Amid the nationwide backlash to AI via data center bans, one Indiana town has moved to prohibit "artificial super intelligence" (ASI).
asserted
town → move → intelligence
Last week, the Bloomington City Council voted unanimously to pass an ordinance that "prohibits the distribution and deployment" of ASI.
asserted
that → vote → ASI
The ordinance, which was backed by the Safe AI Forever Institute, defines ASI as a "system or entity capable of significantly outperforming humans on essentially all cognitive tasks" that can operate without "continuous human oversight."
asserted
that → back → oversight
Covered models are prohibited from being activated by any user interface within the borders of Bloomington, regardless of where the underlying developer is located.
asserted
developer → prohibit → Bloomington
Students at nearby Indiana University won't be booted off of ChatGPT just yet, as the city is yet to designate any models "superintelligent."
asserted
models → boot → ChatGPT
Councilmember Dave Rollo introduced the measure in a memo to the Common Council, the City Council's legislative body.
asserted
Rollo → introduce → Council
He cited concerns from "AI experts" and developers over the technology's potential to cause "catastrophic outcomes" for humanity, including human extinction.
asserted
He → cite → extinction
"The level of risk is simply unacceptable," wrote Rollo, referencing "many AI experts" who placed the probability of extinction between 50 percent and more than 99 percent.
asserted
who → write → percent
Rollo called the policy an "interim public safety parameter…during a period of state and federal legislative inaction," but the requirements for lifting the ban tell a different story.
asserted
requirements → call → story
According to Rollo's memo, the ban will last until the state of Indiana enacts a "comprehensive framework for Provably Safe Superintelligence," after which it would expire after 91 days.
uncertain
it → accord → days
Otherwise, it will stay in place until the council determines a "broad scientific consensus that the development of Superintelligence shall be done safely and controllably."
asserted
development → stay → Superintelligence
To the council, that would have to be proven by a probabilistic risk assessment that finds 99.9999 percent certainty of safety, the standard used for nuclear power plants.
asserted
that → have → plants
"This is an impossible standard to meet," Adam Thierer, a senior fellow at the R Street Institute, tells Reason.
asserted
Thierer → meet → Reason
Second of all, there's nothing in this world that's 99.999 percent safe.
asserted
that → be → world
"It's likely that this would stay in place pretty much forever once it's put on the books," Thierer says.
asserted
Thierer → stay → books
As introduced, the ordinance would have allowed the ban to be lifted if 95 percent of the city's residents voted to strike it down; however, that condition was removed before the council's vote.
asserted
condition → introduce → vote
Other cities, counties, and states have implemented their own restrictions on specific AI-related technologies, including automatic license plate readers, AI pricing, and data centers.
asserted
cities → implement → readers
What distinguishes AI models from license plate readers or data centers is that they transcend state boundaries instantly, all the time.
asserted
they → distinguish → boundaries
Therefore, developers can't easily tailor a large language model to an individual jurisdiction, and local policies can affect models far beyond their borders.
asserted
policies → tailor → borders
That appears to be the goal of Bloomington's policy.
asserted
That → appear → policy
The ordinance, Rollo writes, "could set a precedent and a template for other cities to follow.
uncertain
cities → write → precedent
Over 100 million people live in U.S. cities of Bloomington's size or greater.
asserted
people → live → size
If those (approximately 350) cities passed similar ordinances in the coming weeks or months, this would create a massive regulatory burden for companies creating ASI."
asserted
this → pass → ASI
The memo also says that the SAFE AI Forever Institute and other nonprofits will continue to lobby for these restrictions in other municipalities.
asserted
Institute → say → municipalities
Their strategy, then, is to create a regulatory regime through municipal policies that would effectively govern frontier AI models across the nation.
asserted
that → create → nation
Some states have employed a similar playbook, including California and New York.
asserted
states → employ → California
A patchwork of different state policies is already confusing, but layering on hundreds of municipal policies would create "a compliance hell," Thierer tells Reason.
asserted
Thierer → layer → Reason
"And I think that most of the sponsors not only realize that, but now understand that's the point."
asserted
that → think → that
The ordinance could face legal challenges.
uncertain
ordinance → face → challenges
The Dormant Commerce Clause restricts state laws that unfairly hinder interstate commerce, even in the absence of federal laws.
asserted
that → restrict → laws
If states don't have the power to regulate interstate commerce, then counties surely shouldn't either.
asserted
counties → have → commerce
Thierer also warns that the measure could attract First Amendment lawsuits due to its restriction of people's "ability to learn about these technologies or utilize them."
uncertain
measure → warn → them
No lawsuit has yet been filed.
asserted
lawsuit → file → ?
The ban could be particularly damaging to Indiana University, which is home to the Luddy Artificial Intelligence Center, with hundreds of researchers and students dedicated to studying AI.
uncertain
hundreds → dedicate → AI
Beyond that, AI fluency is quickly becoming essential for providing value in any field.
asserted
fluency → become → field
That now sits at the mercy of the Bloomington City Council.
asserted
That → sit → Council
It seems that these policies won't end in Bloomington.
asserted
policies → seem → Bloomington
"It could be that we're seeing the development of a new faction of the AI safety movement within the Safe AI Forever Institute and others included that are basically out to create this expanded vetocracy at the municipal level," Thierer tells Reason.
uncertain
Thierer → see → Reason
"Just as a lot of people have been showing up at town hall meetings and county planning meetings and zoning ordinance hearings to protest data centers, you could see the same thing now but for AI models."
uncertain
you → show → models
The Aporia
