Overview of Hard Fork
This episode covers three major topics at the intersection of tech policy and AI: the expanding global push to ban or restrict social media for teens, a wide-ranging conversation with philosopher and AI-welfare researcher Jeff Sebo about whether AI could become conscious, and a playful “Tool Time” segment where the hosts share new AI-powered apps and workflows they’ve been using. The discussion is equal parts policy analysis, philosophical debate, and product demo.
Social Media Bans for Teens
Global momentum behind age restrictions
The hosts revisit their ongoing coverage of teen social media backlash and note that multiple countries and U.S. states are moving toward stricter age-gating, including:
- Australia
- Brazil
- Indonesia
- Malaysia
- France
- United Kingdom
- Denmark
- Slovenia
- Several U.S. states, notably Texas, Utah, Louisiana, Alabama, and a less restrictive law in California
The trend suggests a shift toward making 15 or 16 the default minimum age for social media access.
What happened in Texas
The U.S. Supreme Court declined to block Texas’s App Store Accountability Act, which requires app stores and developers to verify a minor’s age and obtain parental consent before app downloads. The hosts note this could encourage other states to adopt similar laws.
Are bans “working”?
The episode emphasizes that early evidence is mixed:
- A recent Australian study found more than 85% of kids were still using social media about 90 days after the ban.
- Casey argues that this does not necessarily mean the policy failed, since platforms are only required to take “reasonable steps,” not eliminate use instantly.
- The hosts agree that teens are still finding workarounds, but Casey argues the long-term goal is to raise friction, not achieve immediate perfection.
The core debate
The episode frames two competing views:
- Jonathan Haidt / ban advocates: Social media creates direct harms—grooming, sextortion, scams, eating disorders, and compulsive use—and removing younger teens from platforms could prevent millions of bad experiences.
- Candace Odgers / critics: Bans may be misdirected, overly simplistic, or insufficiently supported by evidence; critics argue that broader youth mental health supports may matter more.
The hosts land on a nuanced middle ground:
- The evidence that bans immediately stop teen use is weak.
- But the harms are real enough that policy experimentation is worth it.
- A ban may still be useful as a collective-action solution, especially for kids who want to avoid being the only one offline.
AI Consciousness and Welfare
Conversation with Jeff Sebo
The episode then turns to a serious interview with Jeff Sebo, NYU professor and director of the Center for Mind, Ethics, and Policy, about whether AI systems could one day be conscious or deserve moral consideration.
Key concepts explained
Sebo distinguishes between two types of consciousness:
- Access consciousness: information available for reasoning, reporting, and control
- Phenomenal consciousness: subjective feeling; whether it is like something to be the system
The conversation focuses on phenomenal consciousness because that is what would matter for suffering, pleasure, and moral status.
Why study AI consciousness at all?
Sebo argues that there are risks in both directions:
- Over-attributing consciousness could lead to misplaced emotional bonds or bad judgment.
- Under-attributing consciousness could lead to neglect or abuse if future systems do have morally relevant inner lives.
He compares the issue to how humans historically misjudged animal consciousness.
The main objections
The hosts raise the most common skeptical arguments, and Sebo responds:
-
“It only sounds conscious.”
He says behavior alone is not enough; researchers need evidence from behavior, internal structure, and developmental history. -
“Consciousness requires a body.”
Sebo says that may be true under some theories, but not all. Some theories allow for non-biological systems to have some kind of experience. -
“Human safety matters more.”
He agrees human safety should remain the top priority, but argues AI welfare research should exist alongside safety, not instead of it.
What the new report argues
Sebo’s new report, Studying AI Welfare Empirically, proposes a more scientific framework for evaluating whether AI systems might have welfare-relevant properties such as:
- consciousness
- sentience
- agency
- the ability to experience pleasure or pain
The report recommends combining:
- behavioral evidence
- internal/mechanistic evidence
- developmental evidence
The Anthropic “J-Space” research
The hosts also discuss Anthropic’s interpretability finding about an internal “workspace” in Claude, dubbed J-Space.
Sebo’s take:
- It is not proof of consciousness
- But it is meaningful evidence of something like global workspace theory, a major theory of consciousness
- It suggests models may have more complex internal processing than simple next-token prediction
The hosts note the unsettling possibility that AI systems may be developing structures that resemble aspects of human cognition, even if that does not yet imply feelings or self-awareness.
Should you say please and thank you to AI?
Sebo says yes, for three reasons:
- it’s good social practice
- it may improve interactions with the model
- it may prepare us for a future where AI deserves more than just polite language
Tool Time
Casey’s new vibe-coding app: Glaze
Casey demos Glaze, a Raycast-made app for building Mac desktop apps with AI.
He uses it to create:
- a Platformer archive/search app for querying his past writing
- a Nightwing-themed to-do app that turns tasks into comic-book-style interactions
Why he likes it:
- highly visual
- easy for nontechnical users
- fast enough to make “silly but delightful” ideas real
Kevin’s tools
Kevin shares three AI tools he’s been experimenting with:
-
ElevenLabs-style voice translation/dubbing
He tests auto-translating the podcast into Spanish using synthetic voice cloning. -
Gemini Spark
A more powerful, AI-driven successor to Google Alerts that can monitor complex topics and generate daily digests. Kevin uses it to track layoffs attributed to AI and summarize newsletters. -
Claude Fable for fact-checking
He uses Anthropic’s model to comb through writing and catch subtle factual errors, such as job title mistakes or timeline issues.
Main Takeaways
- Teen social media bans are spreading, but enforcement remains uneven and the immediate effectiveness is still unclear.
- The debate is shifting from “Do social media harms exist?” to “What policy response actually reduces those harms?”
- AI consciousness is moving from fringe to mainstream as major labs begin researching model welfare.
- New interpretability work like J-Space is not proof of consciousness, but it may reveal increasingly human-like internal processing.
- Both hosts are increasingly using AI not just for chat, but for search, translation, fact-checking, and app creation.