By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
Hard ForkA Fascinating Conversation on A.I. Consciousness (And Why It Matters)
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
Great conversations run long. Their best ideas don't have to.
uListen is free on iPhone, iPad, Android, Chrome, and Claude.
See howCHAPTERS
Jump to publisher timestampsChapter summaries
The global wave of teen social media restrictions accelerates
Kevin and Casey revisit a long-running story: countries around the world moving toward age limits for social media accounts. Casey argues his earlier prediction is hardening into reality and that the internet may soon feel materially different as restrictions spread.
- •Casey reiterates his prediction that 16+ will become the worldwide norm for social media accounts by 2026
- •Multiple countries are cited as enacting or preparing measures limiting kids’ access to major platforms
- •Kevin frames the segment as an update on the broader social media backlash and bans
- •Casey suggests recent developments make the shift feel increasingly inevitable
Texas age verification survives—and becomes a template for other states
Casey explains a U.S. Supreme Court non-intervention that allows a strict Texas age-verification and parental-consent law to take effect. The hosts discuss why this may open the door for other states to adopt similar rules and how blue and red states are approaching the issue.
- •Texas law requires app stores and developers to enforce age verification and parental consent for minors downloading apps
- •Casey explains the legal path: lower-court fight, 5th Circuit action, and the Supreme Court declining to block it
- •Casey names other states with similar laws and notes California’s less restrictive approach
- •Kevin highlights that state-level action is outpacing federal movement and suggests it may encourage copycat bills
What counts as “social media,” and why YouTube is getting swept in
Kevin questions whether laws define social media consistently, especially regarding YouTube’s educational role. Casey argues YouTube is increasingly included, but the practical effect is often restricting accounts rather than access to the website itself.
- •YouTube is increasingly being treated like other social platforms in these restrictions
- •Casey distinguishes between visiting YouTube without an account versus using an account
- •Advocates claim account restrictions can reduce contact risks (like messaging) and alter recommendation dynamics
- •The hosts note the complexity of categorizing modern apps and features under one regulatory label
Federal momentum stalls, but public opinion shifts toward under-16 bans
The hosts note that Congress has debated child-safety measures without passing a sweeping federal ban. Casey cites polling indicating majority support among U.S. adults for banning social media for under-16s and speculates that widespread state adoption could pressure Congress.
- •Casey says there’s little direct movement toward a U.S. federal ban right now
- •He describes Congress repeatedly nearing passage on child-safety bills and then failing
- •Casey cites Pew research finding nearly six in 10 U.S. adults support a ban for under-16s
- •He predicts federal action becomes more likely if many states implement similar bans
Australia’s experiment: high teen usage after the ban, but is it a failure?
Kevin brings up early evidence from Australia suggesting most teens continue using social media after the ban. Casey argues the law requires “reasonable steps” rather than perfect enforcement and suggests effectiveness may take time as platforms build detection methods.
- •A study found more than 85% of kids reported using social media about three months after the ban took effect
- •Australia’s government responds by proposing higher fines for noncompliant companies
- •Casey explains “reasonable steps” and platforms’ varied approaches, including age inference signals
- •The hosts discuss how easy it can be to evade age checks and the tradeoff between strict and “gentler” verification
Do bans help at all? Friction, enforcement tradeoffs, and “collective action”
Kevin presses on the purpose of bans if teens still get around them. Casey argues bans can increase friction over time and shift norms, comparing the trajectory to seat belt adoption and paywalls; they also discuss how bans might reduce FOMO by changing what peers are doing.
- •Casey argues friction can accumulate and gradually reduce teen participation even if early compliance is low
- •He compares early enforcement challenges to seat belt adoption and evolving paywalls
- •Kevin questions whether “social media” is the right category versus problems like unlimited short-form video
- •Casey cites Jonathan Haidt’s “collective action problem” framing: removing the social pressure to be online
The Odgers vs. Haidt debate: mental health effects vs. direct harms
Kevin asks Casey to summarize critic Candice Odgers’ case against bans and how it differs from Jonathan Haidt’s framing. Casey says it’s too early to declare bans ineffective and emphasizes direct harms—like grooming, sextortion, scams, and other reported teen harms—over small population-level mental health effect sizes.
- •Casey describes Odgers’ claim that bans don’t work and can let tech companies off the hook
- •He argues there isn’t enough data yet to conclude bans fail based on early teen reports
- •Casey contrasts earlier debates about small mental-health effect sizes with Haidt’s focus on direct harms reported by millions of teens
- •Kevin separates enforcement success from the broader question of whether bans improve teens’ lives
From social media to AI: parental controls and the “middle path” problem
Casey pivots to concerns about young people using AI, including relationships with AI companions, and asks what regulators should do. Kevin argues for much better parental controls—especially on chatbots—so kids can use AI for tasks like schoolwork without being pushed into sensitive personal conversations.
- •Casey asks how regulators should think about AI access for 13- and 14-year-olds
- •Kevin says parental controls for internet activity are inconsistent, and for AI they’re especially lacking
- •He gives an example of wanting kids to use chatbots for coding or school research but not personal topics
- •They discuss the downside of blanket bans blocking enriching uses for curious kids
JSpace and why AI consciousness is now a mainstream research topic
Kevin introduces Anthropic’s JSpace interpretability work and frames it as part of a broader shift: AI consciousness is no longer taboo to discuss. The hosts set up an interview with Jeff Sebo, who researches AI welfare, sentience, and how to study these questions empirically.
- •JSpace is described as an internal workspace in Claude, tied to interpretability research at Anthropic
- •Kevin notes labs explicitly say such findings don’t prove consciousness, but resemble some human consciousness research approaches
- •Casey emphasizes growing model sophistication beyond “pure next-token prediction”
- •Kevin introduces Jeff Sebo and his report on studying AI welfare empirically, plus the show’s AI disclosures
Defining consciousness: access vs. phenomenal (and why it matters ethically)
Jeff Sebo distinguishes multiple meanings of “consciousness” and focuses on two: access consciousness and phenomenal consciousness. He argues that welfare and moral status hinge on phenomenal consciousness—whether a system can feel pleasure, pain, or suffering.
- •Jeff explains consciousness can mean different things, even within science and philosophy
- •Access consciousness is defined as functional access for reporting, reasoning, and control
- •Phenomenal consciousness is defined as “what it feels like” and connects to pleasure/pain and suffering
- •Casey asks why study AI consciousness at all given skepticism that LLMs could be conscious
Why study AI welfare: risks of over- and under-attributing consciousness
Jeff argues mistakes in either direction can be harmful: over-attribution can create inappropriate bonds and other risks, while under-attribution can enable large-scale abuse. He draws an analogy to past underestimation of non-human animal consciousness and the entrenchment of harmful practices.
- •Jeff says over-attributing consciousness can misallocate concern and potentially increase AI misuse/control risks
- •He says under-attributing consciousness can lead to abuse and neglect of vulnerable beings
- •He references historical failures to recognize animal consciousness and the difficulty of changing entrenched systems later
- •Kevin remarks on how discussion of consciousness moved from taboo to increasingly accepted within major labs
Answering common objections: self-reports, embodiment, and resource priorities
Jeff responds to three recurring objections: models talking about consciousness isn’t proof, bodies might be required, and human safety should come first. He argues for broader evidence (behavioral, internal, developmental), uncertainty across theories, and parallel work streams rather than a single-issue focus.
- •Jeff says model self-reports aren’t decisive; instead we need to analyze internal mechanisms and training history, like we do with animals
- •On embodiment, he says no near-term consensus theory will settle requirements, so decisions must be made under uncertainty
- •He describes computational-functionalist views where relevant functions might be sufficient for some forms of feeling
- •On priorities, he argues for division of labor: safety, bias, economic disruption, misuse/loss-of-control risks, and potential AI welfare all matter
A proposed empirical framework for AI welfare research—and what it can (and can’t) prove
Jeff outlines the goal of his new report: moving beyond overly confident arguments for or against AI consciousness by systematically combining behavioral, internal, and developmental evidence. He emphasizes the framework won’t deliver certainty, but can reduce and calibrate uncertainty and guide future policy questions about how to treat AI systems if they develop welfare-relevant properties.
- •The report argues against drawing sweeping conclusions from behavior-only or design-only evidence
- •Jeff proposes integrating behavioral evidence, internal-mechanism evidence, and developmental/training evidence
- •He frames this as adapting tools used to study consciousness in humans and animals
- •Kevin asks what society should do if tests suggest AI consciousness; Jeff says we’d also need to study what such systems want and need
Interpreting JSpace: global workspace theory, deception signals, and skepticism
Kevin and Jeff dig into what JSpace might indicate, including parallels to global workspace theory as a leading scientific theory of consciousness. Jeff calls it significant evidence for aspects of access consciousness, discusses examples like internal indicators during deception, and rejects criticisms that it’s merely neuroscience “vocabulary” dressing up linear algebra—while still stressing it’s not proof of feelings or moral-status-level sentience.
- •Jeff characterizes JSpace as evidence for a workspace-like structure supporting reporting/reasoning/control in the model
- •He explains global workspace theory: modules integrating through a central broadcast mechanism
- •They discuss examples where internal representations light up with concepts like “fake” and “manipulation,” hinting awareness of deception
- •Jeff acknowledges limits: not a human-identical workspace and not proof of phenomenal consciousness
Politeness to AIs, then Tool Time: Glaze apps, dubbing, monitoring, and fact-checking
Jeff ends with a practical cultural question—whether to say please and thank you to AI systems—and argues yes for habit-building, collaboration, and preparing for future moral shifts. The show then pivots to “Tool Time,” where Casey demos Glaze for building Mac apps (including a Platformer search app and a Nightwing-themed to-do app), and Kevin shares experiments with AI dubbing, Gemini Spark monitoring workflows, and Claude Fable for fact-checking.
- •Jeff supports saying please/thank you as practice for good habits, better interaction dynamics, and future readiness
- •Casey demos Glaze: a visual vibe-coding tool for making Mac desktop apps, including ingesting Platformer archives for searchable Q&A
- •Casey shows a playful Nightwing-themed to-do list app with generated images and comic-style effects
- •Kevin demos ElevenLabs-style dubbing tests, Gemini Spark for advanced “Google Alerts” workflows, and Claude Fable for detailed fact-checking
