By Yuan Yanan
In November 2022, Open AI, an artificial intelligence (AI) research lab headquartered in San Francisco, launched ChatGPT which quickly attracted global attention. Two months since the release, monthly active users (MAU) of ChatGPT have reached 100 million, making it the app with the fastest-growing user count in history.
The emergence of ChatGPT has opened eyes to the huge potential of its future commercial application, but also raised a lot of discussions and concerns. China Report ASEAN interviewed Li Peng, an assistant researcher at the Institute for AI Industry Research of Tsinghua University, for more context on the application and influence of ChatGPT.
China Report ASEAN: What influence has ChatGPT exerted on the technologies of AI Generated Content (AIGC)? Does ChatGPT utilize the highest level of AIGC’s application?
Li Peng: ChatGPT is a game-changer not only for AIGC but the whole field of AI. We are all experts of languages because we speak them every day, which makes us sensitive enough to recognize even the smallest mistakes an intelligent app makes. However, ChatGPT uses language in a way so far beyond expectations that no other apps can now compete with ChatGPT in dialogue with humans.
Soon after the launch of ChatGPT, Microsoft claimed that it would integrate the chatbot’s game-changing technology into its latest version of the Bing search engine and the Microsoft Edge browser. Baidu, a Chinese technology company specializing in internet-related services, confirmed that it would release the ERNIE Bot, a project similar to ChatGPT, for which closed beta testing is expected to finish in March before release. It remains unclear whether ERNIE Bot will perform better, so ChatGPT is absolutely the highest-level application of AIGC.
China Report ASEAN: Compared with other chatbots such as Apple’s Siri and Microsoft’s Bing AI, ChatGPT realized leap-frog improvement. What was at the core of such improvements? Does ChatGPT have shortcomings?
Li Peng: Its technological model sets ChatGPT apart from other chatbots. The AI chatbots before ChatGPT adopted a relatively small model with a limited database. ChatGPT, in contrast, is based on a creative technological model integrating big data, big modeling, and the feedback it receives from users. It enables ChatGPT to perform much better in academic thinking and writing and produce high-quality results when responding to natural language input such as dialogue, translation, and generation of abstracts for theses.
However, ChatGPT has shortcomings, too.
First, it does not have knowledge of recent events and up-to-date information. According to Open AI, ChatGPT only fetches data prior to the year 2021, which means that it may not be able to integrate material generated after 2021.
Second, the smooth responses of ChatGPT may come at the expense of accuracy because it is incapable of fact-checking. The seemingly fluent and credible text generated by ChatGPT could be made up and false.
Third, the current version of ChatGPT does not cite the materials it references, so users cannot read or check the original texts. Therefore, ChatGPT-generated content may lead to copyright infringement and legal issues.
Fourth, the cost of training ChatGPT, including the computing power and data annotation, is huge. And because the model for the chatbot is quite big, the cost of running it is also high. Therefore, developers of ChatGPT must cut costs to make it more lucrative.
Furthermore, ChatGPT lacks knowledge of vertical industries such as entertainment, medical care, and education as well as the capability to deliver linguistic services for low-resource learning.
ChatGPT is undoubtedly of great value for AI application, and its shortcomings will soon be tackled.
China Report ASEAN: Large-scale application of ChatGPT could cause transformative changes in many industries. How should people respond to such changes?
Li Peng: The powerful capability of ChatGPT to deal with natural language is radical and even disruptive. It is changing AI analysis of natural language, and its potential future application could be quite comprehensive. Industries relying on information collection and primary processing without creativity would be most influenced by ChatGPT.
People should be optimistic about the changes to the means of production. Some jobs could be replaced by AI, but new studies and applications will subsequently emerge. For example, text generated by ChatGPT could have defects in credibility. To solve the problem, people need to fact-check the chatbot’s answers. This is how new jobs are created.
Also, application of ChatGPT in professional fields may not be as easy as for regular users, so good potential to develop middleware remains. For example, Photoshop is a highly professional software for image editing and designing, while Meitu Xiuxiu, often seen as a simplified version of Photoshop, empowers regular users to more easily edit photos and images. The market for middleware is big.
So, people need to develop a dialectical attitude towards ChatGPT and the changes it is bringing because each industrial revolution makes some industries disappear and others emerge. Systematic measures are required to respond to such changes during the transitional period.
China Report ASEAN：What human behavior cannot yet be simulated by AI? What future difficulties for AIGC technologies do you anticipate?
Li Peng: AI is still not capable of mimicking people’s creativity in tasks such as generating new knowledge, approaches, or sophisticated thinking. It also lacks the ability to make insightful discoveries and summaries. These are advantages reserved by humankind.
Three difficulties still need to be tackled by future AIGC technologies. First, developers need to solve the problems mentioned above. Second, ChatGPT now only processes text, so developers need to expand its functionality to multimodal tasks such as image-editing and processing of text in specific fields like chemistry and biomedicine. Third, large-scale application of chatbots like ChatGPT may exert a negative impact on knowledge-sharing and creativity. The internet’s clear benefits made people willing to share information and knowledge online, but if they become excessively dependent on ChatGPT, they will become less willing to create content and share information, which could lead to potential problems.