distribution has been suppressed. Or they can be demonetized, so that Valley-unpopular speakers, even those with large followings, cannot use ad funding to expand their reach. Or facially neutral rules, such as prohibitions on doxing or encouraging harassment, are applied with maximum force only to the unpopular. Combined with the utterly opaque talk-to-the-bot mechanisms for appeal that the Valley has embraced, these tools allow even one or two low-level but highly motivated content moderators to sabotage their target's speech.
Artificial intelligence won't solve this problem. It is likely to make it worse. AI is famous for imitating the biases of the decisionmakers it learns from – and for then being conveniently incapable of explaining how it arrived at its own decisions. No conservative should have much faith in a machine that learns its content moderation lessons from current practice in Silicon Valley.
Foreign government interference. European governments, unbound by the first amendment, have not been shy about telling Silicon Valley to suppress speech it dislikes, which include true facts about people who claim a right to be forgotten, or charges that a politician belongs to a fascist party, or what it calls hate speech. Indeed, much of the Valley has already surrendered, agreeing to use their terms of service to enforce Europe's sweeping view of hate speech--under which the President's tweets and the Attorney General's speeches could probably be banned today.
Europe is not alone in its determination to limit what Americans can say and read. Baidu has argued successfully that it has a first amendment right to return nothing but sunny tourist pictures when Americans searched for "Tiananmen Square June 1989." Jian Zhang v. Baidu.Com Inc., 10 F. Supp. 3d 433 (S.D.N.Y. 2014). Today, any government but ours is free to order a US company to suppress the speech of Americans the government doesn't like.
In the long run it is dangerous for American democracy to give highly influential social media firms a blanket immunity when they bow to foreign government pressure and suppress the speech of Americans. We need to armor ourselves against such tactics, not facilitate them.
Regulation deserves another look. This isn't the first time we've faced a disruptive new technology that changed the way Americans talked to each other. The rise of broadcasting a hundred years ago was at least at transformational, and as threatening to the political order, as social media today. It played a big role in the success of Hitler and Mussolini, not to mention FDR and Father Coughlin.
American politicians worried that radio and television owners could sway popular opinion in unpredictable or irresponsible ways. They responded with a remarkable barrage of new regulation – all designed to ensure that wealthy owners of the disruptive technology did not use it to unduly distort the national dialogue. Broadcasters were required to get government licenses, not once but over and over again. Foreign interests were denied the right to own stations or networks. A "fairness" doctrine required that broadcasters present issues in an honest, equitable, and balanced way. Opposing candidates for office had to be given equal air time, and political ads could to be aired at the lowest commercial rate. Certain words (at least seven) could not be said on the radio.
This entire edifice of regulation has acquired a disreputable air in elite circles, and some of it has been repealed. Frankly, though, it don't look so bad compared to having a billionaire tech bro (or his underpaid contract workers) decide that carpenters communicating with friends in Sioux Falls are forbidden to "deadname" Chelsea Manning or to complain about Congress's failure to subpoena Eric Ciaramella.
The sweeping broadcast regulatory regime that reached its peak in the 1950s was designed to prevent a few rich people from using technology to seize control of the national conversation, and it worked. The regulatory elements all pretty much passed constitutional muster, and the worst that can be said about them today is that they made public discourse mushy and bland because broadcasters were cautious about contradicting views held by a substantial part of the American public.
Viewed from 2020, that doesn't sound half bad. We might be better off, and less divided, if social media platforms were more cautious today about suppressing views held by a substantial part of the American public.
Whether all these rules would survive contemporary first amendment review is hard to know. But government action to protect the speech of the many from the censorship of the privileged deserves, and gets, more leeway from the courts than the free speech absolutists would have you believe. See, e.g., Bartnicki v. Vopper, 532 U.S. 514 (2001).
That said, regulation has many risks, not least the risk of abuse. Each political party in our divided country ought to ask what the other party would do if given even more power over what can be said on line. It's a reason to look elsewhere for solutions.
Network effects and competitive dominance. Maybe we wouldn't need a lot of regulation to protect minority views if there were more competition in social media – if those who don't like a particular platform's censorship rules could go elsewhere to express their views.
In practice, they can't. YouTube dominates video platforms, Facebook dominates social platforms, Amazon dominates online book sales, etc. Thanks to network effects, if you want to spread your views by book, by video, or by social media post, you have to use their platforms and live with their censorship regimes.
It's hard to say without investigation whether these platforms have violated antitrust laws in acquiring their dominance or in exercising it. But the effect of that dominance on what Americans can say to each other, and thus on political outcomes, must be part of any antitrust review of their impact. Antitrust enforcement often turns on whether a competitive practice causes consumer harm, and suppression of consumer speech has not usually been seen as such a harm. It should be. Suppression of speech it dislikes may well be one way Silicon Valley takes monopoly profits in something other than cash. If so, there could hardly be a higher priority for antitrust enforcement because such a use of monopoly strikes at the heart of American free speech values.
One word of caution: Breaking up dominant platforms in the hope of spurring a competition of ideas won't work if the result is to turn the market over to Chinese companies that already have a similar scale – and even less interest in fostering robust debate online. If we're going to spur competition in social media, we need to make sure we aren't trading Silicon Valley censorship for the Chinese brand.
Transparency. Transparency is everyone's favorite first step for addressing the reality and the perception of bias in content moderation. Surely if the rules were clearer, if the bans and demonetizations could be challenged, if inconsistencies could be forced into the light and corrected, we'd all be less angry and suspicious and the companies would behave more fairly. I tend to agree with that sentiment, but we shouldn't kid ourselves. If the rules are made public, if the procedures are made more open – hell, if the platforms just decide to have people answer complaints instead of leaving that to Python scripts -- the cost will be enormous.
And not just in money. All of the rules, all of the procedures, can be gamed, and more effectively the more transparent they are. Speakers with bad intent will go to the very edge of the rules; they will try to swamp the procedures. And ideologues among the content moderators will still have room to seize on technicalities to nuke unpopular speakers. Transparency may well be a good idea, but its flaws are going to be painful to behold if that's the direction our effort to discipline Section 230 takes.
3. What is to be done?
So I don't have much certainty to offer. But if I were dealing with the Section 230 speech suppression immunity today, I'd start with something like the following:
First, treat speech suppression as an antitrust problem, asking what can be done to create more competition, especially ideological and speech competition, among social media platforms. Maybe breakups would work, although network effects are remarkably resilient. Maybe there are ways antitrust law can be used to regulate monopolistic suppression of speech. In that regard, the most promising measures probably are requiring further transparency and procedural fairness from the speech suppression machinery, perhaps backed up by governmental subpoenas to investigate speech suppression accusations.
Second, surely everyone can agree that foreign governments and billionaires shouldn't play a role in deciding what Americans can say to each other. We need to bar foreign ownership of social media platforms that are capable of playing a large role in our political dialogue. We should also use the Foreign Agent Registration Act or something like it to require that speech driven by foreign governments be prominently identified as such. And we should sanction the nations that try to do that.
And finally, here's a no-brainer. If nothing else, it's clear that Section 230 is one of the most controversial laws on the books. It is unlikely to go another five years without being substantially amended. So why in God's name are we writing the substance of Section 230 into free trade deals – notably the USMCA? Adding Section 230 to a free trade treaty makes the law a kind of a low-rent constitutional amendment, since if we want to change it in future, organized tech lobbies and our trading partners will claim that we're violating international law. Why would we do this to ourselves? It's surely time for this administration to take Section 230 out of its standard free-trade negotiating package.
Note: I have many friends, colleagues, and clients who will disagree with much of what I say here. Don't blame them. These are my views, not those of my clients, my law firm, or anyone else.