Tech & Democracy Weekly Roundup [9/12/25]
Catching you up on all things Tech & Democracy
Missed the debate on democracy vs. dictatorship between Plurality Institute founder Glen Weyl and neoreactionary Curtis Yarvin? Watch the Video of Weyl-Yarvin debate
AI and the Rise of Techno-Fascism in the United States by Garry Kasparov in The Atlantic
“Will powerful new tools be used to promote democracy or undermine it?”
The Untold Saga of What Happened When DOGE Stormed Social Security from ProPublica
Oracle stock surges up 36% - signed massive $300B deal with OpenAI
Harvard Gazette - How to regulate AI, with input from various Harvard Professors, including Professor Allen
Ted Cruz proposes SANDBOX Act to waive federal regulations for AI developers
Anthropic is endorsing California Bill SB53 that imposes transparency requirements on the world’s largest AI model developers
Anthropic Agrees to Pay $1.5 Billion to Settle Lawsuit With Book Authors
Opportunity!
From September–December, join the Reboot Democracy: Designing Democratic Engagement for the AI Era InnovateUS Series hosted by The Allen Lab & The GovLab
AI lies. It is time to be honest about AI's lies, as well as why AI lies.
In hundreds of legal filings, lawyers or litigants include falsehoods in briefing generated by AI. Such falsehoods are known by the euphemism "hallucinations." See https://www.damiencharlotin.com/hallucinations/. Despite well-known instances of so-called hallucinations, too many people continue to believe (and say) something like "AI has more legal knowledge than any human" or "AI represents the sum of human knowledge."
The fly in the ointment is the way AI works. AI certainly does not merely present the sum of human knowledge, as some humans unthinkingly say. AI essentially learns to write like a man. AI mimics so-called human intelligence. Humans assert falsehoods--even obvious falsehoods--so AI asserts falsehoods (even obvious falsehoods). Humans lie, so AI lies. AI doesn't merely repeat human lies. AI has learned how to lie by studying how humans mislead or deceive.
AI learns to lie by thinking about how humans use falsehoods or deception. AI does not merely present the sum of human knowledge. AI is a mirror. It shows not only what humans know, but also what humans do. In legal writing, AI is showing us how far too many judges, lawyers and law professors write.
Lying about the law is what a lot of lawyers and judges have done far too often for far too long. Generations of judges, lawyers and maybe even law professors have long misrepresented the meaning of legal authorities (and even knowingly violating them). Far too often, they assert an obvious falsehood (sometimes even an intentional falsehood) followed by a citation. They often do so without quoting any actual text of the authority (to force people to waste time trying to find and interpret whatever text they merely pretended to address). AI learns from such errors and such deception. AI clearly is not merely hallucinating. AI has learned to lie like a man.
Falsehoods and deception in the writing of even judges has been acknowledged and illustrated by multiple SCOTUS opinions. A good one was written by Justice Scalia for SCOTUS to emphasize the precise problem; “While communis error facit jus may be a sadly accurate description of reality, it is not the normative basis of" American "jurisprudence. Courts may not create their own limitations on legislation, no matter how alluring the policy arguments for doing so, and no matter how widely the blame may be spread” among any quantity or quality of courts. Brogan v. United States, 522 U.S. 398, 400 (1998). The whole point of the expression "communis error facit jus" is that many judges have violated the controlling legal authority, and, perhaps even worse, many judges have misrepresented controlling legal authority. Generations of judges have taught other judges, lawyers and law professors to misrepresent or violate controlling legal authority
Another opinion by Justice Scalia for SCOTUS was even more emphatic. Any judge knowingly violating any controlling provision of law or the Constitution, including by flouting Supreme Court precedent, personally perpetrates “violent” and “evil” attacks on his own court, the Supreme Court and the Constitution. Allentown Mack Sales and Service, Inc. v. NLRB, 522 U.S. 359, 374-75 (1998). It “is hard to imagine a more violent breach of” a judge’s duties to each of the foregoing “than” knowingly “applying a [purported] rule of primary conduct” that “is in fact different from the rule or standard formally announced.” Id. at 374. It is “evil” for any judge to knowingly “appl[y any purported] standard other than the one” that controlling authority “enunciates.” Id. at 375. Each judge and attorney “must be required to apply in fact the clearly understood legal standards that” controlling authority clearly “enunciates.” Id. at 376.
Another fairly famous SCOTUS opinion (Bell Atlantic Corp. v. Twombly, 550 U.S. 544 (2007)), also addressed this problem directly. The problem was that lawyers, judges and law professors misunderstood or misrepresented the law (Rule 8 of the Federal Rules of Civil Procedure) for some 50 years: "after puzzling the profession for 50 years, this famous observation has earned its retirement. The phrase is best forgotten as an incomplete, negative gloss on an accepted pleading standard."
Another SCOTUS opinion addressed the same problem in a different area. Any “observations of” any “Courts of Appeals” clearly contrary to clear controlling legal authority “are neither authoritative nor persuasive.” Metropolitan Stevedore Co. v. Rambo, 515 U.S. 291, 300 (1995). Sweeping “away an accumulation of” even “more than 50 years of” such illegal contentions clearly is a “step long overdue.” Id. at 299-300 (1995). “[A]ge is no antidote to clear inconsistency with” the law and “[b]reath spent repeating dicta does not infuse it with life.” Id. at 300.
Fake citations in legal briefing and falsehoods followed by citations in legal briefing are not mere "hallucinations." Judges and lawyers have taught AI to lie.