Ethics in IT — Week 2 · Ch. 5
Freedom of Expression
The First Amendment, Section 230, and the question of who controls editorial judgment online.
13 slides ~16 minutes ETH-06 · The Factionless
By the end of this module, you will distinguish constitutional speech protections, understand Section 230's two protections, recognize DMCA abuse, analyze Reno v. ACLU, and evaluate four content-moderation approaches.
Slide 2 of 13
First Amendment Scope
It binds the government. Not your employer. Not the platform. Not the ISP. Get this right or every analysis that follows is wrong.
The most-misunderstood right in IT ethics. "My free speech!" against a private platform is a category error. The platform may be wrong to remove your post — that's an ethics question, not a constitutional one.
Slide 3 of 13
Five Categories of Speech
Even where the First Amendment does apply, protection isn't uniform. The category determines the level.
Political
Highest, near-absolute. Strict scrutiny on content-based restrictions.
Commercial
Lower, intermediate scrutiny. Government may regulate false / misleading.
Defamation
Not protected when false & injurious. Public figures: actual malice. Private: negligence.
Obscenity
No protection. Three-part Miller test applies.
True Threats
No protection. Statements communicating serious intent of violence to a specific target.
For IT systems that classify content (moderation, ads, search ranking), these categories matter as ethical guideposts even when the First Amendment isn't the legal hook. They're the social contract for what speech does what work.
Slide 4 of 13
Section 230
26 words from 1996 that built the modern internet — and that we now spend most of our time arguing about.
The text: "No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider." — 47 U.S.C. ยง 230(c)(1)
What it does
Two distinct protections: (1) platforms aren't liable for user-posted content; (2) platforms can moderate content "in good faith" without becoming a publisher.
Without (1), no Reddit, no YouTube, no comment sections at scale.
Why it's contested
Section 230 was designed for 1996 bulletin boards. Modern platforms have algorithmic amplification, billions of users, and editorial-grade moderation systems. The law treats them like neutral pipes; they aren't.
Slide 5 of 13
Section 230 in Four Parts
Two protections, one ethical challenge, one set of exceptions.
No Publisher Liability
Platforms aren't liable for content posted by users. Without this, scale-hosting user-generated content is legally infeasible.
Good Samaritan Moderation
Platforms may moderate "in good faith" without becoming publishers. Section 230 protects the moderation decision itself.
The Ethical Challenge
Algorithmic amplification may constitute editorial judgment. Critics argue billion-user platforms with ranking algorithms are different in kind from 1996 BBSes.
Exceptions
230 doesn't cover: FOSTA-SESTA (sex trafficking facilitation), federal criminal liability, IP claims under the DMCA. Carve-outs grow over time.
The two-edged design: Section 230 was passed to encourage platforms to moderate (without making them publishers) AND to encourage them to host more content (without making them liable). It tried to do both. Whether it does either today is the active debate.
Slide 6 of 13
Defamation, Anonymity, & DMCA Abuse
Three ways the legal architecture of online speech gets exploited.
Defamation Online
Same standard as offline: false statement of fact (not opinion), published to a third party, damaging reputation. Jurisdiction complications matter most.
Anonymous Speech
Protected as an aspect of First Amendment political speech. Courts apply balancing tests to subpoenas seeking to unmask anonymous speakers.
DMCA Takedown Abuse
Bad-faith DMCA notices are filed against content the filer doesn't own copyright in — using takedown as a censorship tool. Counter-notice restores it, but the chilling effect is real.
The DMCA pattern: a politician dislikes a video critical of them. They file a DMCA takedown citing footage they don't own. The platform must remove or lose safe harbor. The video is offline for days/weeks while the creator files a counter-notice. The censorship has already worked.
Slide 7 of 13
Case: Reno v. ACLU (1997)
9-0. The internet gets the same First Amendment protection as print.
The challenge: the 1996 Communications Decency Act criminalized knowing transmission of "obscene or indecent" material to minors online. The ACLU argued it was an unconstitutional restriction on free speech.
The decision: the Supreme Court struck down the indecency provisions 9-0. The internet, the Court ruled, deserves the highest First Amendment protection — equivalent to print media, not the lower protection accorded to broadcast.
Why not broadcast?
Broadcast media is "invasive" — thrust upon listeners by limited spectrum allocation. The internet isn't — users actively seek content.
Captive audience?
Court rejected the argument. Online users are not a captive audience the way TV viewers were in 1978's Pacifica.
Least restrictive?
The provisions weren't the least restrictive means of protecting minors. Filtering and parental controls existed.
Slide 8 of 13
What Would You Do?
Your job is to ship the filter. The category is "harmful content." The definition keeps drifting wider.
You are a content policy engineer at a large platform. Legal has asked you to build an automated system that filters "harmful content" — a category defined broadly enough to include political speech the company finds controversial. You've been told this will reduce regulatory risk in certain markets. What does Section 230 tell you about the company's legal position, and what does the ACM Code tell you about your individual obligation?
Section 230 says
The company likely has legal cover to moderate broadly — "good faith" is interpreted liberally. The legal floor is permissive.
ACM says
Avoid harm to the public; honor confidences; act in the public interest. A filter that suppresses lawful political speech is not in the public interest, even if it's legal.
The professional move
Document the definition drift. Push for narrower scope. Refuse to ship a filter whose definition is too broad to defend publicly. The transparency test applies.
Slide 9 of 13
Four Moderation Approaches
Each carries a different power balance. Each fails differently.
Remove & Notify
Remove prohibited content; notify the poster of the policy.
Tension: due process — poster may have no appeal; automated false positives.
Label & Reduce
Leave content; add warnings; reduce algorithmic amplification.
Tension: doesn't prevent harm; the labeling decision is itself editorial.
Deplatform
Remove accounts that repeatedly violate policy.
Tension: most severe; concentrates power; oversight typically internal.
Third-Party Oversight
Independent review boards (e.g., Meta's Oversight Board).
Tension: limited scope; non-binding; appearance of accountability without structural change.
The hardest moderation question isn't "should we remove this?" — it's "who decides, and to whom are they accountable?" Platforms historically answered: us, and to our policy team. That answer is being renegotiated in real time.
Slide 10 of 13
Moderation: Power vs. Process
Two axes. Where a platform sits determines what it tells the world about its values.
The question to ask any moderation system: what is the appeal process? If the answer is "there isn't one" or "we'll consider it," you're in the bottom-right quadrant — where power compounds without process to balance it.
Slide 11 of 13
The "Neutral Platform" Myth
A platform that ranks content has already made an editorial decision. Neutrality was never the offer.
The claim
"We're neutral infrastructure. We just host what users post. Section 230 says we're not the publisher."
Used to defend: minimal moderation, unmoderated amplification, refusal to address downstream harm.
The reality
Platforms design the algorithms that determine what content users see, in what order, and how widely it is amplified.
That is editorial judgment. The decision not to act is also an editorial decision.
The professional question isn't whether platforms exercise editorial judgment — they do, by definition. It's whether that judgment is visible, explained, and accountable. "We're neutral" is an answer that ducks all three.
Slide 12 of 13
What Engineers Can Actually Do
You don't write the policy. But you build the system that enforces it — and that's where the ethics lives.
Make the model visible
If the moderation model uses signals X, Y, Z — document them. If it uses a labeled training set with known bias — flag that. The model's blindspots are part of the system spec.
Build the appeal path
Design the system to support contestation: keep an audit log, support manual review queues, surface the policy text the user allegedly violated. Process is architecture.
Test the failure mode
A moderation system that gets 99% right with 1% false positive across a billion posts is wrong 10 million times. Run the test that exposes the wrong-10-million case.
The professional position: "policy isn't my job, I just write the code" is not available once the code embodies the policy. The ACM/IEEE codes ask you to refuse work whose ethical impact you can't defend. Moderation engineering qualifies.
Slide 13 of 13
Module 6 Summary
Freedom of Expression — key takeaways.
1The First Amendment binds the government. Not platforms, not employers, not ISPs. Get this right or every analysis that follows is wrong.
2Five speech categories — political (highest), commercial (intermediate), defamation (not protected when false), obscenity (no protection), true threats (no protection).
3Section 230 = two protections: no publisher liability for user content; "good faith" moderation doesn't make a platform a publisher.
4Algorithmic amplification is the open question. 1996 BBSes didn't have ranking algorithms. The law treats modern platforms like neutral pipes; they aren't.
5DMCA takedown abuse is a known censorship pattern. Bad-faith notices remove content while counter-notices catch up. The chilling effect already worked.
6Reno v. ACLU (1997): the internet gets print-level First Amendment protection, not broadcast. Active-seeking media, not invasive.
7Four moderation approaches — remove & notify, label & reduce, deplatform, third-party oversight. None is neutral; each carries a different power balance.
8The neutral-platform myth. Ranking algorithms are editorial judgment. The professional question is whether the judgment is visible, explained, and accountable.
Next up: Module 7 — Intellectual Property. Copyright, patents, trademarks, trade secrets, software licensing, and the Waymo v. Uber trade-secret case.