Could generative AI turn out to be the tech industry's Vietnam? And could public backlash lead AI to a better path?
Gary Marcus from Marcus on AI · email · 2026-05 source ↗ Reader 🖥️
A growing public backlash against generative AI — analogous to the anti-Vietnam movement — is poised to force a political repricing of America's multitrillion-dollar AI bet and can, if organized, redirect policy toward FDA-style pre-deployment regulation.
practical · argumentative — Marcus is arguing for action — that organized public backlash can and should push policy (Trump administration, regulators) toward FDA-style preflight checks on AI. The Vietnam analogy is rhetorical scaffolding for a practical call to mobilize.
Structure
| Part | Role |
|---|---|
| Opening observation: commencement speakers (including Eric Schmidt) booed for mentioning AI; Calacanis's tweet drawing a Vietnam parallel | Anecdotal evidence that backlash is real and accelerating |
| Extension of the Vietnam analogy: huge investment, arrogance, no results, ~$1T cost | Argument by analogy — maps Vietnam features onto AI |
| Empirical aside: hallucinations, unreliability, misalignment, poor pilot RoI | Evidence backing the 'mistake' claim independent of the analogy |
| Self-citation: January 2026 Politico 'Black Swan' prediction that Trump would distance himself from pro-AI policy by end of 2026 | Credentialing — Marcus positions himself as having called this early |
| News hook: Trump now 'actively considering' AI preflight checks of the kind Marcus and Rempel Garner proposed three years ago | Evidence that the predicted political turn is underway |
| Call to action: re-read Taming Silicon Valley; 'if the people unite' AI can be steered to a better place | Practical conclusion / mobilization |
Key terms
- Backlash AI Backlash
An organic, broad-based public revulsion against AI — visible in booing of AI boosters, comment-section sentiment, and shifting political winds — that Marcus treats as a coherent political force capable of moving administrations - AI preflight check AI Regulation & Governance (External)
An FDA-analogous regulatory regime in which major new AI applications must be evaluated for safety, with carefully delineated evidence, before public release - Arrogance-fueled mistake
A pattern in which elites pour resources into a campaign despite mounting evidence of failure, sustained by institutional pride rather than performance — Marcus's frame for both Vietnam and the current AI buildout - Mythos Mythos
Apparently a recent AI incident or product whose fallout is, in Marcus's reading, a proximate cause of Trump's policy reconsideration (referenced but not explained in this piece)
Propositions
- explicitcorroboratedPublic backlash against generative AI is growing fast and is now visible enough that AI mentions trigger booing of prominent figuresconf: highsupports ←Presentation: A Few Predicted Talks From QConAI 2030 · p16 — Both independently document fast-growing, visible public backlash against AI as the leading indicator behind predictions of political/regulatory consequences.
- explicitThe current AI buildout structurally resembles the Vietnam War: massive investment, arrogance, no working results, ballooning costconf: high
- explicitGenerative AI's hallucinations, unreliability, and misalignment persist even after massive investment, indicating the investment is not paying offconf: high
- explicitMost AI pilot studies show poor or negative return on investmentconf: medium
- explicitcorroboratedTrump will distance himself from aggressively pro-AI industry policies by end of 2026, and the move has already begunconf: highsupports ←The Trump administration's AI doomer moment · p1 — The administration reviving Biden-era pre-release review it previously ridiculed is a confirming instance of Marcus's prediction that Trump will distance himself from aggressively pro-AI policy.
- explicitdevelopedFDA-style preflight regulation is the right regulatory model for major AI applicationsconf: highsupportsSebastian Mallaby on AI Safety and the Race for Superintelligence · p9 — Marcus independently converges on Mallaby's call for an FDA-style government pre-release assessment body for major AI applications.refines ←Why OpenAI and Anthropic Are Becoming Consultants · p11 — Qualifies the FDA-style preflight proposal: current safety benchmarks are so inadequate that meaningful regulatory vetting criteria may be impossible to write.
- explicitIf the public organizes ('the people unite'), AI policy can be redirected to a better pathconf: high
- implicitdevelopedDiffuse cultural sentiment (booing, comment-section anger) reliably translates into administration-level policy reversalconf: mediumrefines ←The Trump administration's AI doomer moment · p8 — Locates the mechanism of policy reversal in concrete capability demonstrations (Mythos) rather than Marcus's diffuse cultural backlash.
- implicitVietnam-era student protest was decisive in ending US involvement in Vietnam — making student/young-public sentiment a proven lever against entrenched elite projectsconf: high
- implicitThe Trump administration is responsive to the same kind of populist backlash that ended Vietnamconf: medium
- implicitThere exists a discrete frontier between 'major' AI applications (warranting preflight checks) and ordinary ones (not warranting them)conf: medium
Arguments
Generative AI is heading toward, or already is, an arrogance-fueled trillion-dollar failure analogous to Vietnam strength: low
Explicit premises:
- Vietnam was a massive, expensive, arrogance-fueled failure that continued despite evidence of failure
- Generative AI involves multitrillion-dollar spend with persistent hallucinations, unreliability, and misalignment
- Most AI pilots show poor RoI
Implicit premises:
- The features Marcus selects (cost, persistence-despite-evidence, arrogance) are the load-bearing features of Vietnam-as-disaster (not e.g. body count, draft, geopolitics, civilian casualties)
- AI's reliability problems are structural, not transitional engineering challenges that scale or technique will resolve
- Pilot-study RoI is the right measure of AI's economic value (rather than e.g. infrastructure-buildout returns or long-horizon platform plays)
Inferences:
- If the analogy holds, the AI buildout will end abruptly and humiliatingly rather than gradually plateau
- Investors who pull out early — like Vietnam-era doves — will be vindicated; current AI bulls will be the McNamaras of this cycle
Organized public backlash can redirect AI policy toward FDA-style preflight regulation strength: medium
Explicit premises:
- Backlash is already visible (booing, comment sentiment)
- Trump has begun to reconsider his pro-AI stance and is 'actively considering' preflight checks
- Marcus's January prediction of a Trump pivot is now playing out
Implicit premises:
- Cultural backlash translates into policy via political incentives that operate on Trump specifically
- Trump's reconsideration is caused by backlash rather than by, e.g., factional fights inside the administration, Mythos-specific liability fears, or geopolitical recalibration
- Regulators have the capacity and political authority to implement FDA-style preflight checks on a fast-moving software domain
Inferences:
- The window to lobby for preflight regulation is now, while sentiment is hot
- Failure to organize would let the buildout continue despite the backlash
Trump's policy reversal on AI is already underway strength: low
Explicit premises:
- Marcus predicted in January 2026 that Trump would distance himself from pro-AI policy by end of 2026
- Trump is now 'actively considering' AI preflight checks
- This matches the regulation Marcus and Rempel Garner proposed three years ago
Implicit premises:
- 'Actively considering' equals 'beginning to distance himself' rather than tactical hedging
- Public reporting that Trump is considering preflight checks is reliable signal of intent
- Marcus's prior call for FDA-style regulation establishes the framework Trump is converging on (rather than a parallel, independently arrived-at idea)
Inferences:
- If the prediction is confirmed, Marcus's broader Taming Silicon Valley framing gains credibility
- Other Marcus predictions in the Politico Black Swan piece deserve more weight
Narratives
- Commencement speakers including Eric Schmidt booed for mentioning AI
→ Carries the proposition that backlash has crossed from online discourse into mainstream public ritual — without Marcus having to argue it formally - Vietnam War as trillion-dollar arrogance-fueled disaster that continued despite evidence
→ Implicit argument that AI is in the same category — and that history's verdict will be similarly harsh - Marcus's own January 2026 Politico 'Black Swan' prediction proving correct
→ Establishes Marcus as someone whose contrarian calls land — implicitly authorizing the rest of his predictions - Marcus and Rempel Garner proposing FDA-style AI regulation three years ago, now being adopted
→ Frames Marcus as policy-prescient — and the FDA-analogy as the obvious endpoint rather than one option among many
Questions addressed
- ✅ Is the current anti-AI sentiment structurally similar to a major historical backlash?
Yes — Vietnam: arrogance, expense, persistence despite failure - ✅ Will Trump shift away from pro-AI policy?
Yes, and it has already begun — sooner than Marcus predicted - ✅ What regulatory regime should follow?
FDA-style preflight checks on major AI applications - ⭕ What must the public do for the backlash to succeed?
'If the people unite' — but no organizing model, target, demand-list, or measurement of success is specified - ⭕ Through what mechanism does diffuse public sentiment become administration policy?
Treated as automatic; not theorized - ⭕ What counts as a 'major' AI application requiring preflight review, and who decides?
Unaddressed — the FDA analogy papers over the boundary problem - ⭕ Where does the Vietnam analogy break down?
Marcus doesn't stress-test it — no draft equivalent, no body count, AI has paying users and visible (if inflated) utility, and AI is a private capital phenomenon as much as a state one
Gaps
- incompleteThe mechanism connecting cultural backlash to specific policy outcomes is never theorized. Marcus moves from booing commencement speakers to Trump considering preflight checks as if the chain were obvious, but the political-economic transmission is the load-bearing claim of the entire piece.
Read next: Daniel Drezner on ideas and policy; Frank Baumgartner & Bryan Jones on punctuated equilibrium in policy change; histories of how the Vietnam anti-war movement actually influenced Nixon-era policy (Wells, The War Within)closed byThe Trump administration's AI doomer moment — Supplies an empirical instance of the missing backlash-to-policy transmission: data-center backlash from the administration's own base plus fear of blowback from AI-enabled cyberattacks produced an actual policy reversal. - incompleteThe Vietnam analogy is asserted, not stress-tested. Marcus picks the features that map (cost, arrogance, persistence) and ignores the ones that don't (conscription, mass casualties, AI's voluntary adoption, paying customers, dual-use civilian-military structure, capital-market vs. state-budget funding). A reader who respects the contrarian impulse still needs the analogy's load-bearing components separated from its rhetorical ones.
Read next: Khong, Analogies at War; Neustadt & May, Thinking in Time — both on how policy analogies are used and misused - incompleteThe FDA-style preflight proposal is treated as obviously correct, but the analogy to drug approval is itself contestable: AI systems update continuously, have probabilistic outputs, are not single-purpose, and a pre-market gate sits uneasily with software practice. The boundary between 'major' and ordinary applications is undefined.
Read next: Bommasani et al. on foundation-model governance; EU AI Act risk-tier debates; Helen Nissenbaum on contextual integrity as an alternative framing - incompleteMarcus equates 'pilot study RoI is poor' with 'AI investment lacks RoI' without distinguishing infrastructure plays (compute, model training as platform) from end-user productivity studies. Hyperscaler capex logic is not pilot logic.
Read next: Erik Brynjolfsson's productivity-paradox work; recent Goldman/MIT/McKinsey AI ROI studies in tension with each other - illogicalThe piece performs a credentialing move — citing Marcus's January prediction as evidence the current Trump pivot is real and meaningful — but a prediction confirmed by a single ambiguous data point ('actively considering') is weak confirmation. The structure flatters the prediction more than it tests it.
Read next: Tetlock, Superforecasting, on how to evaluate prediction confirmation properly - incompleteMarcus says 'if the people unite' but offers no theory of organization — no target institution, no concrete demand beyond preflight checks, no analysis of who the coalition partners are. Compared to Vietnam-era movements (SDS, draft resistance, veterans' organizations), the anti-AI backlash has no described organizational form.
Read next: Hahrie Han on organizing vs. mobilizing; Zeynep Tufekci, Twitter and Tear Gas, on the fragility of networked movements without organizational depth
Related documents
- Sebastian Mallaby on AI Safety and the Race for Superintelligence
- The growing AI backlash
- The Trump administration's AI doomer moment
- AI Companies Aren't Evil. But They Are Reckless.
- AI's biggest critic has lost the plot
- Travel Through the Lens of AI with Booking.com CEO Glenn Fogel
- Everyone I recommend on AI is optimistic. Except Cory Doctorow.
- No, Anthropic did not call for a pause on AI development
- Three reasons to think that the Claude Mythos announcement from Anthropic was overblown
- NHS England rushes to hide software over AI hacking fears