I think the flood of AI-driven litigation, benefits claims, and other citizen recourse to government services will be a first order factor necessitating the adoption of AI by government.

Clean tie between Fantasia and Nightmare Before Christmas

Will there be a summer version of this program? The (entirely understandable) FT requirement doesn't mesh with academic teaching schedules...

It would appear to be only the be-gin-ning.

Identity-based liability models are frustrated by pseudonymous/anonymous deployment and emergent harms. Bonding resources at the agent credentialing stage subject to penalization ex-post has certain advantages for these likely classes of agentic harms bit.ly/AgentPenalties

Policy certainty has value independent of the specific substance/valence which diminished uncertainty reveals.

I recently found this while moving to take a new role as Asst Prof at @uw_law and reflected on how if it weren't for this @CHPIllinois class with Tom Ulen in 2004 whether I'd even be teaching and writing in law and economics today. So sad to hear of Tom's passing, he was such a generous mentor and insightful scholar.

Privacy is all the more critical these days with automated enforcement tools like ALPRs, facial recognition, and AI assimilating it all into a comprehensive picture of our day to day lives in ways that likely infringe our 4th amendment rights. Great to see this focus on the part of @projectglitch_

The need to update coordinative rulesets springs eternal.

For a law review length piece on bonding/escrow for AI agents, see bit.ly/AgentPenalties (based heavily on the logic of slashing penalties on proof-of-stake networks)

Always great to talk w. @federicoast, especially b/c he hosted an event at @KingsCollegeLon where I presented an early draft of bit.ly/AIjudication (thanks to @TheIHS for support). Color me unsurprised yet delighted a dispute protocol like (@Kleros_io) is working in related areas!

here's my hot take on swarm governance - require bonded resources at any network credentialing stage that are then subject to automated penalties (w. penalty reversal on appeal). Should improve alignment and provide a recovery fund in cases of cascading emergent harms bit.ly/AgentPenalties

I have a project with F. Carugati exploring how sufficient acceleration in information technological change can erode the constitutional basis on which legitimacy and thus state capacity depend. I'd enjoy reading your draft when it's public.

Harms emergent from ai coordination need a tailored institutional response. Deterrence that operates in the same context in which agents act has advantages. Automated penalties on resources bonded at the credentialing stage have specific strengths here. bit.ly/AgentPenalties

Aren't those reflected directly in the utility functions of the cartel members? Unless we impose a subjective measure of merit, isn't expected utility maximization to any cartel member the rawest measure of "merit" within that cartel?

What would make a guild, once successfully established, NOT want to sort on merit (however defined) internally?

AI can't depoliticize judging. At the hard margins, applying AI to legal judgments will relocate the political choice from the courtroom to whoever designs and specifies the model. New project with @whlehr: three kinds of hard cases, why they frustrate algorithmic resolution, and where that residual authority goes when AI is used to resolve legal cases. bit.ly/AIjudication

Payment rails are crucial, yes. But an agent that pays can also harm, and an equally essential question for AiFi is what happens then. Crypto's lessons for agents aren't just payments, they're also slashing: stake up front, forfeited conditionally. Agents need rails AND staking.

Pseudonymous deployment and emergent harms from many agents interacting are ones traditional liability doctrines don't cover as cleanly as your example.

You need to lay off the kumquats, Jason.