Ethical Development of Artificial Intelligence - China's Regulatory Approach and Corporate Compliance Recommendations (Part 1)
Editor's Note:
"What's the hottest trend after the metaverse? Undoubtedly, the answer is ChatGPT.
By the end of January 2023, ChatGPT surpassed 100 million monthly active users, becoming the fastest-growing consumer application in history. This news quickly ignited the ICT industry and the capital market, sparking a profound question for workers: "Will ChatGPT replace you?"
While people were curious and willing to be "guinea pigs," the past two months witnessed internet giants and leading startups launching their own large-scale AI-generated content (AIGC) products, creating a sense of an AI arms race.
Do you know:
- What is the relationship between ChatGPT and AI, AGI, and AIGC as different forms of artificial intelligence, and how has the paradigm shifted?
- How does the global and domestic landscape of AI big data models reflect the new "land-grab" movement?
- What are the similarities and differences in the regulatory landscape and priorities for AIGC development in various countries, including China?
- From the perspective of corporate compliance, what are the key considerations in terms of data, content, licenses, and infrastructure?
Centered around the above questions, the Public Policy team at Sinobravo is pleased to present the "Compliance Series: Ethical Development of Artificial Intelligence - China's Regulatory Approach and Corporate Compliance Recommendations." This two-part series shares our observations and insights into the AIGC field.
Concepts and Paradigm Shift in the Development of Artificial Intelligence
Firstly, let's clarify the meanings of the English abbreviations frequently heard recently:
Artificial Intelligence (AI): A branch of computer science that explores how to build intelligent computer systems. These systems are capable of understanding, learning, and solving problems, achieving human-like intelligence. AI can be categorized into Narrow Artificial Intelligence (also known as weak AI) and Artificial General Intelligence (also known as strong AI).
Artificial General Intelligence (AGI): Refers to computer systems capable of performing any intellectual task at a human level. AGI systems can flexibly apply knowledge across various tasks and domains, possessing abilities for reasoning, learning, planning, and adapting to new situations. In contrast, Narrow Artificial Intelligence excels only in specific task domains, lacking generality.
Artificial Intelligence Content Generation (AIGC): The process of using AI technology to generate content such as text, images, audio, and video. AIGC finds applications in various scenarios, including news writing, creative writing, and advertising design. AIGC typically relies on large datasets and advanced machine learning techniques like deep learning and Generative Adversarial Networks (GANs) to generate realistic and appealing content.
ChatGPT: An AI language model based on the Generative Pre-trained Transformer (GPT) architecture developed by OpenAI. It can understand and generate human-like text, suitable for various AIGC applications. For instance, ChatGPT provides valuable support in customer support, content creation, education, and translation.
Through the analysis of these concepts, we can understand the close relationship between Artificial Intelligence (AI), Artificial General Intelligence (AGI), Artificial Intelligence Content Generation (AIGC), and ChatGPT.
Specifically:
AI as the foundation: The development of AI technology drives progress in areas like AGI, AIGC, and ChatGPT. Research in AGI and the application of AIGC both stem from the continuous expansion and deepening of AI technology.
AGI as the goal: Researchers in the field of artificial intelligence strive to achieve AGI with human-level intelligence. Although complete AGI has yet to be realized, advanced AI systems such as ChatGPT have already demonstrated the potential for generality and adaptability.
AIGC as the application: AI technology applied in the AIGC field provides users with powerful content generation capabilities. AIGC technology helps users save time, improve efficiency, and spark creativity.
ChatGPT as an example: ChatGPT is an excellent representative in the AIGC field, leveraging advancements from the AGI domain to provide users with robust language understanding and generation capabilities. ChatGPT can play a role in numerous applications, such as article writing, code generation, and question answering.
New "land-grab" movement in AI big data models
No "opportunity" arises out of thin air, and the same applies to AIGC, which did not bloom overnight. In fact, in recent years, major players in both the Western and Chinese markets have been engaged in a land-grab movement in the field of AI, each excelling in their own arenas. Let's briefly review the situation of the main international and domestic players:
Global Scope: Microsoft (OpenAI) takes the lead, while Google, Meta, and Amazon join the fray.
Microsoft-OpenAI leads the global AI innovation trend with the GPT mega model. OpenAI was established in 2015 with a $1 billion investment from Elon Musk, Sam Altman (President of Y Combinator), and others. As early as 2019, Microsoft invested $1 billion in OpenAI, and the two collaborated to develop AI technology for Azure cloud platform services. On January 23, 2023, Microsoft announced additional investments in OpenAI worth billions of dollars. On February 2, Microsoft announced the integration of ChatGPT into all its products, including the Bing search engine and the Office suite. CEO Nadella announced that ChatGPT would also be integrated into the Azure cloud computing platform, giving every Microsoft product the same AI capability.
Google, Meta, and others accelerate the deployment of AI mega models and release ChatGPT-like products.
In February 2023, Google released Bard, based on LaMDA, to directly compete with ChatGPT. On March 21, 2023, Google announced the official launch of Bard's public beta, starting with the United States and the United Kingdom. Bard's LaMDA model was trained on 137 billion parameters last year, while PaLM reached a training scale of 540 billion parameters. In May 2022, MetaAI announced the release of the OPT-175B super large model based on 175 billion parameters, which is open to the entire community for free. On February 24, 2023, Meta announced the upcoming release of "Large Language Model Meta AI" (LLaMA) for the research community, aimed at promoting miniaturization and democratization of LLM research.
Following Microsoft and Google, another giant, Amazon Web Services (AWS), announced its entry into the fierce AI competition. On April 13, AWS announced the launch of the Bedrock generative AI service and its own large language model Titan. According to AWS, the released Bedrock model will have the ability to access a range of powerful text and image FM, including Amazon's Titan FM, which consists of two new LLMs, through scalable, reliable, and secure AWS-hosted services.
Domestic Market: Baidu and Huawei take the lead, while Tencent and Alibaba accelerate their deployment.
Baidu: The Wenxin ecosystem rapidly expands and releases the B-end exclusive mega model "Wenxin Qianfan." Baidu has comprehensively laid out AIGC and officially released "Wenxin Yiyan" in March. As early as March 2019, Baidu released the pre-trained model ERNIE 1.0, establishing a three-tier model system of "Foundation + Task + Industry" for the Wenxin mega model. On March 16, Baidu officially released the pre-training, generative large language model "Wenxin Yiyan." On March 27, Baidu officially launched the "Wenxin Qianfan" mega model platform, the world's first all-in-one enterprise-level mega model platform.
Huawei: PanGu, Huawei's large-scale model, has a natural advantage in application areas. Huawei Cloud team initiated the AI large-scale model project in 2020 and released PanGu in April 2021. PanGu has developed a mature system consisting of three stages: basic large-scale models (L0), industry-specific models (L1), and industry-specific scenario models (L2). In terms of technology, PanGu is the world's first Chinese language (NLP) pre-training model with a trillion parameters and 40TB of training data. It better understands Chinese language grammar and semantics, thereby improving model accuracy and quality. Huawei plans to release an upgraded version of PanGu in April.
Alibaba: Tongyi large-scale model and Alibaba Chatbot complement each other. In September 2022, Alibaba officially released the Tongyi large-scale model, which enhances the model's cross-modal capability and efficiency through the unified learning paradigm M6-OFA and modular design. Currently, Alibaba's Tongyi large-scale model serves over 200 scenarios, including cross-modal search in e-commerce, AI-assisted design, open-domain human-computer dialogue, legal document learning, and medical text understanding. In February 2023, Alibaba's DAMO Academy announced that Alibaba's version of the Chatbot, ChatGPT, is under development and currently in the beta testing phase. It will be deeply integrated with DingTalk.
Tencent: Hunyuan AI large-scale model and AI Chatbot are equally outstanding. In April 2022, Tencent first revealed the progress of its Hunyuan AI large-scale model and achieved first place in the authoritative cross-modal video retrieval datasets, including MSR-VTT, MSVD, LSMDC, DiDeMo, and ActivityNet. On March 30, Tencent's Vice President and CEO of Cloud and Smart Industries, Tang Daosheng, announced that Tencent is developing an AI Chatbot, which may be integrated into QQ, WeChat, and Tencent Cloud (serving B2B customers).
Apart from Tencent and Alibaba, other A-share listed companies such as 360, iFLYTEK, and Yuncong Technology are also planning or implementing AI large-scale models:
360: The company is committed to having its own general large-scale model technology and implementing commercialized scenarios based on this technology. On March 29, at the "2023 Digital Security Development and Summit Forum," the founder of 360, Zhou Hongyi, showcased the company's large-scale language model product on the 360 search engine for the first time.
iFLYTEK: At the Shanghai-Hong Kong Innovation Development Forum held on March 31, iFLYTEK's President, Wu Xiaoru, stated that the company has been preparing for some time and will soon release its large-scale model technology.
Yuncong Technology: On the evening of March 30, the company disclosed a fundraising announcement, intending to raise no more than 3.635 billion yuan, which will be used entirely for the development of Yuncong's "Industry Genius" large-scale model project.
-End-