Nov 9, 2020 · 42m · another-podcast
Is content moderation a dead end?
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Benedict Evans and Tony Cowan-Brown examine the structural, legal, and historical complexities of digital content moderation, arguing that online speech challenges stem from unprecedented global connectivity and human cognitive vulnerabilities rather than simple engineering oversights. By evaluating historical precedents from urbanization to automotive safety, they demonstrate why regulating modern public discourse requires continuous institutional evolution rather than technocratic silver bullets.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 71.1% of the talking time here. How this is scored →
The hosts as informed peer 8.0 Guest teaching 2.7 Guest disagreement 1.0 The hosts pushing back 2.0
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 20:30 Questioning platform profit motives versus harm reduction
Cowan-Brown challenges the benign view of platforms by asserting that both major parties complain about bias while platforms prioritize engagement and revenue above all else.
Hardest push from the hosts ▶ 36:38 Pushing on the practical limits of deleting false statementsEvans confronts the easy assumption that platforms should remove misinformation by asking repeatedly whether Twitter should actually delete a verifiably false tweet by a respected journalist.
Biggest teaching moment ▶ 33:00 Gendered blind spots in automated moderation systemsCowan-Brown informs the host that text-centric moderation systems fail female politicians because gendered abuse online relies heavily on visual, audio, and meme formats.
The host holds their own ▶ 34:25 The Mechanical Turk framing of social network vulnerabilitiesEvans synthesizes platform mechanics into a powerful conceptual model where four billion human users act as the CPUs and cognitive biases act as the software exploits.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| The Scale of Universal Connectivity and the Breakdown of Traditional Speech Norms | 8 | 1 | 0 | 0 | Evans establishes a comprehensive historical framework, contrasting the scale of the early internet with modern universal smartphone adoption and explaining the breakdown of traditional social speech norms. | |
| The Political Dimension, Global Incidents, and Defining Hate Speech | 8 | 3 | 1 | 2 | Evans dismantles the algorithmic blame narrative by referencing real-world mob violence on WhatsApp in India and unrest in Myanmar, while Cowan-Brown introduces her experience at NationBuilder regarding the difficulty of defining hate speech. | |
| First Amendment Limitations, Monopoly Power, and Platform Legitimacy | 8 | 3 | 1 | 3 | Evans uses historical AT&T monopoly precedents and international examples like Inner Mongolia to reject simplistic First Amendment definitions of censorship, while Cowan-Brown explores the legitimacy of Silicon Valley product managers deciding speech rules. | |
| The Malware Analogy, Architectural Design, and Business Realities | 8 | 3 | 2 | 2 | Evans draws an extended analogy to Windows malware architecture versus iOS walled gardens to illustrate unscalable moderation, while Cowan-Brown offers observations on political bias debates and software revenue models. | |
| Urbanization Parallels, Automotive Regulation, and Moderation Limits | 8 | 4 | 1 | 2 | Evans draws analogies between platform moderation limits, 19th-century urbanization, and car safety regulations, while Cowan-Brown educates the discussion on how gendered harassment is disproportionately visual rather than text-based. | |
| Human Cognitive Biases, Viral Misinformation, and the Process of Moderation | 8 | 2 | 1 | 3 | Evans frames platforms as Mechanical Turks running on human cognitive biases and presses whether platforms should censor viral journalistic misinformation, concluding with a Burkean perspective on evolutionary social norms. |
Statements from this episode (14)
Evans: Smartphone adoption grew from 100M PCs in 1994 to over 5B users
“When the internet, when Netscape started in 1994, there were like a hundred million PCs on earth. And now something over five billion people have a smartphone.”
Evans: Speech was historically governed by social pressure rather than law
“We have, like, two centuries of working out how speech works. Like, what are you allowed to say and where? And that's, like, a really complex tapestry of Social convention, and peer pressure, and professional standards, and a very, very small piece of law, sor…”
Evans: Online harms stem from total connectivity, not algorithms or ads
“A couple of years ago, there was this whole thing in India of lynch mobs forming based on wild rumors that were spreading on WhatsApp, which was super interesting because WhatsApp is not, doesn't have an algorithm and doesn't have advertising. And so like thos…”
Cowan-Brown: There is no global definition of hate speech
“I mean, when I used to work at nation builder, we had also such a hard time of defining. And again, it's a little bit different there because it's not a platform that amplifies I'll come back to that in a second, but it's not a platform. You don't go there to …”
Evans: Private platform moderation affecting billions is a free speech issue
“Well, it's a private corporation that controls how billions of people talk to each other. If they decide that billions of people can't say that, that is ipse fact to a free speech question. Doesn't matter that they're not the government or not.”
Cowan-Brown: US campaign software companies must pick Republicans or Democrats
“In the US, campaigning software is politicized. So you create software for one political party, not for both. And so you either create software or campaigning software, organizing software for the Republicans or the Democrats, but you can't be seen doing it fo…”
Evans: Breaking Up Facebook Only Relocates Moderation Power to Five Companies
“Now you could argue that's, that means that you should just break up Facebook, but that just relocates the problem because it's now it's five people collectively who are making that decision instead of one, but the decision is still there.”
Evans: Content moderation is unscalable without architectural constraints
“The creation of bad stuff is fundamentally infinitely scalable, because it's people. It's four billion people saying, retweeting stuff that fits their prejudices. And the content moderation or the virus scanning is fundamentally unscalable, because you're one …”
Evans: Facebook generates trivial revenue from news content
“All my friends at Facebook would basically say, we wish we didn't have news at all. We don't make any money from it. We actually don't make any money from news, although the amount of money is trivial. It's a huge amount of pain and aggravation”
Cowan-Brown: NationBuilder failed to profit from European political campaigns
“When I was building out NationBuilder in, in Europe. We didn't make any money from campaigns. Presidential campaigns come and go. So you have a great amount of revenue. You've got great RR for a couple of, you know MRR, sorry, not AR. Good monthly revenue for,…”
Evans: Every platform allowing file sharing hosts child sexual abuse material
“Any internet platform that allows people to share files of any kind has a child porn problem. Every single one.”
Evans: Tech regulation must distinguish platform liability from civic rules like automotive laws
“We regulate emissions and safety, and that's on the car companies. We also have speed limits. That's not on the car companies. And we have parking restrictions. That's also not on the car companies.”
Cowan-Brown: Online abuse targeting women is visual, but moderation tools focus on text
“Most men, the way they have misinformation or get attacked online is actually written content. With women, it's generally pornographic in nature, it's memes, it's visual, but a lot of the censorship and a lot of the content moderation tools that are being put …”
Evans: Big internet platforms operate as Mechanical Turks powered by human biases
“You should think of any big internet platform as basically a Mechanical Turk. So PageRank is a Mechanical Turk. Because PageRank, the whole point of PageRank is instead of trying to work out what that webpage is, Probably should have said this earlier. Instead…”