Governments around the globe are racing to control AI before it becomes too deeply embedded in society. But recent research suggests poorly designed rules could make AI systems less protected than having no regulation in any respect.
Regulatory disagreements within the US are resulting in a patchwork of approaches as states take matters into their very own hands. A key query is who ought to be liable for the security of AI products—the large tech firms constructing the underlying models or the firms that adapt them for a selected task, reminiscent of a customer support chatbot or an AI tutor.
Working this out is trickier than it looks. While it may appear logical to place the majority of the burden on downstream firms directly serving these tools to customers, a brand new study in Proceedings of the National Academy of Sciences finds that could possibly be worse than having no rules in any respect.
“There’s a free-riding behavior that happens,” Benjamin Laufer from Cornell University, who led the research, said in a press release. “The regulation acts as a tool for the overall provider to dump the security burden onto the downstream specialist.”
The researchers’ evaluation relied on a model based on game theory—a mathematical approach to studying decision making. It treated AI development as a two-step game, wherein a “generalist” developer first invests in constructing a broadly capable AI model before a “specialist” adapts it for a selected domain and takes it to market.
In the sport, a regulator sets a minimum safety standard for each players, and the models see this prematurely. They then put money into each the performance and safety of their product, and the revenue is split between them. Investments in each get progressively higher, while the additional revenue each improvement brings in stays flat.
The issue, the researchers found, is that the generalist moves first and knows exactly what the specialist might be legally required to do afterwards. This creates problems when the generalist is about a low bar for safety, or none in any respect, and safety standards for the downstream specialist are also fairly weak.
Within the absence of any rules, each firms put money into safety, since the model assumes a safer product earns more revenue. But when the specialist is forced to speculate a specific amount into safety to fulfill regularity requirements, the generalist can cut its own spending and let the downstream firm close the gap.
That’s since the generalist’s revenue relies on the ultimate safety level of the shipped product, not by itself contribution, so it might get a revenue boost from improved safety without paying for it from its own pocket. The specialist, for its part, has no reason to do greater than the rule demands, so total safety settles on the legal minimum, which is below what would have occurred had there been no regulation in any respect.
On a more positive note, the researchers found that if safety levels on each the generalist and the specialist are set high enough, regulation can actually improve safety while leaving each firms more profitable than they were in an unregulated market.
“Appropriately designed AI regulation could make it possible for various firms involved within the AI development pipeline to collectively arrive at good outcomes for consumers, knowing that the regulation is designed to assist each firm operate in a way that the others can more reasonably predict,” co-author Jon Kleinberg from Cornell University said within the press release.
Nonetheless, the researchers’ model relies available on the market setting an actual price on safety. Because the gap widens between what customers pays for performance and what they’ll pay for safety, the range of circumstances wherein weak rules backfire gets narrower.
The authors also note that the model’s two-player setup is a simplification of real AI supply chains where multiple competing specialists and base-model providers operate across different jurisdictions with different rules.
“People consider AI as a single object, but actually AI involves a really complicated set of stakeholders and actors that every have their very own contributions to the technology,” said Laufer. “To manage in a thoughtful way, we want to contemplate the entire supply chain, not only a single provider or entity.”
Still, the outcomes suggest that taking a very simplistic and light-handed approach to AI regulation may find yourself achieving the alternative of what law makers intend.