Advancements in the field of artificial intelligence have prompted many concerns about the possible disruptive effects this technology will have on everything from employment to war. It also raises many difficult questions about the role that the government should have in regulating AI and managing its effects on our society.
In past columns I’ve argued that policymakers will have to take a proactive approach to regulation in order to avoid the worst of the predicted outcomes. That is what Congressman Jay Obernolte, R-California, and Congresswoman Lori Trahan, D-Massachusetts, are attempting to do with their newly released draft of their Great American AI Act (GAAIA).
GAAIA is a comprehensive effort to federally regulate the development of AI models by implementing harm mitigation standards and enforcing them through independent auditing. The act would also create a framework for researching workforce disruptions and laying the foundation for retraining and assistance programs. Notably, GAAIA would preempt state laws that regulate AI development for three years or possibly more if it is reauthorized.
I recently spoke to Congressman Obernolte to discuss the extent and limitations of GAAIA in its current form. We should note that the current bill is a discussion draft that invites comment from the industry and other policymakers.
In its current form, GAAIA would produce a framework to position the federal government to be able to make informed decisions about labor force disruptions. For example, Section 252 would create reports about labor sectors that are facing threats from AI and Section 255 would produce a study exploring different design options for a program that would provide assistance for displaced workers. Overall, GAAIA is absent of any protections for workers from being displaced in the first place.
According to Obernolte, this was by design.
“It’s not the role of the federal government necessarily to prevent workforce disruption and displacement,” said Obernolte. “Technological revolutions have always been disruptive. Most recently, the internet, but going all the way back to the printing press. But they have always created more jobs for humans than they have displaced. And we are convinced that that’s going to be the case for AI as well.”
Obernolte stated that the federal government should support workers with financial assistance and retraining but denied that the government should intervene in possible disruptions.
Clearly, there are a few important concerns with this approach. The first has to do with whether we’re justified in sharing Obernolte’s optimism that AI will create more jobs than it eliminates. The second has to do with the possibility of successfully training people in vulnerable sectors to use AI to increase their productivity.
The data that we have available to think through these concerns is conflicting, and the uncertainty surrounding future AI and robotics advancements makes it all the more difficult to assess. A Boston Consulting Group analysis predicts that 10-15% of jobs in the US could be eliminated by 2031 while a survey by Morgan Stanley revealed that some firms had already eliminated 15% of their workforce via AI. Other studies find negligible impacts on total jobs numbers.
I asked Obernolte, “So what have you found about the prospects for widespread AI training? And since we’re not protecting workers from getting laid off or replaced, given how difficult it is to retrain these workers, how do we prevent persistent mass unemployment?”
“In the year 1900, over 40% of American workers were employed in agriculture,” he responded. “And yet last year, it was less than 2%. And no one’s wringing their hands about workforce losses in agriculture … So AI is going to be very similar. It is one of the most powerful tools that mankind has ever invented for the enhancement of human productivity.”
The thought of AI replacing human labor en masse certainly seems frightening, especially since unlike the labor shift away from agriculture, AI has the potential to reshape labor markets in years rather than several decades.
But it’s also the case that interventions to artificially preserve jobs seem like an unworkable solution. AI will enhance human labor and replace a lot of it — once we figure out where between those two extremes we land in the aggregate, we can address the welfare of our workers rather than fight an unwinnable battle against technological progress.
Preempting state law
A significant point of contention with GAAIA stems from its preemption of state law. This is important because GAAIA would enforce standards for catastrophic risk mitigation such as a large loss of life or property, but says nothing about concerns like deepfakes, AI scams, and perhaps most importantly, the use of AI models that produce biased outputs in critical decisionmaking such as in hiring and firing.
I asked Obernolte about whether it’s warranted to be concerned that GAAIA would prevent states from protecting their residents from a range of AI abuses.
“I don’t think that it would because we are creating a regulatory framework around development,” he answered. “We are letting the states regulate deployment and using AI to steal someone’s money is by definition a deployment of the AI. That’s not in the development of AI. Now, it is true that when a developer develops a powerful frontier model, we hope that they will think about the ways it can be used to, for example, steal someone’s money.”
Obernolte explained that he sees the role of the federal government as regulating the development of AI rather than its deployment and that states are free to regulate the deployment stage. According to Obernolte, things like deepfakes and scams would fall under deployment.
The trouble with Obernolte’s approach is that many states are passing laws that address those concerns at the development stage. For example, California passed Assembly Bill 489 last year, which regulates the development of AI chatbots that misleadingly pose as healthcare professionals. Legislation like this will likely be preempted by GAAIA, which is why organizations like the ACLU have released statements calling for the removal of the preemption language within GAAIA.
One can plainly appreciate why these problems may have to be regulated at both the deployment and the development stages given that companies who deploy AI models typically have limited means to address these issues themselves. It just makes sense to have developers themselves institute sensible guardrails to restrict how their models can be used.
If I had to make a suggestion to Obernolte and Trahan, it’s that GAAIA should either include remediations for problems that arise at the deployment stage or allow states to do so. By not addressing it in GAAIA and not allowing states to regulate at the development stage, they are eliminating a powerful tool that we can use to limit the abuse of AI.
Government misuses of AI
AI has the potential to be a powerful tool for intrusive government surveillance and earlier this year, the developer Anthropic was blacklisted by the federal government as a “supply-chain risk” for refusing to allow their use for mass surveillance and fully-automated weapons systems. While the legal and political fight over that plays out, it’s apparent how the federal government wants to use AI.
I asked Obernolte if there was any political will in Congress to protect against the government using AI to violate our right to privacy and against unreasonable searches.
“You know, certainly we have the rights that stem from our Bill of Rights around unreasonable searches and seizures that I think are going to have to be updated,” he said. “The courts are going to have to rule in certain circumstances under what can be done warrantlessly and what requires a warrant … And I think that they will eventually need some backup by Congress because we can’t turn judges into legislators.”
Despite constitutional protections, both Congress and the courts have allowed the government to spy on American citizens for decades through the NSA and programs like PRISM. The Trump administration also began employing AI-enabled services from corporations like Palantir and Babel Street to target activists – meanwhile, ICE has been alleged to use personal information to intimidate protesters.
Restricting the acceptable use of AI by our government is at least as important as restricting it for private citizens. Regrettably, as Obernolte states, “If you’re asking a bureaucracy to regulate itself, I think that you have embarked on a fool’s errand.”
Nevertheless, Congressman Obernolte wanted to end our talk on a positive note, reiterating his optimistic view on our future with AI.
“I think we do a terrible job at articulating the optimistic case for AI,” he said. “First of all, AI is already the most powerful tool for the dissemination of human knowledge that mankind has ever created. It can teach anybody anything that they want to learn in whatever learning style is optimal for them … So I believe that AI could be this next big catalyst that creates not only increased workforce productivity, not only increased size of our economy, but literally this rising wave of prosperity that lifts all the boats in America … We’ve got nothing but challenges. But I believe in the end, the benefits are going to far outweigh the drawbacks.”
I don’t necessarily share Obernolte’s optimism for various reasons but time will tell whether AI truly develops into something that is a net positive for the world.
For now, GAAIA is imperfect, but it is a step in the right direction in taking a proactive approach to the challenges that AI is beginning to present.
Rafael Perez is a columnist for the Southern California News Group. You can reach him at rafaelperezocregister@gmail.com.