Chat Generative Pre-trained Transformer (ChatGPT), developed by OpenAI, is an artificial intelligence (AI) large language model that was launched on November 30, 2022. Within just a week of its release, it had over 1 million users, and by the end of January 2023, its monthly active users had surpassed 100 million, making it the fastest-growing consumer application in history and a phenomenal product globally.
Currently, ChatGPT is no longer limited to basic functions such as chat Q&A and entertainment interaction; it has evolved into an efficient productivity assistant tool. Users can achieve intelligent generation of emails, reports, translations, code, and more through chat interactions.
In the long run, the emergence of ChatGPT not only represents a breakthrough in the new generation of chatbots and generative AI but will also bring disruptive changes to the AI industry and the digital transformation of various sectors, thereby reconstructing the paradigm of industry knowledge model building and promoting the scaling and marketization of knowledge-intensive service industries.
This article analyzes the significance and impact of ChatGPT, its application prospects, and proposes governance strategies for the challenges it may face.
Why ChatGPT Became So Popular
The rapid popularity of ChatGPT in a short period can be attributed to its role in triggering a technological revolution, mainly reflected in three aspects.
(1) ChatGPT is seen as a milestone product
The emergence of ChatGPT reignited interest in the AI sector, with industry experts describing its birth as the “iPhone moment” or “Windows system” of the AI era, and it may even be compared to Google’s Go AI program AlphaGo developed in 2016. Microsoft founder Bill Gates praised the significance of ChatGPT, comparing it to the birth of personal computers and the internet. OpenAI co-founder and Tesla CEO Elon Musk publicly stated that “ChatGPT is frighteningly good,” while 360 Group founder Zhou Hong even believes that “ChatGPT is the singularity of general AI development and the inflection point for strong AI to come.”
Specifically, ChatGPT drives disruptive changes in content generation, knowledge creation, information distribution, and retrieval, addressing many pain points of users through extremely simple natural language interactions, further opening new spaces and imaginations for the implementation of AI business models and the digital transformation of knowledge service industries. Thus, the emergence of ChatGPT may signify a revolutionary turning point in the AI field.
(2) ChatGPT opens the path to general AI
Compared to previous AI language models, ChatGPT features multiple technological characteristics such as big data, high computing power, and strong algorithms. In terms of functionality, ChatGPT possesses human-like understanding and reasoning abilities regarding language, context, intent, and logic, demonstrating outstanding performance in high-quality, multi-turn, long-text human-machine interaction dialogues, and has achieved preliminary “intelligent emergence.” At the same time, ChatGPT employs reinforcement learning from human feedback (RLHF), allowing for continuous iterative upgrades through user interactions. Overall, these advanced technological features and powerful interactive capabilities of ChatGPT’s application are propelling human society from specialized AI targeting specific fields and vertical scenarios toward general AI.
(3) ChatGPT will disrupt traditional interaction models
In contrast to the knowledge service model of structured retrieval via search engine keywords in the internet era, ChatGPT shifts to natural language instruction-based knowledge retrieval in the era of large models, transforming interaction from graphical interface touch-based to natural language dialogue-based, heralding the arrival of the natural language interaction era, making human-machine interaction more natural, efficient, and user-friendly.
In the future, people will be able to interact with intelligent machines through natural dialogue, and ChatGPT will meet user needs by accurately and efficiently understanding user intentions and invoking various service functions, thereby changing the current interaction methods of application software. Meanwhile, with the release of GPT-4, ChatGPT will expand from the early single-modal text to multi-modal interactions including audio, video, and images, further promoting the evolution of human-machine interaction from early single-function, single-scenario, single-modal interaction to multi-functional, multi-scenario, multi-modal dialogue interaction.
Trends in Industrial Digital Transformation Driven by ChatGPT
The practical application of ChatGPT in industries will greatly empower the digital transformation and upgrading of industries. A new round of industrial change is becoming an inevitable trend, including accelerating the industrialization of AI, leading AIGC (Artificial Intelligence Generated Content) into an era of “democratization,” promoting human-machine division of labor as the dominant employment paradigm, and expanding AI business value to consumer applications.
(1) ChatGPT accelerates AI industrial layout significantly
ChatGPT has prompted internet giants both domestically and internationally to rush into the field, facilitating the accelerated development of the upstream and downstream industrial chain of AI. For example, Google announced the launch of Bard, Meta released the open-source large model LLaMA, Baidu introduced “Wenxin Yiyan,” JD announced the launch of ChatJD, and major internet companies such as Huawei, Tencent, ByteDance, Alibaba, and NetEase have also indicated that they have laid out ChatGPT-like projects. OpenAI announced on March 1, 2023, that it would officially open the ChatGPT and Whisper model APIs (Application Programming Interface) to the public, allowing developers to integrate the ChatGPT and Whisper models into their applications and products through the API, which will greatly stimulate the demand for computing power in industrial practice and further spur explosive growth in computing power-related industries. At the same time, ChatGPT has accelerated the development of data centers and chip technology research and will trigger a boom in building platforms and related applications, with new industry giants, peripheral industries, ChatGPT-like applications, and the market for intellectual service applications all set to benefit.
(2) ChatGPT pushes AIGC into the era of “democratization”
As a phenomenal product and application of AIGC, ChatGPT may promote disruptive changes in traditional content production methods, significantly lowering the barriers to content creation, and pushing AIGC toward an era of “democratization” with the potential for explosive growth. Compared to traditional AI robots, ChatGPT has stronger language and context understanding capabilities, can respond in context, and continuously optimizes creative content based on user interactions. ChatGPT will change the traditional purely human content generation method, gradually transforming AI from an early creative assistant tool to the main body of content creation, achieving human-machine collaborative content creation and significantly improving content production efficiency and quality. At the same time, ChatGPT will promote AIGC from early text generation to gradually expanding into multi-modal content generation including copywriting, voice, images, video, and code, further enriching the AIGC industry ecosystem.
(3) ChatGPT promotes human-machine division of labor as the dominant employment paradigm
ChatGPT can conduct reinforcement learning based on human feedback, facilitating the evolution of AI from a labor assistance tool to an initial entity with autonomy capable of collaborating and interacting with humans, thereby achieving efficient division of labor and collaboration between humans and machines. Meanwhile, ChatGPT will promote the deep refinement of human intellectual labor, stripping away a large number of repetitive, structured, and highly experience-dependent tasks and delegating them to ChatGPT, accelerating the arrival of the fourth social division of labor era represented by human-machine division of labor, where collaborative human-machine mixed labor will become the dominant employment paradigm in the future. Additionally, ChatGPT may disrupt the old competition model in the labor market, posing a “dimensionality reduction attack” on competitors lacking tool empowerment. Humans should become “teammates” with ChatGPT.
(4) ChatGPT expands AI commercial value to consumer applications
Existing AI products primarily focus on B-end (business) and G-end (government) applications, while attempts at C-end (consumer) applications have been ongoing but lack commercially viable projects that most people are willing to pay for. The launch of ChatGPT has broken this deadlock and brought hope to the AI industry, extending the commercial value of AI from traditional B and G ends to C ends, and achieving commercial model implementation through pay-per-use, SaaS subscription models, integrated solutions with value-added services, and advertising monetization based on user traffic. For instance, OpenAI officially launched the paid version of ChatGPT Plus in February 2023, providing consumers with value-added services such as queue-free access during peak hours, quick responses, priority access to new features and improvements, and opening the API interface, all of which can accelerate the commercial value realization of ChatGPT to some extent and provide funding support for its continuous iterative upgrades.
ChatGPT ignites changes in industrial intelligent transformation scenario models
The practical application of ChatGPT in the field of industrial intelligence will build new bridges for structuring industry knowledge models, thus igniting changes in knowledge management scenario models, which can be specifically reflected in several aspects such as using ChatGPT as an auxiliary tool, enabling ChatGPT to autonomously invoke external tools, and providing interfaces for industry knowledge modeling.
(1) Using ChatGPT as an auxiliary tool to achieve human-machine knowledge co-creation
ChatGPT already possesses many “hard” skills such as information retrieval, processing, and content creation, but it still lacks “soft” skills such as innovation and emotional perception compared to humans. Therefore, ChatGPT can be utilized as an auxiliary tool rather than a replacement or competitor, promoting the integration of human and machine intelligence in various fields such as office work and creation.
Using ChatGPT to partially liberate humans, allowing them to focus on more value-creating work. For example, ChatGPT can assist in completing mechanical and repetitive tasks such as writing emails and retrieving information, saving time to focus on more innovative tasks.
2. Content creation scenarios
Using ChatGPT to enhance and empower humans, creating a perfect partnership between the machine’s “knowledge” and human creativity, where humans provide content ideas and requirements, ChatGPT offers inspiration or generates initial drafts, and humans then intervene for refinement. This collaborative creation method will significantly outperform either purely human or purely ChatGPT efforts.
(2) Allowing ChatGPT to autonomously leverage external knowledge tools
Although ChatGPT has powerful capabilities, it is not omnipotent and does not need to be a universal tool. For instance, Meta’s research team emphasized enabling “large language models to learn to use tools,” meaning that when ChatGPT encounters specific tasks, it can respond to human instructions and act as a dispatcher to directly call the APIs of the tools it needs, thereby providing a new bridge for AI to invoke external tools and meet specific task requirements. Under this approach, ChatGPT can integrate with external tools, quickly match and invoke toolkits, expand its functional scope, and lower the barriers to AI usage. Additionally, ChatGPT can be connected with industrial and service robots to enrich interaction interfaces, enhance interaction performance, and improve human-machine interaction experiences.
(3) ChatGPT provides a new interface for industry knowledge modeling
Compared to the keyword-structured retrieval knowledge service model of search engines in the internet era, ChatGPT shifts toward natural language instruction-based knowledge invocation in the era of large models, transitioning from graphical interface touch interactions to natural language dialogue interactions, making human-machine interaction more natural, efficient, and user-friendly. Moreover, ChatGPT breaks through the traditional model of knowledge extraction by industry experts, enabling efficient modeling and organization through natural language interactions. By forming human-machine teams with human experts, it can achieve industry knowledge modeling. ChatGPT can assist in organizing literature and materials, and through timely feedback from human experts combined with ChatGPT’s reinforcement learning mechanism based on feedback, it can rapidly generate and optimize industry knowledge models. For example, in the medical field, ChatGPT can interact with human doctors continuously, promoting the modeling of professional knowledge in diagnosis, treatment, and other areas, enabling ChatGPT to assist in intelligent triage, preliminary diagnosis, solution provision, and auxiliary treatment.
ChatGPT promotes the scaling of knowledge-intensive service industries
The traditional accumulation of industry knowledge requires experts to extract and transform it, and specialized knowledge services in niche fields can only be provided by domain experts, which incurs high training and hiring costs, is talent-scarce, and cannot be scaled. Meanwhile, traditional intellectual services such as consulting cannot achieve scalable customization for consumer markets. However, as an efficient digital knowledge service tool, ChatGPT has initially realized the coexistence of general large models and specialized small models in the knowledge domain, and is gradually replacing authoritative experts through the model of “technicians + ChatGPT + data,” making the scaling, marketization, and personalization of traditional knowledge-intensive services, with marginal costs approaching zero, possible. This allows for the application of ChatGPT to empower the digital transformation of knowledge-intensive services and the commercialization of intellectual services in specialized scenarios.
On one hand, ChatGPT can be applied as a productivity tool to assist in organizing existing industry knowledge and forming specialized knowledge bases, such as quickly extracting and invoking professional knowledge through natural language interactions, thus allowing ChatGPT to become an AI consultant in specialized fields and serve as a human “external brain” to empower efficient operations in professional scenarios. On the other hand, through sensors and other means, daily office work, operational processes, and individual behavior data can be tracked and accumulated in real time, which can be refined, condensed, and promoted to generate knowledge models through modeling, training, and optimization. Additionally, ChatGPT can expand the application of intellectual services to consumer ends, fully mobilizing users, practitioners, and others to participate in the correction and training of industry data, forming a co-creation data training ecosystem and achieving the industrialization of intellectual services in specialized fields.
Challenges and Governance Considerations of ChatGPT Applications
(1) Challenges faced by ChatGPT applications
From a technical perspective, ChatGPT currently faces multiple limitations, including high costs (including hardware procurement, model training, and daily operations), accuracy (sometimes “seriously talking nonsense”), completeness (knowledge gaps due to lack of internet access), reliability (difficulty verifying specialized knowledge), lack of algorithm interpretability and transparency, and algorithmic biases and discrimination. Moreover, it faces multiple challenges in practical applications.
Firstly, ChatGPT may lead to “machine currency” driving out “human currency,” meaning that content generated by machines could potentially overshadow content produced by human civilization, leading to a proliferation of machine-generated falsehoods that obscure the truth.
Secondly, ChatGPT may challenge the existing scientific research and education systems, including impacts on basic education systems and resulting in academic misconduct and cheating.
Thirdly, the copyright of content generated by ChatGPT is difficult to define. On one hand, if generated content is copyrighted, it may lead to some form of content monopoly, resulting in copyright conflicts (including between humans and machines, as well as between different machines). On the other hand, if there is no copyright at all, it completely disregards the contributions of model developers, which would undermine the motivation for model development and innovation.
Fourthly, user privacy protection faces a “dilemma.” User data can only release its practical value through sharing and analysis, and the larger and more diverse the data, the more beneficial it is for building large models, but excessive data mining may infringe on user personal privacy, thus facing the “dilemma” of data value mining versus user privacy protection.
Fifthly, a gray industrial chain phenomenon has emerged. Since ChatGPT is not yet open to the Chinese market, a series of gray industrial chain phenomena have arisen, such as building mirror products, buying and selling ChatGPT accounts, and reselling ChatGPT-generated content.
In addition, ChatGPT may also bring challenges such as fake news, fraud, network and data security risks, digital divides, and structural unemployment.
(2) Governance strategies for ChatGPT
As an emerging technology, ChatGPT faces numerous challenges in its current industrial practice applications, and future governance responses could consider the following aspects.
Firstly, identify potential risks of ChatGPT in different application scenarios and categorize them for governance based on risk levels.
Secondly, add digital watermarks or identification markers to ChatGPT-generated content products to reduce negative impacts and related regulatory risks arising from misuse of the model.
Thirdly, adopt an agile governance approach, objectively viewing the innovations of ChatGPT with an open and inclusive attitude, continuously tracking its development dynamics, allowing for exploratory applications in some industrial fields within controllable limits, and insisting on a balanced approach between regulatory compliance and innovative development.
Fourthly, use technical means including AI countermeasures and identification of ChatGPT-generated content, and actively explore accountability mechanisms and behavioral norms for generative AI, ensuring that “who generates, who is responsible” and making the product usage trajectory traceable to individuals, leading technology toward good and promoting human-machine symbiosis.
[This article is a research output supported by the China Postdoctoral Science Foundation project “Human-Machine Value Coexistence in Manufacturing Enterprises: Influencing Factors, Driving Paths, and Evolution Mechanisms” (Grant No: 2023M731900) and the major project of the Science and Technology Innovation 2030 “Governance Challenges and Strategies for Artificial Intelligence in Key Areas” (Grant No: 2020AAA0105300).]
AuthorUnit: Tsinghua University School of Public Administration; Tsinghua University Institute for AI International Governance
Source: “China Development Observation” 2023 Issue 6
Image Source: China News Service
Content copyright belongs to “China Development Observation” magazine. Please indicate the source and original author when reprinting.
Click the card below to follow and spread the “China Development Observation” WeChat public account (ID: ichinado). We will strive to provide valuable returns.
“China Development Observation” Magazine:
Room 7, Block A, Yicheng Wealth Center, No. 22 Ronghua Middle Road, Beijing Economic and Technological Development Zone (Postal code: 100176)
Website: www.chinado.cn
Submission Email: [email protected]
WeChat Public Account: ichinado (China Development Observation)
Phone: 010-82097226 68353228 88335431
Get more think tank information,
Welcome to follow the “China Development Press” WeChat public account:
