The Conservative Bias Panic Comes for Gmail’s Spam Detection

Originally written for and cross-posted at Lawfare

Over the past decade, concern over Big Tech bias against conservative speech and speakers has dominated the narrative from the right. From repeated calls to repeal Section 230 of the Communications Decency Act to Donald Trump’s executive order attacking Twitter for censorship and subsequent petition from the National Telecommunications and Information Administration to the Federal Communications Commission (FCC) to reinterpret Section 230 to the passage of state anti-deplatforming laws in Florida and Texas that are now on their way to the Supreme Court, the memetic notion of liberal Silicon Valley tech workers secretly intervening to thwart the interests of conservatives has increasingly manifested in legal challenges to alleged discrimination by technology. The movement has now trained its sights on an unexpected target: Gmail’s spam detection.

Continue reading “The Conservative Bias Panic Comes for Gmail’s Spam Detection”

So You Want to Reform Section 230

I’ve had a number of conversations over the past several months that start with a riff on the same central question:

How should we reform Section 230?

My response is always the same:

What problem or problems are you trying to solve?

Invariably, the response will be something like:

There’s bipartisan consensus that we need to reform 230.

Blurg. In the minds of too many smart folks, the contours of Section 230 have transformed from levers and dials to achieve a wide array of policy objectives—broadly speaking, means for intermediating the flow of user-generated content across Internet platforms—to ends in themselves. What’s the problem? 230. What’s the solution? Reform 230.

This dynamic obscures what are, in my view, a fairly wide range of varying, sometimes overlapping, sometimes disparate, and almost always underspecified problems and solutions that are lurking beneath the tautological “we need to reform 230 because 230” framing. It’s not novel to point out that in developing policy, we ought to identify problems, diagnose root causes, and articulate and iterate on solutions, but we’ve almost completely lost that discipline in 230 conversations as they have begun to drown in the froth of partisan power politics.

230 reform looks like a speedrun of what happened with net neutrality, which took more than a decade to transition from a weighty academic discussion to political charlatanism. We’ve quickly glitched our way past the opening levels where we’re supposed to have some serious discussion about what we’re actually trying to accomplish to the part where Senators are literally holding hearings framed as boss fights:

Card advertising a Senate Commerce Committee Hearing as a "FREE SPEECH SHOWDOWN" labeling Sen. Ted Cruz as "THE FREE SPEECH CHAMPION" and Twitter CEO Jack Dorsey as "THE CZAR OF CENSORSHIP."

Nevertheless, I’ll try in this post to distill some themes that have come up in the course of a bunch of “reform 230” conversations. I hoped when I started that perhaps I could pull together an effective taxonomy of problems and solutions, but as I wrote, I became more convinced (and dismayed) that much of what’s being proposed is not much more than flinging things at the wall to see what sticks. There are some good ideas emerging from the primordial ooze, but Cam Kerry’s observation that we need serious thought- and consensus-building before we chaotically rend the Internet asunder is right on the money.

Continue reading “So You Want to Reform Section 230”

Section 230 as Telecom Law

Today (June 23), I’m excited to be appearing at a session of the Everything You Need to Know About Section 230 in 5 Hours series, hosted by Kate Klonick and the Yale ISP. My session, with Olivier Sylvain and Tejas Narechania, is entitled What 230 Meant for Telecom and Agencies (agenda here).

The focus of the event is the interaction between the FCC and Section 230, in light of President Trump’s recent Executive Order on Section 230 that in part contemplates the FCC issuing regulations that change the scope of Section 230’s application in various ways. Given that focus, I thought it’d be helpful to poke through the FCC’s history of interacting with 230, which is somewhat richer and more complicated than folks might expect.

Maybe a bit surprisingly, the FCC has opined about Section 230 dozens of times over the 20+ years since it was enacted, which might seem to contradict the notion in 230-world that the FCC has no business saying anything about Section 230. To the contrary, 230’s in the Communications Act, and Democratic and Republican FCC Chairmen alike have found repeated occasion to contemplate Section 230.

Nevertheless, a trip through Section 230’s history doesn’t give a lot of support for the Trump Executive Order’s notion that the FCC can interpret the scope of Section 230 to impose a goofy Rube-Goldberg-style ban on political discrimination by platforms. Instead, I see roughly two eras of Section 230 at the FCC:

  1. 230 as Broadband Policy Support. The first era, starting with 230’s enactment at the Clinton-era FCC and lasting through most of the Bush-era FCC, focused primarily on 230’s findings and policy statements as justifications for a wide array of non-regulation of Internet services as well as preemption of state and local Internet regulations, and overarching broadband competition and deployment policies.
  2. 230 as Regulatory and Classificatory Tool. The second era—perhaps the most promising for the Executive Order’s proponents—includes efforts by Bush-era FCC Chairman Kevin Martin to aggressively interpret Section 230 as a basis for the FCC to punish discriminatory behavior by ISPs, while attempting also to uphold his predecessor’s use of 230 as a basis for interpreting the Communications Act’s classificatory scheme to generally avoid regulating ISPs. But the good news for Trump ends there, as neither of the two Obama-era Democratic FCC Chairmen were willing to revisit the Martin’s use of Section 230 as a substantive basis for a non-discrimination mandate—nor was Trump’s hand-picked Republican FCC Chair, who picked up the Bush-era 230-as-deregulatory-classification-support conception but expressly rejected the use of 230 as a source of non-discrimination authority.

What follows is a mostly complete potted history of the FCC’s opinions citing Section 230. (Apologies for the citations to Westlaw, which I’ll replace with public domain cites if anyone needs it—just e-mail me!) This is mainly intended as a quick reference for folks who are relatively well-steeped in 230 and want to see its intersection with telecom law, but others may find it of more general interest too.

Continue reading “Section 230 as Telecom Law”

The Antitrust Concern Troll Bar Association’s Playbook for Not Solving Problems

Hal Singer has a delightful post up on the dynamics of gatekeeper power in the big tech companies and the role that competition has to play. Though I don’t agree with Hal on all the details, I agree in broad strokes that we (a) need stronger antitrust standards and enforcement and (b) need to supplement antitrust law with other regulatory measures that address issues like privacy, security, accessibility, and other public goods that the market and antitrust aren’t well positioned to provide.

However, I’m most delighted that Hal calls out a member of the Antitrust Concern Troll Bar Association, who criticizes the New Brandeisian antitrust movement for trying to make antitrust do too much, but, as Hal puts it:

. . . never explicitly calls for legislation outside of antitrust, such as privacy or nondiscrimination protections, which would demonstrate the sincerity of his argument.

Inject this directly into my veins. This line of argument is raised all the time by folks who ultimately have a deregulatory agenda, but instead of admitting that, always concern troll about the unintended consequences of addressing a problem with any particular solution and argue that it really should be addressed with some other solution that conveniently isn’t on the political table at the moment.

It’s a pattern. Taking a page from Harold Feld’s seminal breakdown of similar behavior at the FCC, I’ve distilled this into the Antitrust Concern Troll Bar Association’s Playbook for Not Solving Problems:

  1. If someone is talking about solving a problem with sector-specific regulation, observe that the problem would really be better solved by a general purpose regulator with a stronger commitment to enforcement that won’t be susceptible to capture or harm innovation within the sector—say, the Federal Trade Commission.
  2. If the general purpose regulator moves to solve the problem, point out the agency’s history of overreach and lack of legal authority, and note that the problem would really be more appropriately addressed by antitrust law. (For bonus points, observe, sadly, that the agency really is susceptible to capture after all.)
  3. If antitrust law is substantively incapable of solving this problem and people start agitating for antitrust reform, you have two options:
    • Point out that the reach of antitrust law is limited for Very Good Reasons™ (I suggest vague, pseudoscientific allusions to neoclassical economics or perhaps a call for deference to Robert Bork’s Force Ghost) and observe that sector-specific regulation would actually be a better solution. (Repeat step 1.)
    • Argue that antitrust law is actually working as intended and that the market will fix this problem. (Go to step 6.)
    • Either way, make sure to label the proponents of fixing antitrust law with a condescending term like “hipsters.” (This will show that you are both principled and cool.)
  4. If antitrust law is substantively capable of solving the problem but its enforcers are corruptly refusing or incompetently failing to act, praise their well-reasoned, grounded, and economically-informed restraint and observe that the market will fix this problem. (Go to step 6.)
  5. If antitrust enforcers look like they might actually act to solve the problem (don’t worry—this hardly ever happens), point out that they are out of control and/or corrupt and observe that the market will fix this problem. (Go to step 6. Really, anytime you’re in doubt, just go to step 6.)
  6. When the consequences of the market not solving this problem get really bad, gaslight the people pointing out the consequences and question whether it’s really a problem at all. Raise a lot of technical-sounding questions about whether This Is Really a Market Failure™ (spoiler alert: it never is!).
  7. If the consequences are so bad that they start to attract attention from powerful politicians, point out that the next Congress might need to begin a painstakingly slow inquiry into sector-specific regulation. (Repeat step 1.)

Trump’s 230 Executive Order and FCC Inside Baseball

A colleague asked me offline, in reference to the White House’s Executive Order attempting to retaliate against Twitter for flagging misinformation in Trump’s tweets:

[W]hat are your thoughts about whether the FCC has any authority to issue Section 230 regulations?

Barrels of ink have been spilled on the broader context of the Order itself, so I won’t retread the basics. Instead, I want to dig in a bit on the inside baseball of what might happen if the FCC actually tries to implement the provisions the Order that attempt to have the Commission:

expeditiously propose regulations to clarify:

(i) the interaction between subparagraphs (c)(1) and (c)(2) of section 230, in particular to clarify and determine the circumstances under which a provider of an interactive computer service that restricts access to content in a manner not specifically protected by subparagraph (c)(2)(A) may also not be able to claim protection under subparagraph (c)(1), which merely states that a provider shall not be treated as a publisher or speaker for making third-party content available and does not address the provider’s responsibility for its own editorial decisions;

(ii)  the conditions under which an action restricting access to or availability of material is not “taken in good faith” within the meaning of subparagraph (c)(2)(A) of section 230, particularly whether actions can be “taken in good faith” if they are:

(A)  deceptive, pretextual, or inconsistent with a provider’s terms of service; or

(B)  taken after failing to provide adequate notice, reasoned explanation, or a meaningful opportunity to be heard…

Harold Feld at Public Knowledge has done a very thorough job addressing whether the FCC can actually do any of this in two articles, which I highly commend to your attention, and which answer the question firmly: no, the FCC cannot do thisbut why not is complicated. Here, I want to briefly reiterate a few points made at length by Harold and supplement them with some discussion of the First Amendment.

Continue reading “Trump’s 230 Executive Order and FCC Inside Baseball”

Creativity and Closed Captions

(reviewing [reading] [sounds]: Closed-Captioned Media and Popular Culture by Sean Zdenek)

I’m delighted to join my colleagues at Authors Alliance with this cross-posted contribution to their ongoing series on Authorship and Accessibility, an outgrowth of a collaboration between AuAll, Silicon Flatirons (where I’m a faculty director), and the Berkeley Center for Law & Technology, which held a roundtable on the topic with technologists, authors, academics, lawyers, and disability advocates in Berkeley last year, summed up in this report co-authored by my students in the Colorado Law Samuelson-Glushko Technology Law & Policy Clinic.

Consumer groups and advocates have long been concerned about the quality of the captions that convey the aural components of video programming to viewers who are deaf or hard of hearing with video programming. While inaccurate and incomplete captions are often the butt of jokes, they aren’t so funny for people who are deaf or hard of hearing and rely on captions to understand the aural component of a video. For example, a single wrong letter on news captions might mean the difference between a story about a war in Iraq and a war in Iran.

That’s why consumer groups have fought hard for caption quality. Those efforts that culminated in the FCC’s 2014 adoption of wide-ranging caption quality standards for television, which require consideration of  that require captions to be accurate, synchronous, complete, and properly placed on the screen.

The FCC’s rules aim primarily at establishing a baseline of compliance to ensure that captions deliver a transcription of a program’s soundtracks that is as close to verbatim as possible given the unique attributes of sound and text. There are lots of good reasons that advocates have focused on verbatim captions over the years; in addition to  incomplete and incorrect captions, there is a lengthy and complicated history of simplifying and censoring the content of captions, which most recently entered the public eye in the context of Netflix’s censorship of captions on the rebooted Queer Eye for the Straight Guy. Verbatim is a principle that corresponds neatly to the goal of equal access: the captions should give viewers who are deaf or hard of hearing as near an equal experience in watching video programming as their hearing counterparts listening to the soundtrack.

However, advocates have also long urged their counterparts in the video industry to take captions seriously not just as a matter of accessibility, but as a matter of creativity. If filmmakers obsess over every aspect of a movie’s cinematography and sound design, why not the captions? In a production that spends millions of dollars to get all the details right, captions that are front and center for a film’s deaf and hard of hearing audience shouldn’t be an afterthought—they should be a core part of the creative process.

Sean Zdenek’s 2015 book Reading Sounds is one of the first efforts to rigorously explore the creative dimensions of captioning. Zdenek, a technical communication and rhetoric professor, endeavors to explore captioning as “a potent source of meaning in rhetorical analysis” and not simply a legal, technical, or transcription issue.

Zdenek’s exploration is an essential encyclopedia of scenarios whether captioning leaves creative choice, nuance, and subtlety to captioners and filmmakers. While captioning spoken dialogue seems on first blush to pose a relatively straightforward dialogue, Zdenek identifies nine (!) categories of non-speech information that are part of soundtracks, including:

  • Who is speaking;
  • In what language they are speaking;
  • How they are speaking, such as whispering or shouting;
  • Sound effects made by non-speakers;
  • Paralanguage—non-speech sounds made by speakers, such as grunts and laughs;
  • Music, including metadata about songs being played, lyrics, and descriptions of music; and
  • Medium of communications, such as voices being communicated over an on- or off-screen television or public address system.

Tricky scenarios abound. What if one speaker is aurally distinct from another, but his or her identity is unknown? (Imagine Darth Vader being identified as “Luke’s Father” in the early going of The Empire Strikes Back. How should a captioner describe the unique buzzing sound made by Futurama’s Hypnotoad? How should the captioner describe an uncommon dialect that may not be familiar to a hearing viewer, or which may have been invented by the filmmaker? What are the lyrics to “Louie, Louie,” exactly?

Zdenek expands into a variety of other problematic scenarios such as undercaptioning (the omission of non-speech sounds), overcaptioning (making prominent the exact content of ancillary speech happening in the background that a hearing viewer may be unable to parse precisely), and transcending the context of a scene to convey information that the viewer shouldn’t know. Delayed captions are all too familiar to deaf and hard of hearing viewers, but Zdenek explores the subtle relationship between caption timing, punctuation, the spoilage of time-sensitive elements afforded by the ability to read ahead of the dialogue, such as reading the aural punchline to the a visual setup, and the inadvertent creation of irony by captions that linger on the screen for too long. Zdenek even highlights the need to caption silence in dynamic contexts, such as a phone ceasing to ring or a person mouthing inaudible dialogue—scenarios that call to mind the controversial “silent” scene in The Last Jedi, which many hearing theater-goers were sure was a glitch but was an intentional choice by director Rian Johnson.

Zdenek also explores the role of captions in situating video in broader cultural contexts. For example, should a captioner identify a narrator who is a well-known actor with whom the audience will likely be familiar but who is uncredited in the film? How should music, such as the iconic NBC chimes, be described in text? And how can captioners be trained to capture cultural significance—especially if a captioner is a computer program converting text to speech automatically?

Zdenek does not offer complete solutions to all these questions and scenarios. But he extrapolates in unsparing detail (much of it presented in audiovisual context on the book’s companion website) how they arise and what considerations captioners and filmmakers might take into mind in thinking not just about how to comply with captioning law, but to author captions.

In doing so, he has also created a compelling reference for lawmakers and policy advocates to develop a richer, more nuanced understanding of the role that captions can play in advancing the civil rights of Americans who are deaf or hard of hearing to access video programming on equal terms. Zdenek is identifying dimensions of captioning that the next generation of video accessibility policy needs to consider and address.

Spectrum, Analogy, and the Line from Physics to Policy

In their 2017 essay Not a Scarce Natural Resource: Alternatives to Spectrum-Think, my colleague Pierre de Vries and former student Jeff Westling lay out an intriguing thesis: roughly speaking, that the oft-uttered-in-DC-circles analogy of the wireless radio spectrum as a scarce natural resource (or collection of resources) doesn’t withstand scrutiny.

According to Jeff and Pierre, ‘spectrum’ is an intellectual and legal construct amenable to multiple meanings, not a ‘resource’ that can be consumed:

The word spectrum is usually used rather loosely, with a variety of denotations. Since the word’s implications depend on which definition the writer had in mind, we propose five possible meanings (however, we suspect that writers often do not have any clear definition in mind, with
spectrum simply referring to the broad topic of radios and regulation). We find none for which all three attributes scarce, natural and resource hold simultaneously.

So the thesis goes, the relevant focus of most spectrum inquiries is not some identifiable ‘slice’ of the spectrum that is consumed like a resource, but rather the right to transmit on a particular frequency at a particular time in a particular geographic area without harmful interference occurring at the relevant receivers. (Though it’s not the focus of Pierre’s and Jeff’s inquiry, a close cousin of the spectrum-as-resource analogy, the ‘spectrum’-as-real-property analogy, is similarly pathological.)

The thesis seems obviously right to me, but I’ve never quite been able to put my finger on why it matters. I have long suspected that most spectrum policy folks who invoke  resource and property analogies do so as convenient shorthand for describing a mode of spectrum policy, but back up their real arguments with nuanced complications of the analogy that ultimately lead to similar results as those they’d reach if they started with the more accurate but unwieldy conception of spectrum rights as a right to transmit without interference.

However, it occurred to me last week in preparing to teach my telecom law students about spectrum that one particularly critical shortcoming of a reliance on the resource and property analogies is the risk that your world view about spectrum management (a political / legal question) will inform your view of how electromagnetic physics work (a scientific question). The informing should go the other direction—spectrum management is a malleable human construct that ought to be informed by the physics of transmission and reception.

The insight occurred to me in puzzling over how to convey to law students, in relatively simple but accurate terms, how radio transmission and reception works. (Luckily, my colleague Dale Hatfield offered me a brilliant analogy involving moving your hand in a pool and capturing the frequency and intensity of the generated waves with a bobber at the opposite end, which neatly illustrates or can be easily extended to illustrate all the fundamental concepts—transmission and reception, frequency and wavelength, amplitude, attenuation, interference, etc.) But it seemed critical to start with the physics because the potential for interference at the receiver ultimately dictates a significant proportion of telecommunications law and policy, including necessitating the involvement of regulators like the Federal Communications Commission to manage the practices around transmissions and reception to some degree or another. It also seems especially critical because many new-to-spectrum-policy law students, don’t understand even on an intuitive level (especially given that most of them don’t listen to AM or FM radio or watch broadcast TV) what a radio is or how it works.

Later in the week at a student-focused event I was tasked with moderating, I tried to press this point further. But the speakers (all lawyers) steadfastly declined my multiple invitations to explain transmission and reception, instead jumping immediately to resource and property analogies. It went something like this:

[Me:] How does a radio work? Please explain transmission and reception at a basic level.

[Panelist:] Well, spectrum is just like when you buy a house…

To be fair, these were smart lawyers sensibly using legal analogies to connect with a student audience at a law school. Fair enough.

But here’s the rub. When we start from the perspective of transmission and receptions, the possible modes of spectrum management unfold in a pretty obvious way—as do the advantages and shortcomings of each method. That is, when we start with the physics, the pathologies and tradeoffs of the different modes are fairly easy to understand. For example:

  • One mode for setting spectrum management, popularized by Ronald Coase in The Federal Communications Commission, is the use of property to order the right to transmit without interference, which facilitates (in theory, at least) all kinds of neat efficiencies through the ability to buy and sell licenses on the secondary market. But it’s pretty obvious to anyone with a basic understanding of the physics that it’s much more difficult to set property boundaries when you’re dealing with the complexities of interference, like out of band emissions, harmonics, noise floors, and so forth.
  • Another mode is the use of unlicensed spectrum bands, pioneered by Mike Marcus among others, which allow innovative new technologies to be developed without incurring the cost of spectrum licenses. There’s an obvious analogy to the commons, and a basic understanding of the physics quickly reveals the possibility of a tragedy-of-the-commons-like problem where over-proliferating transmitters cause so much interference to receivers that a band becomes useless for everyone.
  • Interference can proliferate under any mode of spectrum management. But a basic understanding of the physics of transmission and reception makes clear that detecting interference is a non-trivial task compared to, say, monitoring for trespassers.

With these understandings in mind, we can use property and resource analogies to a limited extent, understanding that they’ll only go so far because assigning the right to transmit without interference involves fundamentally different physics than assigning real property or natural resources.

But if we back into an understanding of how a radio works from a simplified resource or property analogy we risk thinking that the underlying radio physics work according to the analogy, which they often won’t. As a result, the pathologies of the management technique may no longer be obvious because, hey—maybe radio transmissions do fit within neatly circumscribed property lines, just like houses and yards. We then have to wrench the analogies to fit the physics in—“well, spectrum is like a natural resource, except that it’s infinitely renewable,” or “well, spectrum is just like real property, but with shifting and hard-to-measure five-dimensional boundary lines.” Some of the metaphors’ shortcomings are easy to understand, but some are quite persistent, such as the presumption that setting property boundaries is relatively easy.

This is sort of like what happens in law and econ, where we set policy based on economic theories rooted in potentially erroneous assumptions of how people act. With law and econ, though, people may start behaving according to the incentives we set for them, therefore making our models look correct even though the effect is largely observation bias, supercharged by the force of law, leading to the (more or less) expected policy outcomes.

But with spectrum policy, the physics of electromagnetism don’t change just because we misunderstand them, so flawed understanding leads to flawed policy, which leads to flawed outcomes—which will never self-correct. (Of course, spectrum management deals with macro- and microeconomic issues in addition to physics, so this overstates the case a bit. But you get the point.)

Of course, I don’t mean to argue that resource and property analogies aren’t useful for spectrum policy. The case for analogical thinking has been a mainstay in technology law ever since Judge Frank Easterbrook lambasted cyberlaw as ‘the law of the horse’. But, as Larry Lessig counsels in his rebuttal to Easterbrook, we are well served to understand how technology (or in Jack Balkin’s framing, technology’s societal salience) changes the aptness of our analogies. And understanding radio physics is critical to understanding not only why resource and property analogies are useful, but where they come up short.

Update (9.14): Pierre kindly pointed out via e-mail that this analysis ought to be complicated with the role of engineering in addition to physics. The distinction (and feedback loops) between the two are really important and interesting and probably worth a whole other post.