Notes — Cass Sunstein on Liberalism and Rights in the Age of AI

Notes on Cass Sunstein in conversation with Tyler Cowen — Conversations with Tyler #262, recorded 10 October 2025 at Harvard Law School; published 26 November 2025.

Four questions — Adler’s reading frame

Q1 — What is it about as a whole? A wide-ranging interview keyed to Sunstein’s 2025 book On Liberalism: In Defense of Freedom, moving from the durability of liberalism itself through immigration, the generational health of liberal thought, a run of named-thinker verdicts (Parfit, Mises, Hayek, Mill, Popper, Waldron, Ullmann-Margalit), the origins of illiberal ‘wokeism’ inside liberal premises, indigenous rights, and then a sustained legal-doctrinal section on how AI bears on the First Amendment, manipulation, libel, and AI-run trials, before closing on Bob Dylan as liberal exemplar and Sunstein’s forthcoming work on animal rights.

Q2 — How is it argued? Cowen’s characteristic pressure-testing format: rapid, specific hypotheticals (a toaster’s free-speech rights, a wall’s height at the southern border, a tweet one should be able to sue over) rather than open prompts, designed to force Sunstein from abstract commitment to concrete application. Sunstein answers in a consistent register across topics — state the doctrinal or philosophical starting point, then work a hypothetical through it — and repeatedly distinguishes categories that look similar but are legally or conceptually load-bearing (falsehood vs manipulation; a content-based vs content-neutral restriction; AI’s own rights vs the rights of humans behind the AI).

Q3 — Is it true, in whole or part? Sunstein’s First Amendment analysis (AI as a non-rights-holding speaker whose human producers and users retain their own rights) tracks settled doctrine — the communist-visa case he cites (Kleindienst v. Mandel-adjacent reasoning) and the New York Times v. Sullivan actual-malice standard are both real and correctly characterised. His autobiographical claims (DHS service, jury duty, the Tulsa Dylan museum visit) are first-hand and not independently checkable here [?]. His literary and historical judgements (Orwell’s ambivalence in 1984, the Hayek–Mill correspondence book, Parfit’s philosophical trajectory) are contestable interpretive claims, explicitly offered as such — Cowen pushes back on several (Parfit’s individualism, Hayek’s harm-principle ambivalence) and Sunstein concedes ground rather than holding a fixed line. The claim that ‘no legal system has a right not to be manipulated’ is a strong generalisation Sunstein does not fully substantiate beyond noting ‘incipient movements’ in the EU and US [?].

Q4 — What of it? The episode’s most portable idea for the wiki is Sunstein’s answer to why liberalism does not maintain itself: not internal contradiction, but the absence of any mechanism within liberalism for generating the norms of cooperation, charity, and mutual support a flourishing society also needs — liberalism does not undermine these, but has no resources to replenish them either. This directly extends Enlightenment‘s open question (there flagged as Pinker’s ‘incomplete recursion’ on what maintaining liberal norms requires) with a distinct, legally grounded answer. The AI free-speech material — AI as toaster, not speaker; rights sit with producers and users, not the model — is a clean, quotable doctrinal frame absent elsewhere in the wiki. The manipulation material extends Sunstein’s own behavioural-economics corpus (nudge, choice architecture) into a proposed new legal right.

Glossary

Liberalism (Sunstein’s working definition) — a political order organised around individual freedom, the rule of law, and pluralism, which does not itself supply the norms of cooperation, charity, and mutual support a flourishing society also needs. [§ Opening: is liberalism self-undermining]

Nudge / choice architecture — Sunstein and Thaler’s term (from Nudge, 2008) for a change to how choices are presented — a default, a frame, a prompt — that steers behaviour without removing any option or changing incentives. Not discussed at length in this episode but the background against which his manipulation and algorithm-aversion remarks sit; see Choice Architecture. [§ On manipulation, libel, and nudging]

Right not to be manipulated — Sunstein’s proposed new legal right, analogous to the early-20th-century emergence of a right to privacy: a claim against being tricked into bypassing one’s own reflective, deliberative capacities (as distinct from being told something false). He proposes starting with egregious commercial cases — hidden terms, concealed long-term financial commitments. [§ On manipulation, libel, and nudging]

Content-based vs content-neutral restriction — First Amendment doctrine distinguishing a law that regulates speech because of what it says (presumptively unconstitutional) from one that regulates speech for a reason unrelated to its content, such as time of access (subject to a lower bar). Sunstein uses this to separate a ban on asking an AI about Israel (content-based, invalid) from a curfew on AI access (content-neutral, needs a justification but is not automatically invalid). [§ On the First Amendment rights of AI]

Incompletely theorized agreement — Sunstein’s term (from his own earlier work, invoked here via Parfit) for a shared practical conclusion reached by people who hold different, even incompatible, underlying justifications for it — e.g. supporting free speech on Kantian, truth-seeking, or utilitarian grounds simultaneously. [§ On Parfit, Mises, and Hayek]

Algorithm aversion / AI aversion — the empirical tendency to trust human judgment over an algorithm’s even when the algorithm is no less accurate, sometimes rational (context-specific human knowledge advantages) and sometimes not (a bare preference for human involvement irrespective of accuracy). Sunstein distinguishes the two and separately researches AI aversion as a related but distinct phenomenon. [§ On trial by AI]

Actual malice standard (New York Times v. Sullivan, 1964) — the US libel standard requiring a public-figure plaintiff to show a false statement was made knowingly or with reckless disregard for the truth. Sunstein wants it loosened, not for damages, but to allow a cheap, fast remedy (a nominal payment plus a correction) even for merely negligent falsehoods. [§ On manipulation, libel, and nudging]

Key claims by section

Opening: is liberalism self-undermining [§ Opening: is liberalism self-undermining]

Sunstein’s central claim in On Liberalism: the risk to liberal societies is low-probability self-undermining and higher-probability external undermining by illiberal and anti-liberal forces. Liberalism does not manufacture the norms of cooperation, charity, and mutual support a flourishing society needs, and — crucially — has no particular resources to defend those norms once they are under attack. Sunstein locates the threat partly in ‘something illiberal in the human heart,’ reading Orwell’s 1984 as secretly ambivalent about the tyranny it depicts rather than simply opposed to it.

On liberal tensions around immigration [§ On liberal tensions around immigration]

Sunstein grounds immigration policy in two liberal foundations that do not by themselves generate concrete policy: Kantian dignity (persons as ends, never merely means) and Millian utilitarian concern for well-being, including the well-being of those excluded or deported. He recounts visiting the southern border while working at DHS under Biden, and argues that ‘lawful pathways’ (expanded legal channels, e.g. the H-2B visa programme) plus effective deterrence (walls, technology, personnel) can enforce immigration law without inherent brutality — brutality is a contingent feature of how enforcement is carried out, not a necessary feature of enforcement itself.

On new generations of liberal thinkers [§ On new generations of liberal thinkers]

Asked whether liberal thought is regenerating below the age of fifty, Sunstein names economists (Ed Glaeser, John List, Esther Duflo, Sendhil Mullainathan) rather than philosophers as liberalism’s current active frontier, and suggests philosophy’s star system rewards slow-building careers (Rawls did not become ‘Rawls’ until well into his career) in a way economics, with its capacity for a single startling empirical finding, does not.

On Parfit, Mises, and Hayek [§ On Parfit, Mises, and Hayek]

Sunstein reads Derek Parfit as liberal principally through his search for convergence among Kantian, contractarian, and utilitarian foundations for the same conclusions (echoing Sunstein’s own ‘incompletely theorized agreements’), while conceding to Cowen’s pushback that Parfit’s anti-individualist, anti-self later work sits uneasily with liberal commitments — Sunstein likens it to Kahneman and Tversky’s early heuristics-and-biases method: exotic hypotheticals whose reliability outside familiar contexts is doubtful. On Mises, Sunstein admires the ‘cranky, ill-tempered, elbows-out enthusiasm for freedom,’ rawer and earlier than Hayek’s more systematic case. On Hayek, Sunstein rejects the central thesis of The Road to Serfdom — that government economic regulation on the New Deal model leads toward Stalinist or fascist tyranny — as false, while still calling the book one of the best of the last two centuries; he separately agrees Trump-era state equity stakes in private companies (lithium, Intel, rare earths) are a genuine bad sign, just not evidence for Hayek’s strong claim.

On Mill and liberal wokeness [§ On Mill and liberal wokeness]

Sunstein traces the intellectual origin of ‘woke’ to Mill’s The Subjection of Women — a canonical liberal text identifying the subordination of one group by another and the adaptive preferences that subordination produces. In Mill’s hands this is a genuinely liberal insight; Sunstein argues it curdles into illiberalism when it produces ‘arrogance of the social engineer’ — reflexive shaming and a refusal to listen to the people being described, on both sides of the subordination. He is careful to distinguish this failure mode from serious scholarship miscategorised as ‘woke’ (naming Catharine MacKinnon’s work as a case wrongly tarred this way).

On the value of Popper, Humboldt, Dewey and other liberal thinkers [§ On the value of Popper, Humboldt, Dewey and other liberal thinkers]

A rapid-fire named-thinker segment: Popper praised for method (how to determine truth) and freedom of inquiry; Dewey rated highly (‘great, great, great, great, great’); Edna Ullmann-Margalit — Sunstein’s late co-author — singled out as an underappreciated philosopher who did ‘the best work there is’ on the invisible hand, social norms, presumptions, and considerateness; Jeremy Waldron admired generally but Sunstein rejects his anti-judicial-review argument as wrongly treating the case against judicial review as a priori rather than contingent on the actual competence of courts versus legislatures.

On indigenous rights and liberalism [§ On indigenous rights and liberalism]

Sunstein is candid about the thinness of his own engagement here, citing Christine Webb’s The Arrogant Ape and Will Kymlicka as starting points, and frames the open question as: what would we need to learn from an unfamiliar society’s norms and internalisation of law that would actually revise liberal theory, as opposed to simply being worth celebrating on its own terms.

On the First Amendment rights of AI [§ On the First Amendment rights of AI]

Sunstein’s central doctrinal move: AI itself — like a toaster or a talking vacuum cleaner — has no First Amendment rights, but the humans engaging with it (a right to receive information) and the humans behind it (producers, co-authors) retain theirs, exactly as a film projector has no rights but the filmmakers behind a suppressed film do. This yields a working framework: a government ban on discussing a topic with an AI is a content-based restriction and presumptively invalid; a time-of-day access restriction is content-neutral and needs a justification but is not automatically barred; a university banning AI-assisted coursework is permissible because it targets academic-integrity conduct, not viewpoint. He extends this to human–AI co-authored output (his own books quote AI-generated text) and to open-source code as speech, with regulability turning on the government’s actual justification (national-security threat: plausibly regulable; protecting incumbent competitors: not).

On manipulation, libel, and nudging [§ On manipulation, libel, and nudging]

Sunstein separates falsehood (a false factual claim) from manipulation (inducing someone to bypass their own reflective capacities without necessarily saying anything false) and argues for a new legal ‘right not to be manipulated,’ modelled on the early-20th-century emergence of a privacy right — starting with egregious commercial cases like concealed recurring-payment terms. On libel, he argues the Sullivan actual-malice standard, while correctly protective of contestable political speech, leaves ordinary people with no cheap remedy for damaging, merely negligent falsehoods, and proposes allowing suits for a nominal sum plus a correction even without a showing of actual malice.

On trial by AI [§ On trial by AI]

Sunstein treats AI-run trials as ‘very thinkable’ in the near term but flags two barriers: algorithm/AI aversion (partly rational — context-specific human knowledge advantages — partly a bare preference for human involvement) and the Sixth Amendment jury-trial right, which would require constitutional amendment to displace. His own jury service (deliberately seated by a judge who wanted a jury-behaviour scholar on the panel, over the lawyers’ objections) illustrates his substantive point: juries bring an accuracy-independent ‘conscience of the community’ judgment that a purely predictive AI could not replicate, and that most people would not want replaced even if AI matched jury accuracy.

On favourite novels, movies, TV, and Bob Dylan [§ On Bob Dylan]

Closing register shift: A.S. Byatt’s Possession as favourite novel for its specificity about romance; Star Wars as the law-and-liberty film that has shaped his thinking; a close reading of Bob Dylan as liberal exemplar — self-invention (‘he not busy being born is busy dying’), rootlessness reframed as freedom rather than loss, and ‘Like a Rolling Stone’ read as an anthem of liberty rather than the bitterness Dylan himself once called it.

On animal rights [§ On animal rights]

Sunstein’s next book, Animals Matter, proposes a ‘bill of rights for animals’ and revisits a topic he abandoned after facing ‘ridicule and death threats’ during a Senate confirmation process over his pro-animal-welfare views. He also previews a planned book on disagreement, How to Disagree, built around what he calls the Rapoport Rules (via Daniel Dennett): restate an interlocutor’s argument so they would recognise and be grateful for it, credit what you have learned from them, and state what you agree with before criticising.