Senator Bernie Sanders of Vermont and Representative Greg Casar of Texas have introduced legislation that would prohibit the development of artificial superintelligence, according to a report from Democracy Now!
The proposal, titled the “Ban Artificial Superintelligence Act,” marks one of the most aggressive congressional attempts yet to place hard limits on the frontier of artificial intelligence research. Rather than focusing on disclosure requirements, safety testing or licensing regimes â the approach favored by most AI bills introduced in recent years â the measure seeks to draw a line that developers would not be permitted to cross at all.
What “superintelligence” means
The term generally refers to a hypothetical AI system whose capabilities exceed those of humans across essentially all cognitive tasks, not merely in narrow domains such as chess, protein folding or code completion. Several leading AI laboratories have publicly described building systems of that kind as an explicit long-term goal, a stance that has drawn alarm from a range of researchers who argue that humanity has no reliable method for controlling a system substantially smarter than its creators.
That concern has produced an unusual coalition. Warnings about advanced AI have come from computer scientists who helped build the field, from labor organizations worried about mass displacement of workers, and from civil society groups focused on surveillance, disinformation and the concentration of power in a handful of technology companies.
An unusual pairing of arguments
Sanders, an independent who caucuses with Democrats and has long been among the Senate’s sharpest critics of corporate consolidation, has repeatedly raised questions about what automation means for American workers and about the extraordinary wealth being accumulated by a small number of technology executives. Casar, a progressive Democrat from Austin, has built much of his public profile around labor issues and workplace protections.
Both themes run through the case for a ban. Supporters of such limits argue that the decision to build systems capable of replacing human judgment at scale is being made by a few private firms, with little democratic input and enormous financial incentive to move quickly.
Steep odds in Congress
The bill faces a difficult path. Congress has struggled for years to pass even modest technology regulation, and the AI industry has assembled one of the largest lobbying operations in Washington. Critics of a prohibition are likely to argue that “superintelligence” is too vague a category to legislate against, that a ban would be difficult to enforce, and that restrictions imposed only in the United States could push research offshore to jurisdictions with weaker oversight â including geopolitical rivals.
Supporters counter that the same arguments were once made about chemical weapons, human cloning and nuclear proliferation, areas where international prohibitions and norms have had at least partial effect. They also note that a legal definition problem is not the same as an impossibility, and that the alternative â waiting until such systems exist before deciding whether they should â offers no second chance.
Even if the legislation stalls, its introduction may shift the terms of the debate in Washington. Most AI policy discussion to date has concerned how advanced systems should be built and deployed. A bill asking whether some systems should be built at all pushes a question that has largely lived in research papers and industry blog posts onto the congressional record.
Neither lawmaker’s office has indicated what committee action, if any, the bill is expected to receive. Read More
Leave a Reply