Home » How China wants to secure the development of generative AI

How China wants to secure the development of generative AI

by admin
How China wants to secure the development of generative AI

How China wants to secure the development of generative AI

When Baidu finally introduced its longed-for ChatGPT competitor Ernie (Enhanced Representation through Knowledge Integration) to the public at the end of March 2023, a meme quickly made the rounds on Chinese social media: A black sedan with open double doors parked in a dark alley next to a dirty, dented dumpster . ChatGPT is written on the limousine, while the dumpster bears the Chinese name for the Ernie bot.

Advertisement

But there’s far more to the discussion about whether China’s flagship chatbot is a failure than just dissatisfaction with a bad piece of software. Behind the critical look at Ernie is the question: Can China do AI? Our author Eva Wolfgangel investigates this question in the latest issue 1/2024 of MIT Technology Review.

There are now around 130 major language models in China. However, since these are usually only accessible to Chinese users, the Chinese models cannot easily be tested with the now common benchmarks and classified into a corresponding leaderboard.

Reports about the quality of the models are therefore rather anecdotal. Experts do not want to be quoted publicly with an assessment. However, after an update in October, many of the interviewees see Ernie at about the same level as ChatGPT 3.5 – about a year behind development in the USA. However, a recently published comparison by the online service China Talk sees the chatbot Kimi from the start-up Moonshot AI ahead – among other things because the language model can process up to 200,000 Chinese characters as input.

See also  Women's Volleyball World Championships Comprehensive: The Chinese team won the first place in the group and advanced to the top 16 - Sports - CGTN

In addition to technical problems, Chinese developers are primarily struggling with regulatory obstacles. In March of this year, China became the first country in the world to pass regulation of generative AI. The first draft of the regulation was quite strict, says Helen Toner from the US think tank Center for Security and Emerging Technology (CSET), – which then led to a conflict of goals: “On the one hand, the state wants to support the industry and become a leader in the field of AI “, she says. “On the other hand, he wants to maintain control at the same time.”

China wants to be at the forefront of key technology. The new issue of MIT Technology Review therefore examines how far the People’s Republic is in various areas: from artificial intelligence to the chip industry to women’s rights. Highlights from the magazine:

Yiqin Fu, a political scientist at Stanford University, says the law was too general in many places, making it unclear for companies whether they were behaving correctly: “They wanted to behave in accordance with the law, but they didn’t know how.” That has now changed: In October, the Information Security Standards Committee published a draft for “basic requirements” for the security of generative AI with technical standards: Among other things, providers of AI models must randomly select 4,000 data points from each training corpus and examine them for possible prohibited statements . At least 96 percent of these must be considered acceptable or the corpus must be blacklisted. Accepted training data must also be filtered for prohibited content. Providers must also create a database of 1,000 questions to test the model’s refusal to respond. It must refuse to answer at least 95 percent of the questions it should not answer. These questions must cover tricky and sensitive topics such as politics, religion and the like.

See also  China holds back investments in the EU and Europe moves away from Beijing: now the Politburo embraces Big Tech to continue growing

The strict regulation slows down technical progress – and thus stands in the way of the goal of achieving AI technology leadership by 2030. “China has taken a firefighting approach,” says Jinghan Zeng, professor of China and international studies at Lancaster University – wait and then regulate when incidents occur. It has now become clear to the party that generative AI must be regulated immediately “so that it continues to support Chinese values.”

In China, however, there is a phenomenon that shapes politics there and thus also the population as well as the economy: Fang-Shou, derived from the two verbs Fang (to recover) and Shou (to tense). Meaning: A phase of tension or increased control is always followed by a phase of opening and loosening – and vice versa. Zeng therefore does not believe that regulation in China is significantly holding back innovation. “The European approach will slow down more than the Chinese one,” he predicts, “compared to the Chinese one, this one is much more restrictive.” And China has a clear advantage: access to a lot of data. “China is still far behind the USA, but that will change.”

(wst)

To home page

You may also like

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More

Privacy & Cookies Policy