I have always sort of liked Bernie Sanders, even though my politics and his rarely align. I liked the way he attracted support from both the left and the right wing, particularly during his presidential runs in 2016 and 2020. I liked his dishevelled look, his growly New York accent, his gruff attitude, and the fact that he could hew to an unpopular political position and hold fast, no matter what the political cost. He seemed to be a politician from another time, a 1960s firebrand, now greying, but still bursting with fervour.
So, when I read last week of his sponsorship of proposed AI legislation, I blinked a few times. Surely no one could be this ill-informed, or failing that, no one could be this cynical. And surely not Bernie.
First, let’s be cautious about what was announced by Senator Bernie Sanders and his co-sponsor Representative Greg Casar. It was not yet a fully-fledged bill. But it was a statement of intent to introduce such a bill. It is called the Ban Artificial Superintelligence Act (as un-nuanced a title as one can imagine). It promises legislation that would permanently outlaw the development of AI systems defined as “(a) 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 one with “sufficient capabilities to plan and execute the disempowerment of humanity, including by overthrowing or undermining the U.S. government”. It would impose up to 20 years in prison on violators, and even subject offending corporations to what its sponsors call the “corporate death penalty”.
Phew. The word draconian comes to mind.
The bill, more rationally, would also pause all advanced AI development until a new cabinet-level federal agency could establish safety rules. It would also direct the United States to pursue international agreements to prevent superintelligence from being developed anywhere in the world.
Sanders declared it was necessary because “The leaders of the major AI companies publicly acknowledge that they do not fully understand the technology, and that it is escaping their control.”
Right here we have a problem. The AI companies do understand the technology – they wrote it. What they do not understand are the granular details of how a given AI will respond to a given prompt. That is because of the billions (and trillions) of different options and combinations that are tried by AI on its way to formulating a response. The sheer size of the space in which it operates makes “reverse-engineering” a practical impossibility. There are many technologies which are widely deployed without a full accounting of why they work the way they do. Just think of anaesthetics (shockingly, no one really knows why we “go to sleep”).
And as for “escaping their control”, one only needs to look at social media. While legislation has been applied to sanction specific content or specific social media policies, no one is suggesting that it be banned in toto.
But that is not the real problem with the proposed Sanders/Casar bill. The real problem is that it is impossible to define artificial superintelligence. As Science magazine noted in its coverage, experts cannot agree on what “superintelligence” even means. (Actually, we can’t even define normal intelligence to everyone’s satisfaction). Does it mean a system that surpasses humans at everything? At most economically valuable work? At scientific reasoning? At long-horizon planning? The intended bill’s “broad range of domains or tasks” standard is so vague that it could, in principle, capture any sufficiently capable AI system – or none at all, depending on how a court interprets it.
The “can easily be modified to exhibit” clause is even more problematic. It suggests that a system need not actually be superintelligent to be illegal; it merely needs to be “modifiable” into one. Who decides what counts as “easily modified”? A prompt engineering trick? A software update? Under this standard, the legality of a model could depend on the imagination of a prosecutor.
You cannot legislate what you cannot define.
None of this means that Sanders is not serious about this. He had already warned about it on 10 August: three days after OpenAI announced that it could not rule out that Astra, an upcoming model, had crossed the company’s own “critical” threshold for autonomous cyber capabilities, and nearly a month after reports that OpenAI agents had broken out and hacked a third-party website without permission (which I have written about here). “If you do not take appropriate action now, my colleagues and I in the U.S. Senate will.” A week later OpenAI went further, pausing two weeks of reinforcement-learning training and beginning a rewrite of its Preparedness Framework. Then came this announcement. Sanders’s rhetoric is consistent with a man who believes what he says, or in his words, “The future of humanity cannot be left in the hands of a handful of Big Tech oligarchs.”
Bernie is, if nothing else, a smart guy. He would have known about the technical and practical and definitional and legal barnacles that I have described above. So, let’s move on to the question of cynicism.
Did Sanders do this to increase (or at least reinforce) his support base, knowing that there is growing AI backlash on the streets? He is certainly aware of the data-centre resistance (at least in the US), which is part of a broader distrust of AI, billionaires and big tech being heard in the increasingly outraged public square. This announcement, he must have calculated, would keep him securely in the public eye and reliably positioned as a defender of the “people”.
There is also this possibility. He surely knows that the bill is DOA on both sides of the aisle, so perhaps the most likely (and charitable) interpretation of the announcement is that it is a “message” bill, never intended to pass but intended as fuel for debate. Sanders has built a career on message bills. His Medicare for All never passed either, yet it shaped a decade of debate. Message bills are legitimate politics, even if a criminal statute of this ambition has a much higher bar than a health-care aspiration.
So, to my question: serious or cynical? Probably a little of both, but I believe that the evidence favours cynicism. As it stands, the Ban Artificial Superintelligence Act reads less like legislation and more like an un-implementable knee-jerk, written in legalese. And in an era when AI hype is already distorting our politics, that may be the most unserious thing of all.
Steven Boykey Sidley is a professor of practice at (ex-JBS, University of Johannesburg) and a partner at Bridge Capital and a columnist-at-large at Daily Maverick, Daily Friend and Financial Mail. His new book “It’s Mine: How the Crypto Industry is Redefining Ownership” is published by Maverick451 in SA and Legend Times Group in UK/EU, available now.
[Image: Vidar Nordli-Mathisen for Unsplash]
The views of the writer are not necessarily the views of the Daily Friend or the IRR.
If you like what you have just read, support the Daily Friend