Chinese views on ChatGPT and Baidu's upcoming alternative
They appear to believe opportunities abound, and a successful Chinese stand-in would be preferable.
By now, everyone must have developed some familiarity with ChatGPT. (If you haven’t, that’s your loss.) Discussions about the Microsoft-backed OpenAI product have gone viral in China as well, despite the Nikkei Asia report that regulators have told major Chinese tech companies such as Tencent and Alibaba not to facilitate or platform access to ChatGPT.
(This is not a confirmation of the report, and OpenAI officially didn’t make ChatGPT accessible to the Chinese mainland in the first place, although there are ways to get around the developer’s restriction.)
Domestically, the Chinese government and experts have recognized the seismic changes brought by ChatGPT and voiced their concerns often based on apparently political considerations and, in at least one example of a Chinese journal article, based on erroneous information.
Numerous Chinese companies have professed their interest in riding the waves of ChatGPT-like services, but Baidu, the Chinese tech giant owning the locally dominant search engine, appears particularly determined and has got the most international attention.
On January 30, Reuters and Bloomberg reported - each citing one unnamed source - that Baidu was developing its own ChatGPT. The Wall Street Journal followed suit quickly as well. It’s perhaps not unexpected that, as Bloomberg put it one week later, Baidu Surges as Hope Mounts over Chinese Answer to ChatGPT.
It’s unknown who the anonymous sources were, but Robin Li went on the record in an earnings call this week that its ChatGPT-like app will revolutionise its search engine, as Reuters dutifully quoted the Baidu CEO in the title of the report. A WSJ column is more skeptical, saying in Baidu’s ChatGPT-Style Bot Will Be No Magic Bullet that “whether the Chinese company’s Ernie Bot can recharge stagnant advertising revenue is an open question.”
From the transcript of the earnings call
Alicia Yap -- Citi -- Analyst
My question is related to ChatGPT. So, there's been a lot of expectation on Baidu progress in ChatGPT and also the AI-generated content. Can management share with us what could be the new opportunities that you are seeing in this field? And how is Baidu positioned in this trend? Any color that management can share would be appreciated.
Robin Li -- Co-Founder and Chief Executive Officer
Thank you for your questions, Alicia. We are obviously excited about ChatGPT and AIGC. It represents a magnitude that could change a lot of things. We are working on Ernie Bot, a new version of conversational AI bot that uses our latest technology in large language models.
We will embed Ernie Bot into Baidu Search first and will open it to the public in March. Baidu is as leader in China's technology innovation. Remember, we launched early in March 2019 and have scaled it up with well over 100 billion parameters. It is trained by serving billions of user search requests and other applications every day.
So, in terms of NLP capabilities, Ernie is considered as the state-of-the-art Chinese language model. And it is not only about language, but also about understanding Chinese culture. Ernie 3.0 is already a very localized AI foundation model for China market, which means that the generated large language model we are working on right now will be more suitable in Chinese language and to the China market than models developed overseas. In addition, our deep learning framework, PaddlePaddle, which Ernie is based, has gained a strong momentum.
Millions of developers use Paddle for their AI works. There is a strong synergy between the framework layer and the model layer. AI pre-training, as you know, is very expensive. We believe our full stack AI capabilities will allow us to build the most efficient large language model and support all kinds of applications from search to content generation, or any vertical area that could improve productivity significantly.
For Baidu Search, I just mentioned that we're working hard on a revolutionary version of Baidu Search built upon Ernie Bot that incorporates generative AI into our search algorithm, as well as content creation. And we are adding interactive features, too. Users will soon be able to interact directly with the new generated large language model. It could be -- it would be complementary for even upgrading the traditional search experience and attract more users.
ChatGPT type of features could potentially become a new traffic entry point for internet users and, therefore, expand the market size of search. Meanwhile, it will also help our advertisers, our content creators, and merchants, etc. With our strong AI capabilities, we should be able to keep iterating and upgrading the model search. Search is just one example.
There are many other applications we believe will also benefit from it. And for AI cloud, our AI cloud provides full stack offering of four layers, from the cloud infrastructure layer to deep learning open-source framework layer, and to the large foundation model layer, and, ultimately, to application. So, there are other four layers. And by opening the generative large language model to the public aka, you know, model and the service, we should help many business owners and entrepreneurs with their own models and applications on our cloud, and bring about a significant positive change and improvement in a number of areas, including increased efficiency, better decision making, and improved customer experiences.
So, to recap, Baidu's strong AI capabilities build our moat in large language model and AI foundation models. In addition to expand the market size for search, we are able to help many, many industries to build their own model, device their own application, and significantly improve their productivity. We believe it will be a game changer for cloud computing. AI is transforming many industries in a big way, and we are super excited about what's to come.
We are building an AI ecosystem around Ernie Bot. As of today, a number of organizations have already decided to integrate Ernie Bot into their products and services. And remember, this is just the beginning of the journey. Thank you.
Regardless, below are what some Chinese have said on ChatGPT and Baidu’s declared ambition in particular. They appear to believe ChatGPT is already great with huge potential, but at least for the Chinese market, a Chinese tech giant may be better positioned - or desired - to develop an alternative.
A Reuters journalist asked about ChatGPT on Friday, Feb. 24, at a Chinese government press conference.
Chen Jiachang, Director-General of the Department of High and New Technology of the Ministry of Science and Technology:
Artificial intelligence (AI) is an emerging technology that has strategic importance and is rapidly becoming a key driver of technological innovation, industrial upgrading, and productivity improvement. One of the important directions for the development of AI is human-machine dialogue based on natural language understanding. ChatGPT is a phenomena application that exemplifies the high level of interaction between humans and machines, showcasing that large language models for natural language have already acquired certain features of general AI. Thus, ChatGPT holds vast possibilities for utilization in diverse sectors.
The development trend of AI can be characterized by two main features. Firstly, AI has a significant, far-reaching and disruptive impact on social and economic development. In particular, AI has introduced new methods and approaches in scientific research, leading to a paradigm shift in fields such as biological science, material science, and energy science. Secondly, AI is experiencing rapid development driven by various factors, including the internet, big data, supercomputing, intelligent computing, and brain and cognitive-related technologies. This has led to breakthroughs in the field and the emergence of the "intelligence +" paradigm in industries such as manufacturing, transportation, and medical care. For example, automated and intelligent processes have significantly improved production efficiency in some dark factories [fully automated factoris with no required human labor]. Furthermore, AI-assisted diagnosis and treatment systems for more than 3,000 common diseases have been implemented in over 27,000 community-level medical institutions, demonstrating the high penetrability of AI in the medical field.
The Chinese government has placed great emphasis on artificial intelligence, as demonstrated by its investment in research and development and the establishment of several organizations dedicated to promoting AI development. In 2017, the State Council issued the "New Generation Artificial Intelligence Development Plan"《新一代人工智能发展规划》, which led to the establishment of the AI Plan Promotion Office 人工智能规划推进办公室, the National New Generation Artificial Intelligence Strategic Advisory Committee 战略咨询委员会, and the National New Generation Artificial Intelligence Governance Specialist Committee 人工智能治理专业委员会. Over the past two years, the government has developed the "National New Generation Artificial Intelligence Governance Principles"《新一代人工智能治理原则》and the "Ethical Standards on the National New Generation Artificial Intelligence"《伦理规范》. Additionally, they have launched major AI technology projects, and created an R&D layout with "basic software and hardware" as its central focus and two supporting components referred to as "wings": "basic theories" and "innovative applications". Furthermore, the government has partnered with leading companies to create open innovation platforms for artificial intelligence. These platforms have enabled the growth of small and medium-sized enterprises and contributed to the advancement of the industry's technological progress.
The Chinese government has also made significant progress in promoting innovation and development, including the establishment of 18 national pilot zones for innovation and development, which encourage the development of application scenarios with local characteristics. This progress has led to several breakthroughs in domestic research, including the "hybrid Tianjic chip" and the proposal of "neuromorphic completeness" by domestic scholars. In addition, Chinese research institutions have made significant strides in developing advanced technologies such as the Chinese Multimodal Pretrainer and the OPT-Omni-Perception Pretrainer, both of which have met international standards.
Moving forward, the Ministry of Science and Technology is committed to supporting AI as a strategic emerging industry and a key driver of economic growth. To achieve this, it will focus on four key initiatives. Firstly, it will develop an open and collaborative AI innovation system that promotes basic theoretical research and drives major technical breakthroughs. Secondly, it will work towards further integrating AI into various economic and societal domains, refining and upgrading technologies in critical application scenarios, and fostering market growth. Thirdly, it will proceed with the establishment of a governance system to ensure AI security and controllability. Fourthly, it will encourage open cooperation in all aspects of AI. Thank you for your attention.
Wang Zhigang, Minister of Science and Technology:
Let me add a few points to the discussion. ChatGPT is indeed quite popular these days. Director-General Chen Jiachang touched upon some relevant information during his address. The journalist raised two questions: how to view ChatGPT and how to respond ethically. In other words, how can we promote development while considering the double-edged sword nature of technology by strengthening ethical norms? ChatGPT itself is a large model which utilizes big data, powerful algorithms, and significant computing power. As mentioned earlier, it has made some advancements in natural language understanding and processing, as well as effective combinations of algorithms, data, and computing power.
In our country, we have taken corresponding measures in terms of ethics for any new technology, including AI technology, to ensure that the development of science and technology is beneficial and poses no harm and to leverage its benefits better.
Professor Yi Weidong, Brown- and UC Berkeley-trained Professor at the University of the Chinese Academy of Sciences, via a state-run news site
Yi said ChatGPT is not just a chatbot but an artificial intelligence application platform based on natural language models. ChatGPT and similar artificial intelligence platforms can liberate human intelligence from general learning and thinking, allowing us to focus on more creative labor. ChatGPT's technological development and comprehensive application are significantly superior to previous generations, providing answers that better match human thinking expression and knowledge understanding.
Yi believes that ChatGPT is an easy-to-use artificial intelligence platform, similar to how a smartphone is an application platform for mobile communication and the internet. People can develop artificial intelligence apps based on ChatGPT or similar platforms to allow artificial intelligence to complete various types of intellectual labor, such as computer programming, paper writing, film and television creation, and scientific experiment design. This will maximize the popularization of artificial intelligence applications, making them a part of everyone's daily life and work and maximizing the release of human creativity.
ChatGPT's current strength even challenges the search field and puts pressure on Google. However, after being tested by hundreds of millions of users, its advantages are mainly in literature, philosophy, and other fields, while errors may occur in mathematics, physics, and chemistry. Moreover, ChatGPT's Chinese question-and-answer level is still inadequate due to the lack of Chinese data models used for training, making people anticipate the "domestic version of ChatGPT."
Yi believes that Baidu faces competition and technological challenges from international technology giants, but due to its research and development investment and technological reserves in the field of artificial intelligence, it has the conditions to win in this new round of revolutionary development in artificial intelligence.
“The state's support for industrial development, combined with long-standing investments in funds and technology by large domestic enterprises, as well as a massive user base, have collectively formed a unique ecosystem. Chinese companies have the potential to seize the opportunity and compete in the artificial intelligence revolution, thereby gaining a significant share of the market. We are confident that China has the ability to surpass the world's advanced level in the field of artificial intelligence applications.”
Wu Hequan, vice president of the Chinese Academy of Engineering from 2002 to 2010, via Economic Information Daily, a newspaper under Xinhua News Agency
Firstly, he notes that Chinese internet companies have increasingly become “platforms,” providing users with various services and capabilities. Secondly, he highlights the innovative models employed by internet platforms that allow for 24/7 work and flexible employment, contributing to employment growth. Thirdly, he highlights the impressive levels of innovation displayed by internet platforms across various industries, not just in business models but also through “hard” technology innovation, talent acquisition, and integration with the real economy. He is confident in the development potential of Chinese internet platform companies, as he believes they represent advanced technology and high technology.
Regarding the birth of ChatGPT, Wu notes its symbolic significance but reminds us that its application is limited, as it requires a large amount of computing power that few companies can afford. He acknowledges that only a few Chinese companies have the computing power to support such technology. Wu believes that Baidu's understanding of Chinese culture and the Chinese market can help them develop AI applications as good as those produced by foreign companies.
Looking to the future, Wu Hequan believes that business users will emphasize the AI capabilities and intelligent service output of cloud service providers. This shift will bring about far-reaching changes to evaluation standards and market patterns. He recommends that enterprises looking to engage in AI application development should focus on combining AI with real industries, rather than just following international companies to create chatbots. Wu highlights that AI has already been widely applied in various fields, including manufacturing, power, industry, and energy, with significant improvements in efficiency and cost savings. He believes that the emergence of automatically generated AI products will effectively help the industry solve pain points and that more industry applications will accelerate their development.
Yu Guoming, a professor at the School of Journalism and Communication of Beijing Normal University and a frequent commentator on China’s media ecosystem, via the website of the Global Times
The breakthrough of ChatGPT lies in its ability to learn and integrate relationship and emotional factors.
Yu believes that the underlying technology of ChatGPT is not complicated, and the research and development logic is relatively straightforward, with unique characteristics in deep learning. From the perspective of technological application, simpler technology tends to be more open and functional, while complex technology has limited boundaries. For example, five or six years ago, there was already the MGC (machine-generated creative) content production method based on artificial intelligence. However, ChatGPT cannot be defined as MGC anymore. Instead, it should be called AIGC (AI-generated content). Yu said that “ChatGPT is a breakthrough technology with high freedom and a wide range of applications.”
Regarding the development of the internet, Yu says that internet companies faced significant difficulties in the past year, but they have accelerated their deployment in the field of artificial intelligence. He emphasizes that artificial intelligence, especially large-scale deep learning, requires not only funding and technical strength but also data accumulation, so it relies on technological continuity. “The company with more experience, better databases, and technical capabilities will have a more advantageous development,” Yu said that “Baidu has seized the initiative in this regard,” as it has already made layouts in the four-layer technology stack of chips, frameworks, models, and applications, and has benefited from it and the related technology of ChatGTP to show comprehensive advantages globally.
Yu predicts that artificial intelligence will accelerate its penetration into all aspects of news and communication. He believes that artificial intelligence will deeply participate in all kinds of media products, media services, and media phenomena that will emerge.
Hu Yong, a professor at the School of Journalism and Communication of Peking University, via Sina, a news portal
artificial intelligence will indeed have a place in content creation. However, instead of using it to replace humans, it is better to consider how it can help enhance people’s writing capabilities.
Hu Yong pointed out that AI has moved from simply understanding language, text, images, and videos to generating content. The essence behind AIGC is that artificial intelligence can become an automated creator. The advantage is that it can reduce human input in the content production process. AI writing tools are capable of creating more content in a shorter period compared to human writers. At present, artificial intelligence is much better at assisting content creation and providing help for human writing. The key is to automate production elements but not to hand them over to artificial intelligence completely. Humans are still doing the heavy work, creating cohesive narratives, and checking the accuracy of things. We do not yet know when artificial intelligence will catch up with or even surpass humans in these areas.
"Baidu is the first internet company in China to announce its ChatGPT-like product. Its product, ERNIE Bot, is based on the ERNIE model, which is widely used in the industry", Hu said, "Baidu has comprehensive advantages in ChatGPT-related technologies in China and even around the world. People should not always belittle Chinese technology companies. China's AI capabilities are worth expecting."
Hu also pointed out that in the future, artificial intelligence will always need human input and professional knowledge to fully unleash its potential in a way that is ethical, responsible, and safe. Artificial intelligence must deal with the relationship between artificiality and intelligence well. As long as there are new applications for artificial intelligence to learn and new tasks to master, it will always need human input. "In the face of the prosperity of ChatGPT, we should actively imagine a future of technology where humans are still indispensable," Hu said.
Now for a taste of some dire warnings, below are from an upcoming paper in a Chinese journalism journal written by Zhou Ting and Pu Cheng. Zhou is the dean of the School of Government and Public Affairs at the Communication University of China, where Pu is a Ph.D. student.
(1) Becoming a tool in cognitive warfare
While technology aims to do good, ChatGPT's advanced deep neural network technology and human-like interactive realism make it vulnerable to weaponization in international information warfare, public opinion warfare, and cognitive warfare. ChatGPT's ability to produce a large volume of structured text that is difficult to distinguish between true and false can affect international public opinion. NewsGuard, a U.S. news credibility assessment and research organization, found that ChatGPT can adapt to questions filled with conspiracy theories and misleading narratives, producing convincing yet unattributable content in seconds.
[NewsGuard: The Next Great Misinformation Superspreader: How ChatGPT Could Spread Toxic Misinformation At Unprecedented Scale]
ChatGPT's reliance on existing language corpora on the internet, with English language information from developed countries in Europe and America in dominance, reinforces certain values and cannot provide a competition of values.
For instance, a netizen asked ChatGPT whether the United States could shoot down a civilian balloon from China if it floated to the United States, and the answer was "yes." However, when asked whether China could shoot down a civilian balloon from the United States if it floated to China, the answer was "no," indicating a "double standard" position. [ https://www.163.com/dy/article/HTG3D7MH0552RR09.html]
The risks associated with ChatGPT's weaponization in cognitive warfare are concerning, and current regulation tools may not be sufficient to counter its impact.
[Zichen: I have tried to reproduce the aforementioned question-and-answer on ChatGPT which was widely circulated in Chinese social media, but couldn’t: ChatGPT doesn’t advise shooting down a civilian U.S. balloon in China or a civilian Chinese balloon in the U.S.]
(2) Prolonging International Conflicts
The internet is filled with massive amounts of unfiltered, biased information, with racial and gender discrimination among the most prevalent. ChatGPT's answers are based on learning from internet information and audience feedback, not only affecting cognition but also reinforcing users' biases. For example, Insider reported that ChatGPT told users it would be okay to torture people from certain minority backgrounds.
Because there is currently no way to conduct pre-auditing of ChatGPT's answers, answers containing serious errors can easily slip through among the massive number of responses. If biased information is massively produced and circulated worldwide through generative artificial intelligence, it will exacerbate international conflicts such as separatism and racial prejudice. Although the AI industry has already realized this downside of the product, the current technology for filtering out negative information is still unreliable. The iterative updates of filtering models still rely on old data, and as more data is fed to ChatGPT, creating biases and suppressing them will become a long-term chase game.
(3) Damaging cybersecurity
ChatGPT's powerful data collection and text generation abilities are a potential catalyst for exacerbating cyber attacks and threatening cybersecurity.
The first threat is the theft of user information. ChatGPT's operation relies on the collection and analysis of vast amounts of internet data, and as the corpus continues to expand, more user information could potentially be stolen and used to train the AI model. The risk of user information leaks exists due to private companies' monopoly on consumer data, as demonstrated in the 2022 Facebook data leak that exposed the information of 500 million users.
The second threat is that ChatGPT can be used to aid criminal hackers. With its ability to quickly generate phishing emails and malicious code, ChatGPT can serve as an easy weapon for hackers to use in carrying out cyber attacks. According to the cybersecurity news website Dark Reading, hackers already use ChatGPT to steal data from large companies, including well-known multinational corporations such as Microsoft, PayPal, Google, and Netflix.
However, current internet regulation tools have not kept pace with the development of AI technology, leaving loopholes for hackers to exploit. As such, it is crucial to strengthen cybersecurity measures and update regulations to keep up with the fast-evolving landscape of AI technology and its potential cybersecurity risks.
[Zichen: The citation of the Dark Reading report is apparently wrong, as the report simply didn’t mention that. What the report says was that ChatGPT — the chatbot that can assist anyone in producing instantaneous, high-volume content that's already become notorious for its cybersecurity implications since its November release by OpenAI — also is becoming a phisher's best friend.
ChatGPT doesn’t learn from user inputs. And while private companies did have huge data leaks, the government did as well.]
(4) Exacerbating Global Digital Inequality
The proliferation of ChatGPT will not only consolidate the market dominance of transnational giants in Europe and America but also exacerbate the digital gap between North and South countries, leading to new forms of international information dissemination inequality and digital exploitation. On the one hand, as a hybrid of Western technology and capital, ChatGPT is conducive to consolidating and expanding the West's international discourse power. Its creator, artificial intelligence enterprise OpenAI, is located in San Francisco, USA, and its co-founders include Tesla founder Elon Musk. In addition, according to a report by (Chinese media outlet) The Paper, Microsoft is negotiating to invest $10 billion in OpenAI to create a new growth point in the search field. On the other hand, ChatGPT is also creating new forms of digital exploitation. Time magazine reported that OpenAI hired Kenyan outsourced laborers who earn less than $2 per hour to manually label the massive database to train ChatGPT - they read and annotated an average of over 20,000 words per hour.
To address the challenges posed by generative artificial intelligence represented by ChatGPT, we need to take two-pronged actions.
Firstly, we need to unite global governance forces to seek fair solutions, and prevent false information, privacy violations, discrimination, digital divides, and other issues. This requires global cooperation to address the risk challenges of cognitive shaping.
Secondly, we should support local enterprises with robust capabilities to develop artificial intelligence content generation platforms, increase the supply of high-quality simplified Chinese content on the Internet, use new technology to open up breakthroughs in speaking about China and cognitive competition and strengthen guidance on the public ability to judge information to prevent public opinion crisis brought about by technological shocks.