AI Regulation
A groundbreaking modeling study has revealed that poor regulation of artificial intelligence (AI) can be worse than a complete absence of legislation when it comes to the safety of AI products and services.
The difficulty in implementing strong federal regulations for AI has led to isolated and limited-scope initiatives, but it is unclear how this patchwork of laws will alter the development of products and services and the behavior of companies.
To better understand this, researchers from Cornell and Carnegie Mellon universities in the US developed a theoretical model to estimate the effects of AI regulation, both for companies that produce general-purpose generative AI models – such as those behind popular chatbots – and for companies that apply these models, creating a variety of applications, such as customer service tools or medical diagnostic systems.
“The goal of regulation should be the mutual benefit of everyone in society, and that can include those who develop the technology, but also end users and the public,” argues Professor Benjamin Laufer. “There isn’t much security regulation for AI, and many possible regulations are only proposed at this stage. In a way, regulation is like groping in the dark, so it’s worth analyzing the effects these regulations might have on incentives.”
In the model developed by the team, it is possible to define a minimum security requirement – whether for the AI company in general, for the company developing the downstream technology, or for both – and then estimate the security and performance of the products in the market.
Good for consumers and good for businesses.
The result was surprising: When regulations targeted only companies developing application technology, the resulting products proved to be less secure than if there had been no regulation at all.
“An undue exploitative behavior emerges,” said Laufer. “Regulation functions as a tool for the general supplier to transfer the burden of safety to the value chain specialist.”
On the other hand, centralized and balanced regulation, which assigns responsibility to both primary development companies and application companies, results not only in safer products for consumers, but also in higher profits for all companies involved.
This happens because, when general AI producers and companies in the value chain need to meet a specific security goal, it reduces the risk for both categories of companies, since they don’t need to rely on each other’s word that certain security investments will be made.
The model created by the researchers is still quite simplified, but they intend to expand the work by analyzing the real impacts of regulation on the development and security of AI models.
Source: www.inovacaotecnologica.com.br
Source link
