In an ideal world it’s certainly possible that the U.S. federal government could implement meaningful guardrails for AI development that protect innovation, national security, consumer privacy, public safety, and labor/creatives, while preventing market monopolization by the sector’s biggest players.
But we don’t live in that ideal world. We live in a country where an historically corrupt autocrat with a bad comb-over has disemboweled all of our public interest regulators and Congress is too corrupt to function. Thanks in no small part to a tech sector bankrolled by people who openly hate democracy.
With Chinese AI development resulting in a surge of significantly cheaper options that threaten U.S. dominance, and a bipartisan coalition of people violently opposed to data centers (or justifiably terrified by the unsubtle lack of ethics in big tech), it’s extremely likely that 2027 sees some form of AI regulation.
It’s also extremely likely these new laws are going to either be terrible and ghost written by the tech industry’s biggest players — cementing and legalizing all of their worst impulses and locking in their dreams of cloud software domination. Or terrible and written by people who think they’re helping, but have no idea how any of this works, resulting in bad proposals that confuse people and don’t pass.
Enter Bernie Sanders, who on Tuesday attended a conference alongside authoritarian bigot Steve Bannon, apparently to capitalize on the growing bipartisan animosity toward data centers and unethical tech giants. Sanders also recently announced he was introducing new legislation that would suspend AI development and “ban superintelligence”:
“The Ban Artificial Superintelligence Act would permanently ban the development and deployment of superintelligent AI and temporarily pause advanced AI development until a federal regulator has established safety rules. It would also direct the U.S. to pursue international agreements to prevent superintelligence from being developed anywhere in the world.”
As a whole I’m bored with Dem lawmakers proposing legislation they know won’t pass to convey the idea they’re “doing something” (Elizabeth Warren loves this). The argument is that this at least “starts a conversation” about what a real law would look like, but I’m not sure that’s true. We spent the last two decades endlessly talking about the need for privacy laws and data broker regulation, only for corruption to cripple every last effort.
I also don’t really love Dem lawmakers thinking the best way to reach Trump voters is to appear on stage with fascists, lending fascists tech policy credibility they don’t have (it’s what resulted in years of people deluding themselves that you could partner with fascists on antitrust).
That’s not to say I don’t think people like Sanders’ shouldn’t propose new laws. And maybe the final law actually has some salient teeth, whenever he finally unveils it. But I do think those proposals need to be very simple and extra well considered and crafted to gain any meaningful momentum in a sea of hysteria, noise, corruption, and hypercapitalism.
Sanders’ short summary tries to define “superintelligence” this way:
- An artificial intelligence system that exhibits or can easily be modified to exhibit
capabilities that match or exceed human cognitive performance and capabilities
across a broad range of domains or tasks.- AI systems that have sufficient capabilities to plan and execute the disempowerment of humanity, including by overthrowing or undermining the U.S. government.
Technically all sorts of software could be “easily modified to exhibit capabilities that match or exceed human cognitive performance across a broad range of domains or tasks.” You could argue your current spreadsheet or music software can do that.
Recent videos suggest Sanders may not really understand how AI works, which is a problem if you’re going to be proposing major tech policy. It’s also pretty clear Sanders is one of many people freshly spooked by software superintelligence they don’t understand and can’t actually define.
The last week or two have been flooded with a lot of young AI industry folks, high on their own science fiction farts and spooked by the rapid pace of software development, publicly insisting that modern software automation has become self-aware and superintelligent and has a not insignificant chance of simply deciding to wipe out humanity (this is, to be clear, not true).
These folks aren’t necessarily worried that AI is going to be used by rich people to ratfuck creatives and labor (it is), or automate a lot of flawed human systems in a way that provide corporations accountability cover from heartless and greedy systemic policy (also true; see: AI adoption in Medicare rejections).
These guys are out here insisting, falsely, that software has achieved human-level sentience and will, on a lark, simply decide to decimate the eastern seaboard for fun. This is, again, not a thing that’s actually true or happening, not that it matters to the ad-engagement based clickbait press. A lot of these folks are also coming to the party very late, after they’ve been well paid and the horses are all out of the barn.
The worries are driven by the very real threat of agentic AI hacking at incredible new speed and scale (see: Hugging Face). While those worries are real, in the process a lot of these folks have anthropomorphized software in a way that’s just not accurate, misrepresented what’s happening, and are very clearly terrifying people in a way that misdirects attention away from the more realistic but mundane harms of AI.
Many of the early issues with automated hacking software doing precisely what it was programmed to do have involved very basic human errors, like failing to double check that a third-party vendor properly isolated your agentic hacking swarm from the internet. Or failing to monitor your automated hacking software in real time. There are problems, but they’re human problems, not HAL 9000 type problems.
Some of the actual human problems could simply be addressed by the U.S. actually enforcing antitrust law without treating it like “anti-business” or “extremism.” Or funding and staffing the regulators we currently have. Or finally passing simple privacy and campaign finance reform laws. It often feels like we’re trying to build new bridges without first ensuring the ground is stable.
Sanders’ bill also proposes establishing a new cabinet-level federal agency to “safeguard the public from the dangers of artificial intelligence.” Ideally this agency would be staffed with objective experts with genuine public interest bonafides, though historically that’s not how the corrupt U.S. works (see: the FTC, FCC, or EPA, where genuine public interest officials, like Gigi Sohn, can’t pass a corrupt congressional nomination process).
Sanders’ bill also proposes a total “pause” on all development while we figure things out:
Pausing advanced AI development until a new, federal AI regulatory body is up and running and has established clear rules and model review process to ensure safe and secure development and deployment of AI.
Obviously none of this is happening. Congress is too corrupt to pass meaningful public interest legislation surrounding privacy, gun rights, health care, or much of anything else. Into that intentional vacuum will step the wealthiest tech executives, who are dead set on ghost writing their own AI legislation in 2027.
All of the top AI executives this week responded to the hyperbolic “Skynet will kill us all” news cycle with claims they want to voluntarily slow the rate of AI development. I tend to think that given the dawning realization that none of this stuff is profitable, they’re preparing investors and the press for slowed CAPEX ahead of a market correction caused by their own hype and funny math.
But the idea that big tech companies are going to create a self-regulating body that limits their pursuit of wealth and power is silly. In reality, they’re going to flood Congress in the new year with all sorts of bad laws that ban cheaper Chinese alternatives, open source models, open weighted models, on device models, and anything that imperils their dreams of walled garden software automation domination.
I don’t pooh-pooh Sanders for at least trying to get the ball rolling on AI legislation not crafted by the planet’s shittiest oligarchs, but if we’re going to have alternative visions to the shitty AI laws tech giants dream up in the new year, they’re really going to need to be exceptionally well crafted and not driven by vibes. They also can’t, you know, inadvertently ban the entirety of modern software.