By LI Linxu

As the rapid development of generative AI is garnering headlines, regulators around the world are grappling with how to manage this disruptive technology.

China has begun to solicit public opinion on a draft regulation for the management of generative AI products or services like ChatGPT or ERNIE Bot.

The so-called generative AI refers to a kind of technology based on algorithms or models that are able to generate content such as images, texts, voice, video and codes.

The draft regulation lays out a series of ground rules that generative AI services have to follow, including the type of content they are allowed to generate, and the procedures of putting them into the market.

For the governance of generative AI, the draft regulation is pioneering in many ways, particularly in proposing principles, drawing boundaries and clarifying responsibilities, which will set guardrails for the healthy development of this fast-evolving technology, said Rao Gaoqi, research fellow at the Institute of Language Intelligence under Beijing Language and Culture University, during an interview with S&T Daily.

PHOTO:VCG

Curbing disinformation

Cracking down on disinformation has long been a focus for tech regulators across the globe. With the emergence of generative AI, the task of fighting off disinformation is more urgent than ever.

ChatGPT's propensity to fabricate quotes and false facts has already raised huge issues on trust and reliability.

Disinformation is prohibited under the draft regulation, specifying that AI-generated content shall not disrupt economic and social order.

Product or service providers shall ensure that content generated from generative AI is true, accurate, and shall take measures to prevent the generation of disinformation, as per the draft regulation.

From the point of rule design and enforcement, the key to understanding this requirement lies in the latter half of the clause, that is, taking countermeasures against disinformation, said Xiao Youdan, research fellow at the Institutes of Science and Development under the Chinese Academy of Sciences, in an interview with S&T Daily.

The draft regulation proposes a full set of prevention rules to take aim at disinformation along the whole process, said Teng Rui, associate professor of law in Huazhong University of Science and Technology, during an interview with S&T Daily, noting that in its enforcement, attention should be paid to the cohesion between the new regulation and relevant laws and regulations.

Clearly, preventing disinformation is one of the most important compliance requirements in the draft regulation, said Shi Xiaonan, partner at Merits & Tree, in an interview with S&T Daily. Given AI's black box effects, regulators should take full account of the regulation's enforceability.

PHOTO:VCG

Protecting IPR

The protection of intellectual property rights (IPR) is also highlighted in the draft regulation.

This month, news of an AI-generated image winning a prestigious photography award shocked the world, causing an outcry in the art circle and exposing the IPR dilemma associated with generative AI.

Generative AI is not a lawless territory, said Rao, noting that given its complex application scenarios, how to properly define the boundaries of IPR is testing the wisdom of regulators.

China has now drawn IPR red lines for generative AI. Providers of generative AI products or services should respect IPR and business ethics, and avoid IPR infringement, according to the draft regulation.

Meanwhile, the pre-training and optimized training data used for generative AI shall not contain content that infringes IPR.

Focusing on data source, data processing and content generation, the draft regulation sets up a series of targeted IPR protection measures, said Teng, adding that the regulation philosophy behind these measures is in line with relevant laws and regulations, and is conducive to cultivating an environment of science for goodness.

PHOTO:VCG

Addressing privacy concerns

As the popularity of generative AI rises, so do concerns over violation of personal privacy.

Italy has chosen to temporarily ban ChatGPT over privacy concerns.

Compared to an outright ban, China's regulation effort is a tentative and beneficial direction toward reining in such technology, Tian Jingqiao, lawyer specialized in cybersecurity law, said to S&T Daily, adding that the regulation's real effects remain to be seen and some of the enforcement rules may still require further elaboration and refinement.

Illegal acquisition, disclosure, and use of personal information, privacy and business secrets are prohibited by the draft regulation. Service providers shall establish a mechanism to handle user complaints and make timely remedies.

Although there are some worries regarding the compliance costs, all experts agree that establishing a robust regulation system is critical for the healthy development of generative AI.

If the technology is left unregulated, the consequences could be dangerous, said Teng, though this means that relevant companies need to invest more resources on compliance, it is essential to ensure the development of generative AI on the right way.

"It is absolutely right to install brake systems in a car before letting it hit the road," said Xiao.