Now Reading
[free access] Why ChatGPT is not Chinese ?

[free access] Why ChatGPT is not Chinese ?

Two days apart, OpenAI publishes its ChatGPT 4 while Baidu launches Ernie, his chatbot. In this article, Jean-Yves Larguier discusses the major challenges for China in this strategic industry.

For a few years, it was fashionable in China to at least seemingly ignore, Silicon Valley. The dynamism of the Chinese tech scene was enough on its own, and entrepreneurs could admire the country's successes in artificial intelligence. But that sweet tale shattered in November 2022 with the release of ChatGPT by OpenAI. Testimony of the curiosity aroused by the performance of ChatGPT, Chinese netizens all sought to try it. Despite the difficulty for the general Chinese public to access the application, some have found ways on e-commerce sites to get an access code via a foreign phone number and connect from abroad. However, this flourishing trade was quickly brought to a halt by these same platforms., by order of the authorities.

ChatGPT's breakthrough in generative AI has shed light on AI performance. China has long placed artificial intelligence at the top of its technological priorities and enjoys the support of consistent public policies. As Pascal Boniface reminds us in his book “Geopolitics of artificial intelligence”, China has had its own “sputnik moment” : “It took place in May 2017 lorsque Alpha Go, artificial intelligence program developed by the British company DeepMind (redeemed in 2014 by Google), beat the Chinese go game world champion Ke Jie ».

Since, China is engaged in a real technological race around artificial intelligence, with planned applications in key areas such as the internet, the scientific research, autonomous driving, health, education, security and military. These technological advances are expected to play a major role both economically and geopolitically., and all powers must invest in AI or risk being downgraded.

In October 2018, during its speech before the Politburo of the Communist Party of China Central Committee (PCC), Xi Jinping pointed out that " AI is a strategic technology that paves the way for this new era of scientific and industrial revolution and industrial change, and that it has an effect of “goose head” (head goose) with strong stimulating ability ». China and the United States are vying to take the lead in this new field.

China had long hoped for a “technological reset”, a kind of reset to put China and the other powers on the same starting line. Out of the game at the time of the industrial revolution, behind in information technology, China holds with AI its opportunity to compete in this technological transformation. It even aspires to become the leader in artificial intelligence innovation here. 2030, a crucial issue to become the first world power. China wants to make AI its future economic engine by deploying it in key areas.

It is important to study China's strategy in this technological race and analyze the advantages and challenges it faces in this global competition.. The arrival of ChatGPT confronts China with the reality of American technology and leads it to assess its positioning after years of investment, while raising questions as to why ChatGPT is not a Chinese invention ?

La Chine surprise par la performance de l'Intelligence Artificielle de ChatGPT

Over the last few weeks, the generative AI tool developed by OpenAI has been widely commented on in the media, and in particular on Chinese social networks, where many expressed outrage at ChatGPT's lead on projects carried by national champions.

ChatGPT is a chatbot, a computer program that can understand and generate natural language. It is based on GPT-3.5, a broad language model (LLM), trained on large datasets, using a combination of supervised and reinforced learning from human feedback. GPT (Generative Pre-trained Transformer) is a predictive natural language model that can generate text based on an estimated probability of the next sequence of words. However, he has no reasoning ability. To initiate his training, the model uses the training method called “deep transfer learning”, which uses layers of neurons pre-trained on large amounts of data to improve its performance. Large language models require significant computing power. To speed up and improve the training process, specialized chips for intensive computing, such as graphics processing units (GPU) and tensor processing units (TPU), are essential.

The GPT-3.5 model, trained on 570 gigabytes of text, account 175 billions of parameters, hundred times more than its predecessor GPT-2. Each of these parameters in Artificial Neural Network technologies is comparable to a neuron that interacts with others to understand and react to its environment.. In addition to the number of parameters, the architecture of the model, the density of the neural networks as well as the richness of the training corpus, fine-tuning of the model are also essential for the quality of a generative AI. In December 2022, ChatGPT would have passed the Turing test, an AI assessment in which a human is unable to distinguish between a response generated by a chatbot and a human response. To go even further, the ChatGPT model will also have to face other more complex tests to understand the nuances of language and resolve sentence ambiguities, such as the Winograd Schema.

If the third version of ChatGPT has shaken up the tech giants, the latter are now preparing to integrate chatbot solutions into their applications : Microsoft recently announced the integration of OpenAI's GPT model into Bing, offering a user experience similar to that of ChatGPT in its search engine. Google, on the other hand, unveiled BARD, its own chatbot powered by its LaMDA language model, which will soon be available to the general public. This chatbot technology could revolutionize the user experience of search engines, but also the productivity of companies by supplementing or even replacing the human for certain tasks. Generative AI is now being considered in many areas, such as customer service, education or computer programming. China is also engaged in this race and is preparing to unveil its own advances in generative AI

ChatGPT has put competition in artificial intelligence back in the spotlight. No way for Chinese companies to be left behind. Baidu, China's largest search engine company, announced the launch of a conversational chatbot service for the end of March 2023, which will be integrated into its search engine. Called Ernie, pour « Enhanced Representation through kNowledge IntEgration », ou « Wenxin Yiyan 文心一言 » en chinois, the solution was first introduced in 2019 and has since expanded its capabilities to include language comprehension and generation and text-to-image generation. Baidu thus becomes the first of its Chinese counterparts to join the global race, knowing that ChatGPT and Google are not available in China. The question is whether Baidu will be able to compete with them in the international market.. According The Register explains that his pre-trained language model, called "ERNIE 3.0 Titan », understand 260 billions of parameters, be more than the 175 billion parameters of ChatGPT's GPT-3 model.

An ambitious and dynamic China in Artificial Intelligence

China has several major assets, first of all with its financial power. In 2021, the AI ​​industry in mainland China received funding from 17,21 billions of dollars from private investment, supplemented by public subsidies from the central government and local communities under the National Development Plan for Next-Generation Artificial Intelligence (NGAIDP).

Universities and institutes participate in this dynamic, by training more AI engineers than the United States, and conducting important research and conferences. The Beijing Artificial Intelligence Academy has developed Wu Dao 2.0, the largest pre-trained AI model in the world with 1,75 trillion parameters (10 times more than GPT-3, but results still confidential), to simulate conversational language and understand images. Wu Dao has the distinction of having been trained on Chinese corpora different from those of the United States, cultural and linguistic differences playing an important role in generative AI.

China also has a considerable advantage thanks to its large amount of data from both e-commerce platforms, of the health sector, payment applications than surveillance cameras. Unlike Western companies which are more assertive and better protected with regard to the use of their data, citizens are in a less favorable position with regard to controls and the sharing of their data. Chinese companies benefit from very favorable regulations for the use of this data collected by the government and the private sector, which creates a significant market for companies specializing in artificial intelligence focused on practical applications.

The Chinese state encourages the grouping of AI players around regional hubs. In particular to Beijing where 1048 companies were identified in October 2022, i.e. one third of the country's companies. This allows for a high concentration of talent, with a community of 40 000 AI experts. Other centers have developed around Shanghai and Shenzhen. The Chinese government also encourages private sector companies, such as Alibaba, CloudWalk, iFlytek a Sensetime, to collaborate with the People's Liberation Army and security services to share their technological innovations and their infrastructures, thus creating a military-digital complex. This strategy of military-civilian fusion is considered to be one of the factors contributing to the rapid growth of the AI ​​industry in China in recent years.

Chinese tech giants such as Baidu, Bytedance, Huawei, Inspur, Tencent, Alibaba and SenseTime own advanced expertise in AI, as well as a large amount of data needed to develop a model. Alibaba has placed AI at the center of its international research laboratory DAMO Academy. L’enterprise currently testing its generative AI and chatbot tool, that it could possibly combine with the group's communication application, DingTalk. As for Tencent, son AI Lab developed the Different Dimension Me tool for generating images in the DALL-E family of OpenAI. Like the GAFAMs, the tech giants in China are all developing AI-based solutions to integrate them into their tool suites or super-applications.

As members of " the national AI team », major Chinese technology companies are on a mission to develop a AI ecosystem as complete as possible. Baidu is not exclusively focused on conversational and generative AI, but has also invested in other AI technologies such as autonomous driving with its platform Apollo, in partnership with around forty renowned car manufacturers. Baidu also offers cloud AI services by providing modular AI solutions on their compute infrastructures, using in particular its AI processors and accelerators developed by its subsidiary Kunlun Technology. In addition, Baidu opened PaddlePaddle, a deep learning platform (Deep Learning) and open source, used by a community of nearly 5 million developers, where many pre-trained AI models are shared.

Thanks to its strengths, China dominates some AI subfields, such as facial recognition and computer vision, where she excels, as well as natural language processing (NLP) for chinese, and machine learning for data analytics and e-commerce, where China is very competent. According to AI annual report. Index 2022 from Stanford University, China leads in number of AI publications in journals and conferences, as well as patent applications. However, regarding granted patents, China is still lagging behind the United States, as well as in terms of the quality of the publications.

See Also


After all the investments and flashy announcements about Chinese artificial intelligence, one wonders why no Chinese actor was the first to surprise the world with a ChatGPT.

ChatGPT : a temporary or permanent setback for China's AI ambitions ?

Several factors explain why the Chinese were not the first to develop such a mature product as ChatGPT. First of all, looking at the different AI companies in China, we see that priority is given to defense-related applications, security and the maintenance of public order, thus relegating research efforts on open-domain conversational AI to the background.

Ensuite, generative models present particular technical and governance complexity, being less regular and predictable than traditional software. These models therefore require careful verification and long training to meet the ethical and moral standards of the country., a point on which China will not tolerate deviation. Generative AIs and chatbots will have to toe the official policy line, like the Chinese internet within the Great Firewall. An AI engaging in a critical rant against the Communist Party is unthinkable and would spell corporate bankruptcy. This pressure weighs on Chinese entrepreneurs, limiting innovation in generative AI for fear of sanctions from the authorities. The Chinese press quickly warned against the biases of ChatGPT likely to lead to misinformation and calls for a stricter regulations. Furthermore, the real performance of these Chinese models remains largely unknown.

A social factor must also be taken into account, because generative AIs have the potential to replace humans in a number of basic or repetitive tasks, which could lead companies to cut many jobs. To reconcile economic growth and maintaining employment, the Chinese government will have to be vigilant about the deployment of AI in certain sectors. Unlike Western countries, China may be less inclined to seek productivity gains to avoid worsening unemployment among young graduates. Since a few weeks, this concern is also relayed on Chinese social networks.

However, Mingke Luo, CEO and Founder of MRS.ai, a Chinese company specializing in artificial intelligence, suggests that " what was perhaps most lacking, this is the mindset of OpenAI researchers : 300 scientists with a vision and a desire to advance generative AI, working away from the economic constraints of investors and their search for short-term profits ». For this Shanghai-based AI expert : « the non-profit nature of the project was an asset to allow these researchers to go beyond their limits ». Also, it seems that with the takeover of private technology players, China has yet to develop a forward-thinking vision for the use of generative AI. The United States seems better able to exploit the full potential of this technology and exert global influence. This will lead US companies to better monetize AI and attract more of the world's top talent, as they already do.

Finally, the major obstacle facing China is limited access to the best computing capacities, consequence of US restrictions on the purchase of US AI chips, as well as the design and manufacture of the most advanced semiconductors. The United States aims to limit the ability of Chinese technology companies to create large neural networks by blocking access to chips high-end computing. Although most Chinese AI startups are application-focused and don't need high-performance semiconductors to handle massive data, those doing basic research may struggle using less powerful chips that slow down the computational process and increase costs. US tech sanctions raise concerns about China's ability to keep up with AI advances. Despite China's reserves of chips for its server farms, it is inevitable that at some point, China will be at a disadvantage in terms of computing power, what will impact AI projects in China.

Conclusion

Countries that do not bet on artificial intelligence will be downgraded as much as the countries of the 19th century that did not embrace the industrial revolution. This is therefore a major issue of sovereignty. China is definitely on the right track with the United States. The difference will be in their respective ability to attract the best talent, to unleash the innovation potential of the private sector and to propose a societal project for the use of beneficial and sustainable AI.

As dazzling as it is colossal, the technological leap of generative AI and chatbots is causing both excitement and concern. Although it will take some time to see its impact on economies and societies, The race is on, fueled by power rivalry. The 14 mars 2023, OpenAI already unveils an improved model, and GPT-4, which has a hundred times more parameters than GPT-3. China struggles to demonstrate convincing results, but she may not have revealed all her cards yet. Competition in the field of AI will greatly depend on the capacity of each country to master the high-performance chips and the necessary computing infrastructures..

AI pioneers will be able to influence the standards for regulating AI algorithms. China has already taken action with the Cyberspace Administration of China (CAC). Europe is preparing a bill on artificial intelligence, AI Act, to establish a legal framework for AI, the regulation of algorithms and an ethics of AI, hoping to set a global standard. Control systems will be put in place, however this could lead to the development of models specifically trained for certain regions on which applications will be developed, accentuating the decoupling of information and technologies. At the International scale, the great powers of AI will certainly compete for influence to spread their own vision of artificial intelligence.

Scroll To Top