Since the outbreak of ChatGPT at the beginning of 2023, many people have been in awe at the artificial intelligence technology’s capabilities: Will AI rule mankind one day? For the publishing industry, the concern is focused on how AI technology will change the content-centric industry. To this end, Bookdao interviewed Du Yu, an author of AIGC: The Era of AI Creation, to answer several questions about how AI creation will affect the publishing industry.
In February this year, China Translation Publishing House published AIGC: The Era of AI Creation, which introduces AIGC's business implementation scenarios and industry applications to all practitioners, entrepreneurs, investors, and government departments who are concerned about future technology.
In this book, the authors also express their openness to the arrival of the era of AI creation: “Although there are concerns, human creations will eventually help human development. AIGC is undoubtedly a change in productivity that will send the world to enter the era of AI creation. In the new era, the improvement of creator productivity is mainly manifested in three aspects: replacing repetitive works for improved efficiency; separating creativity from production, content creators can obtain inspiration and ideas from AI-generated works; the combination of massive pre-trained data and the randomness of AI system expand the boundaries of innovation, and creators can produce outstanding ideas that could not be imagined in the past.”
Even so, AIGC is not perfect. A series of issues such as “copyright ownership of AIGC” and “abuse of AIGC in illegal and criminal activities” are now the focus of debate. At present, academia and industry are trying to solve these problems from various aspects. But no matter what, the rapid development of AIGC has become an irreversible trend, and the prelude to the era of AIGC is slowly opening.
With the support of big data learning, the iterative speed of AIGC is also amazing: from the birth of “The Turing Test”, a test method for machine intelligence measurement, in 1950, it was not until 2017 that the world's first AIGC novel On the Road was created. In 2017, the world's first poetry anthology 100% created by AI was published; in 2020, OpenAI launched GPT-3, which has 175 billion parameters, and has amazing performance in the text translation, question answering, and generation; in 2021, OpenAl launched DALL·E, a text to image generator; at the end of 2022, ChatGPT was launched, and the global active users easily exceeded the 100 million mark in less than two months; in March 2023, OpenAI released a large-scale multi-model GPT-4, with the capability to provide text output from input in both image and text format the latest system gives a human-level performance on various professional and academic benchmarks, once again refreshing the perception of AI.
Bookdao: ChatGPT exploded in popularity at the beginning of the year. Does it mean that AI creation will hit the cultural and technological fields in a big way, or will it disappear like a shooting star?
Du Yu: ChatGPT may be popular for a much longer time than expected. It is a revolution in the productivity of Internet content, and it will also be used as the underlying infrastructure to be connected to various industries at the application level. It is not like the AlphaGO that the application scenarios targeted at the time were so narrow.
As for technological breakthroughs in infrastructure, a certain reaction cycle is required from technology to products to extensive commercial applications. This is not only a matter for application entrepreneurs but also needs time to follow up on the development of related industries and corresponding policies. At this point in time, it is more like when the IOS and Android systems were just popularized and the outline of the mobile Internet era started. This is not going to be popular like a meteor for a while and then disappear.
Bookdao: In 2017, Microsoft’s Bing AI published its first collection of poems, followed by a collection of paintings. What are the breakthroughs and differences between the current ChatGPT and Bing?
Du Yu: The most difference may be its versatility. Bing’s poetry writing is more professional and only used for writing poetry, while ChatGPT can also write novels, scripts, weekly reports, and complete many other things. Of course, the “versatility” here is discussed on the premise of text mode, and painting is a matter of another field. Bing’s painting should be more benchmarked against some of the current AI image-generating tools.
In addition, ChatGPT is more flexible in writing poems, unlike Bing's poetry writing program, which can only trigger intentions, and cannot control some specific details of the generated poems. In ChatGPT, language description can be used to describe the detailed requirements of poetry, and further requirements can be raised after generation, and better poetry works can be written through multiple rounds of dialogue.
Bookdao: What are the most likely usage scenarios for AI creation in the news publishing industry? Based on your observation, which fields are already in use?
Du Yu: The earliest application of AI creation in the news industry was some news reports with a relatively fixed format but high time requirements. This technology is relatively mature, and it has been used in disaster reports such as earthquakes, or financial and sports reports with strong timeliness.
As for the publishing industry, the more mature structured application technology includes structured writing of titles and abstracts, which can be helpful for editors during editing and proofreading. In addition, the emerging AI painting and ChatGPT will bring new possibilities to the publishing industry, such as the realization of book illustrations with the help of AI painting tools, and the use of ChatGPT to assist editors, etc.
Bookdao: For the publishing industry, what pros and cons might AI creation bring?
Du Yu: The most significant positive impact is that the author's creative efficiency will increase, which will greatly benefit the author in terms of creative inspiration and paragraph writing. As a result, there will be more high-quality work in the publishing industry. At the same time, with the improving efficiency of editors with the help of AI tools, the process of publishing is becoming faster.
On the other hand, with AI lowering the creative threshold, many low-quality submissions may also emerge, and the publishing industry may face more difficulty in screening submitted works. In addition, the current laws and regulations need to justify copyright-related issues arising from the use of AI creation.
Bookdao: Foreign periodicals and publishers do not accept authorship for AI creations. Was it copyright or other reasons?
Du Yu: Under the current legal system in many countries, the subject of authorship must be a person or an organization composed of people. Giving authorship to AI creation may add a lot of unnecessary trouble in legal affairs.
Most of the journals that do not accept AI authorship are academic journals. The authorship is not only a right, but also an endorsement of the rigor and academic norms of the article, and there is no doubt that AI cannot do this. If AI generates wrong academic content, or draws on an important document without a reference, who will ultimately be responsible? So now some academic publications will reject manuscripts created with ChatGPT, or require authors to indicate the part created with ChatGPT.
Bookdao: At the Beijing Book Fair, the Publishers Association of China established the Academic Publishing Committee, specifically mentioning that the focus of the next work is to promote the establishment of academic publishing standards in the era of AI. What do you think of academic publishing and even publishing norms in the era of AI, and what aspects should the publishing industry prepare in advance?
Du Yu: First, the establishment of the contribution channel and the establishment of an editorial team dedicated to AI participation in content creation. As AI creation tools are becoming available, more people may use AI for creation. In this case, it may be necessary to distinguish the manuscripts produced by AI from those completely created by humans and form an editorial team specialized in AI creation to review the content.
Second, formulate publication specifications for relevant AI-created content. For example, whether to disclose the part of AI participating in the creation, whether to limit the percentage of content created by AI and so on.
Third, the reform of book categories and distribution channels. Whether it is necessary to add a new category to AI-created content under the current book classification, at which level and where to add it, and whether to add regulations for the distribution channels of AI-created content, etc., are all issues that need to be considered.
Bookdao: AI creation is more about achieving human-like expression through intelligent learning, but it is not human itself after all. How many stages do you think there are still to go from now to completion?
Du Yu: Rather than a few stages, it is a cyclical iterative process. The three important factors that affect AI creation are computing power, algorithms, and data. In each stage, there may be a breakthrough at a single point, and then the other two will quickly follow up, and finally, the iterative performance will infinitely close to the state of human beings.
Bookdao: think the publishing industry has got left to prepare for the era of AI?
Du Yu: I think the publishing industry will always exist, but the working model may be different.
For example, editors in the publishing industry in the past may be more concerned with editing and proofreading author content that is not smooth or does not conform to publishing standards. In the future, editors in the publishing industry may focus on the rationality and creation review of AI-created content. Just like the transformation from paper publications to electronic publications, publishing houses have always existed, but there may be some adjustments and changes in the working mode. Here, I prefer to understand the time left by AI for the publishing industry as the time to keep up with the new era. The publishing industry will not be eliminated by AI, but publishing institutions that cannot keep up with the times may be eliminated.
Bookdao: From the Metaverse in 2022 to AIGC in 2023, what role can the publishing industry play as a content industry in this feast of technology, and to build a profitable business model?
Du Yu: A promoter of scientific and technological knowledge, a discoverer of high-quality authors in related fields, and a facilitator to bring cutting-edge content to the public. The value should not be limited to profit in terms of publication sales, but more importantly, benefits the entire Chinese science and technology industry and the growth of China's younger generation of science and technology.
At present, AIGC's monetization model in B-side services is more feasible. Traditional industries urgently need AIGC technology to reduce costs and increase efficiency. Many companies have a high willingness to pay for technologies that can improve business efficiency or significantly reduce business costs. Moreover, because there are obvious differences in industry and business logic, and the mainstream AIGC models are relatively general if products can be developed for specific business needs, there are still great opportunities. Finding a business scenario that can be implemented, and locking in a segmented scenario to train AIGC and make product differentiation in specific fields, is the best way to commercialize.