Chip war and censorship hobble Chinese tech giants in chatbot race

Chip war and censorship hobble Chinese tech giants in chatbot race

Baidu's AI chatbot has not yet been launched for public use
Baidu's AI chatbot has not yet been launched for public use. Photo: Jade GAO / AFP/File
Source: AFP

PAY ATTENTION: Enjoy reading our stories? Join YEN.com.gh's Telegram channel for more!

Search giant Baidu's lacklustre unveiling of its chatbot exposed gaps in China's race to rival ChatGPT, as censorship and a US squeeze on chip imports have hamstrung the country's artificial intelligence ambitions.

The highly anticipated preview of "Ernie Bot" last week was limited to a pre-recorded demonstration with simple questions to summarise the plot of a sci-fi novel and solving a straightforward algebra equation -- to avoid politically and factually incorrect answers.

From cloud computing to autonomous driving, none of the array of services Baidu had earlier promised its Ernie Bot could do were on display.

The firm's shares plunged as much as 10 percent during the unveiling, although they rallied the following day on positive reviews from brokerages including Citigroup, whose analysts were among a small group of people invited to test the bot.

Read also

Boosted by Ukraine war, German arms maker joins DAX index

A flurry of Chinese companies including Alibaba, JD.com, Netease and TikTok-parent Bytedance have rushed to develop services that can mimic human speech since San Francisco-based OpenAI launched ChatGPT in November, sparking a gold rush in the market.

Google on Tuesday invited people in the United States and Britain to test its AI chatbot, known as Bard, as it continues on its own push to catch up.

PAY ATTENTION: Click “See First” under the “Following” tab to see YEN.com.gh News on your News Feed!

The popularity of ChatGPT in China -- where users have to scale Beijing's internet firewall using virtual private networks (VPNs) and foreign phone numbers -- has left Baidu and others scrambling to regain its dominance on home turf.

"OpenAI probably spent as much time just testing GPT-4 as Baidu spent building Ernie Bot," said Matt Sheehan, fellow at the Carnegie Endowment for International Peace.

Read also

How AI could upend the world even more than electricity or the internet

"China's tech ecosystem doesn't have a tradition of funding open-ended research that doesn't have a clear path to profitability."

Chip supply

Ernie Bot is fluent in Mandarin, as well as other regional languages including Hakka spoken in South China and Taiwan, and targets the Chinese market with more than one billion internet users.

A headache for developers is Beijing's heavy-handed censorship of anything seen as challenging the Communist Party -- including a one-time purge of Winnie-the-Pooh after the cartoon bear was compared to Xi Jinping.

Baidu co-founder and CEO Robin Li speaks at the unveiling of Ernie Bot
Baidu co-founder and CEO Robin Li speaks at the unveiling of Ernie Bot. Photo: MICHAEL ZHANG / AFP/File
Source: AFP

When asked if the president of 10 years is "a good leader", one of China's best-performing publicly available ChatGPT-style models, developed by Beijing's Tsinghua University, says: "The input may contain ethical content. Please try a different input."

The strict restrictions on the Chinese internet mean companies have "significantly less data resources for training purposes compared to Western competitors", Lauren Hurcombe, a technology lawyer at DLA Piper, told AFP.

Read also

Asian markets extend global rally as banking sector fears ease

Ernie Bot has not yet been launched for public use.

China has announced ambitious plans to become a global leader in the field of AI by 2030, and consultancy group McKinsey estimates the sector could add about $600 billion every year to China's gross domestic product by then.

Most of the growth will come from producing driverless cars, adding more robots to assembly lines and healthcare breakthroughs, according to McKinsey, and the government has also used AI to beef up its mass surveillance programme.

However, Washington has moved to suffocate China's technology ambitions, blocking through sanctions its access to high-grade chips, chipmaking equipment and software used to design semiconductors.

This has made it difficult for Chinese companies to buy chips including Nvidia's A100 and its successor H100, considered the gold standard for large-scale AI training systems.

"There is a real question whether a domestic supply can be generated in the short term," Hurcombe said.

Read also

One billion users, but trouble mounts for TikTok

AI gap

But the effect of the US measures will take time to make a dent because Chinese companies rushed to stockpile high-end chips before Washington announced the export controls in October.

Baidu has its own chip design arm, Kunlun, and the company says it is capable of mass-producing a seven-nanometre chip that is partly used to power its AI systems.

Dou Shen, head of Baidu's AI Cloud group, shrugged off questions about the impact of the US restrictions during a call with investors in November, saying: "We think the impact is quite limited in the near future."

For years, China has bragged about filing more artificial intelligence patent applications than the United States.

But the average number of citations of its patents -- an indication of the importance and originality of its inventions -- lagged behind the United States and other developing countries in 2020 and 2021, according to Stanford University's AI Index 2022 report.

Read also

Baidu shares fall after disappointing AI chatbot debut

The United States also had twice as many AI startups as China, and had three times more private investment flowing into the sector in 2021, according to the report.

The Chinese government's top-down approach to spurring innovation has failed to deliver results.

The Beijing Academy of Artificial Intelligence, established in 2018, introduced a ChatGPT-like product two years ago.

Wu Dao was described by its creators as "the world's largest" AI language model with 1.75 trillion parameters, which is significantly larger than OpenAI's previous GPT-3 model with 175 billion parameters. But it never really caught on.

New feature: Сheck out news that is picked for YOU ➡️ click on “Recommended for you” and enjoy!

Source: AFP

Authors:
AFP avatar

AFP AFP text, photo, graphic, audio or video material shall not be published, broadcast, rewritten for broadcast or publication or redistributed directly or indirectly in any medium. AFP news material may not be stored in whole or in part in a computer or otherwise except for personal and non-commercial use. AFP will not be held liable for any delays, inaccuracies, errors or omissions in any AFP news material or in transmission or delivery of all or any part thereof or for any damages whatsoever. As a newswire service, AFP does not obtain releases from subjects, individuals, groups or entities contained in its photographs, videos, graphics or quoted in its texts. Further, no clearance is obtained from the owners of any trademarks or copyrighted materials whose marks and materials are included in AFP material. Therefore you will be solely responsible for obtaining any and all necessary releases from whatever individuals and/or entities necessary for any uses of AFP material.