China‘s Multi-Billion Dollar Bet on AI Self-Reliance

The AI Arms Race Heats Up

The global artificial intelligence race is kicking into high gear as the US and China jockey for dominance in this critical emerging technology. Amid an escalating tech cold war and the continued decoupling of the world‘s two largest economies, China is going all-in on its quest for AI self-reliance.

Nowhere is this more evident than in the scorching hot field of generative AI. Using massive deep learning models trained on internet-scale datasets, generative AI systems can create convincingly human-like text, images, video, audio and code from just a simple text description. When OpenAI unveiled its GPT-3 language model in 2020, it set off an arms race to build ever larger and more capable models.

Now, Chinese tech titans Baidu, Alibaba, Tencent and Huawei are investing billions to develop their own state-of-the-art "foundational models" to rival American counterparts like GPT-3, LaMDA and Megatron-Turing. The goal is not just to have the most advanced AI, but to create a robust domestic ecosystem of startups, researchers and developers that can drive Chinese innovation without relying on US tech.

Baidu‘s $145M Bet on Chinese AI Startups

Baidu fired the latest salvo in March 2023 with the launch of a 1 billion yuan ($145 million) fund to invest in Chinese startups working on generative AI. The fund, which will cut checks of up to 10 million yuan ($1.4 million) for early-stage companies, aims to jumpstart an ecosystem around Baidu‘s Ernie Bot large language model.

Ernie (short for "Enhanced Representation through kNowledge IntEgration") is Baidu‘s flagship AI system that powers its search engine, smart speakers, and self-driving cars. With the latest Ernie 3.0 Titan model, Baidu claims it has surpassed GPT-3 in Chinese language understanding and generation.

Ernie 3.0 Titan boasts 260 billion parameters (the "weights" that store the model‘s knowledge), edging out GPT-3‘s 175 billion. It was trained on 4.9 terabytes of cleaned Chinese text data spanning domains like encyclopedias, news, social media, and web pages. On the CLUE benchmark for Chinese language understanding, Ernie 3.0 achieved state-of-the-art results of 90.5% accuracy, beating the previous record of 88.1% set by Huawei‘s PanGu-Alpha model.

To spur adoption of its platform, Baidu released an API for startups and developers to easily build applications on top of Ernie 3.0. Early use cases include AI-powered legal contract review, intelligent tutoring systems, and automated 3D game character generation. Baidu has also open sourced parts of Ernie‘s codebase on GitHub to encourage community collaboration.

Baidu‘s AI fund follows in the footsteps of OpenAI, which launched a $100 million startup fund (later increased to $175 million) in 2021 to support companies building on its GPT-3 model. By fostering a domestic AI ecosystem centered around Ernie Bot, Baidu hopes to solidify its position as the leading provider of AI technology in China.

China‘s AI Ambitions and America‘s Response

Baidu‘s big bet on AI is emblematic of the Chinese government‘s broader push to achieve parity with the US in key strategic technologies. In recent years, China has:

• Publically stated its goal to become the world leader in AI by 2030
• Launched major national initiatives like the New Generation AI Development Plan to coordinate funding and research efforts
• Invested billions of dollars in AI startups and research centers
• Attracted top global AI talent to its universities and companies
• Published more AI research papers than any other country (see data below)

*Source: Allen Institute for AI*

As of 2023, China is home to over 1,100 AI companies with a combined market value of $140 billion, according to CB Insights. The government has designated several national AI champions, including Baidu, Alibaba, Tencent and Sensetime, to lead development of core technologies like computer vision, natural language processing, and robotics.

This rapid buildup of Chinese AI capabilities has raised alarms in Washington, D.C. The US government has taken steps to slow China‘s progress and maintain American leadership in AI, including:

• Imposing export controls on advanced semiconductor technology to China
• Limiting Chinese investment in US-based AI companies
• Increasing federal funding for AI research and development to $6 billion per year
• Creating the National AI Initiative Office to coordinate AI strategy across government agencies
• Forming international partnerships with allies like the EU, UK, India, and Japan to counter China

The battle for AI supremacy has become a key front in the escalating US-China tech rivalry, with both sides pouring resources into outspending and out-innovating the other. As the two AI superpowers continue to decouple, the stage is set for an era of unprecedented competition that could have profound implications for the global balance of power.

Is China‘s AI Ecosystem Ready for Prime Time?

For all its headline-grabbing investments and initiatives, questions remain about the maturity and real-world readiness of China‘s homegrown AI ecosystem. Some key challenges include:

• Model Reliability: When Baidu‘s Ernie Bot was unveiled in March 2023, beta users discovered it had a tendency to hallucinate information and make factual errors. This reflects the difficulty of getting large language models to consistently generate truthful and reliable outputs, a problem that also afflicts US models. More robust testing and validation is needed before deploying these models in high-stakes domains.

• Talent Shortages: While China now graduates more AI PhDs than any other country (2,300 in 2020 compared to 1,900 in the US), it still faces a significant talent shortage, especially for experienced engineers. A 2018 Tsinghua University report estimated that China‘s AI talent pool was just one-fifth the size of America‘s. Upskilling workers and attracting global talent remains an ongoing challenge.

• Semiconductor Dependence: China‘s AI chip startups like Cambricon and Horizon Robotics have made impressive advances, but still rely heavily on foreign suppliers for key components and manufacturing equipment. With the US choking off access to cutting-edge chips, China faces an uphill battle to develop native semiconductor capabilities that can support its AI ambitions.

• Research Originality: Although China now publishes more total AI papers than the US, America still leads on the number of highly-cited studies that represent significant original breakthroughs. The top 100 most-cited AI papers from 2020-2022 included 44 with solely US-affiliated authors compared to just 12 for China. More fundamental research is needed to drive Chinese AI innovation.

• Ethical Concerns: The Chinese government‘s use of AI for mass surveillance, predictive policing, and censorship has raised alarm bells about the country‘s commitment to ethical AI development. Stronger governance frameworks are needed to ensure responsible deployment of AI in both commercial and public sector contexts.

The Road Ahead

Despite the challenges, China‘s whole-of-society push for AI leadership shows no signs of slowing down. President Xi Jinping has declared that AI is critical to China‘s future economic competitiveness and national security. The 14th Five Year Plan calls for major investments to "seize the high ground" in AI and other frontier technologies.

Beijing‘s strong top-down industrial policies, coupled with a vibrant grassroots startup ecosystem, have created structural advantages for Chinese AI development that will be hard for other countries to match. With tech giants like Baidu writing big checks and the government greasing the wheels, Chinese AI companies are poised for explosive growth in the coming years.

At the same time, realizing the full potential of AI will require addressing very real challenges around ethics, safety, robustness and international cooperation. As AI systems become more powerful and pervasive, ensuring they remain under human control and aligned with our values is paramount.

Both the US and China have a shared responsibility to pursue AI development in a way that benefits humanity as a whole. This could involve initiatives like jointly funded research projects, standards setting bodies, and confidence-building measures to reduce the risk of an unconstrained AI arms race.

If China and the US can find ways to collaborate while still competing, it would go a long way towards ushering in a new golden age of AI – one that uplifts people around the globe. The alternative, an AI "splinternet" bifurcated between US and Chinese spheres of influence, would deprive people everywhere of the technology‘s full benefits.

As the world‘s AI superpowers charge ahead, the decisions they make today will shape the trajectory of the field for decades to come. Will we choose a path of unchecked competition and national self-interest, or find ways to cooperate in pursuit of beneficial AI for all? Time will tell, but one thing is clear – China has made its multi-billion dollar bet on AI loud and clear, and plans to reap the rewards for a long time to come.