The development of human history is filled with disruptive technologies, each revolution causing a new round of panic, which we can temporarily call "the growing pains of generational change."
Just as when the spinning jenny was invented, many textile workers faced the fear of being replaced and gathered at night to try to destroy this tool that could liberate human productivity.
They resisted, but could not stop the wheels of technology from rolling forward.
The same story is happening again in the AI era. Since July, Tomato Novel's AI training supplementary protocol has attracted attention, causing many authors to post in opposition to AI training. Subsequently, Tomato Novel launched a feature to opt out of the relevant terms, in response to creators' demands.
In fact, since ChatGPT went viral, people's anxiety and fear about AI have never ceased. This year, as generative AI accelerates its penetration into various industries, this anxiety has become more specific, with artists, filmmakers, web novelists, and taxi drivers all sending signals in the name of their groups.
At present, perhaps we should discuss how people caught in the cracks of the era can better navigate this transition period. How should technology providers and platforms help people leverage AI to become truly "super individuals"?
AI Writing: Can "Assist" but Can't "Output"
In this AI training controversy, there was an interesting episode: a netizen discovered that an author on a reading platform had written over 200 novels in 3 months, a speed that was likely AI-generated content. However, looking at the readership, almost none of the books had normal reading numbers.
Casually browsing through these templated titles and settings, the anxiety of being replaced by AI writing seemed to lessen suddenly. After all, AI writing without a human soul is like a slacker who loves to pad word counts, completely lacking the ability to free human hands and consciously work to earn money for people.
However, the fact that AI can't be a good ghostwriter doesn't mean it can't be a good assistant. Those who use AI to assist in writing have already won big.
Last October, a novel called "Land of Machine Memories," "100% created by AI," won second prize in the Jiangsu Youth Science Fiction Writing Contest. This novel was created by Shen Yang, a journalism professor at Tsinghua University, through a dialogue format, gradually prompting AI to create. He specified a Kafka style, and after 66 dialogues, output about 43,061 characters, from which he copied 5,915 characters to complete the competition version of "Land of Machine Memories."
Similarly, in January this year, 33-year-old Japanese writer Rie Kudan won Japan's top literary award, the "Akutagawa Prize," with her novel "Tokyo Resonance Tower." She mentioned that 5% of the award-winning work was directly generated by AI, and that she often communicates with AI outside of writing, confiding thoughts she "cannot discuss with anyone else."
Chen Qiufan, vice chairman of the Science Fiction Literature Committee of the China Writers Association, who won the "Galaxy Award," China's highest science fiction award, also mentioned that he is currently using AI to help with creation, seeing AI as an "assistant" that can discuss creation 24 hours a day, greatly improving writing efficiency.
In terms of subject matter, science fiction is mainly fictional. If human authors are more likely to be constrained by the physical world, then AI's imaginative and occasionally incoherent combinations might bring authors some inspiration that breaks conventions.
However, compared to science fiction, AI might be better at more structured content like mystery novels, web novels, and short dramas.
Taking mystery novels as an example, authors can use prompts to determine the core storyline, suspense points, character settings, and conflicts, like pinning key nodes on a whiteboard. As for how to connect these nodes and how to render specific scenes and atmospheres, this can be left to AI to complete.
The same applies to web novels. Genres like fantasy, xianxia, and rebirth are relatively fixed, and unlike literary creation, web novels have strong commercial attributes. Authors earn income through paid reading and tips, so content creation is more likely to cater to readers' and market demands, leading to popular and templated themes, structures, and worldviews in web novels.
The intervention of AI can precisely help creators deal with this templated content.
For example, the "AI naming" feature in Tomato's AI writing tool can help authors name characters and items when conceiving novel characters. A screenwriter once told Silicon Star that naming is a headache in daily work, but now they can just feed the character settings to AI, and AI can generate ten names in seconds for them to choose from.
Similar to character names, there are specific settings for skills, sects, treasures, and monsters in fantasy novels. It might take an author flipping through the entire "Classic of Mountains and Seas" and pondering until they're bald to come up with expressions differentiated from other novels. Now, this can be left to AI to generate, with the author only responsible for reviewing.
In addition, in the early stages of writing, authors' research work is also very tedious. When the protagonist is a professional like a doctor, police officer, or professor, the author's lack of relevant knowledge can easily lead to unrealistic creation and plot holes. AI's knowledge base and search capabilities can help authors eliminate knowledge blind spots.
In this process, the reason why AI is only an assistant and not a replacement is mainly because, from a technical perspective, the content generated by AI is based on the probability distribution of existing data, not true creation.
According to the learned probability distribution, AI predicts the next most likely word or phrase based on the generated text sequence (i.e., the previous text). In this process, the AI model is only combining and expanding within the existing knowledge framework.
But creating entirely new concepts or ideas is precisely the most important aspect of literary creation. Just like those 200 books that no one cares about, this means that without human prompts, pure AI writing without a soul may struggle to move forward.
How to Get Through the Growing Pains of AI?
However, even with just an auxiliary role, the current anxiety of creators is unavoidable. In fact, each round of new technology diffusion relies on the passage of time to reduce people's resistance.
In 1962, Professor Rogers summarized the basic laws of innovation diffusion in a social system in his book "Diffusion of Innovations" and proposed the S-curve theory of innovation diffusion.
The curve shows that in the early stages of new technology development, acceptance is low due to various reasons. As the penetration rate increases, once a certain critical point is broken, the speed of penetration rate increase will accelerate. When the penetration rate exceeds 40%-50%, the speed of increase will gradually slow down.
This pattern has been repeatedly confirmed in scenarios such as smartphones, ERP systems, mobile banking, and e-commerce for small and medium-sized enterprises. AI, as a disruptive new technology, is no exception. According to the curve, considering that the current technology is still not mature, it hasn't even truly entered the diffusion stage.
Currently, Tomato Novel is not the only web novel platform in China that has set up AIGC tools. Various platforms have successively joined the queue to promote AI writing.
As early as July last year, China Literature Group launched the industry's first large model "Yuewen Miaobi" and the writer's assistant Miaobi version; the Qimao platform also cooperated with Baidu's "Wenxin Yiyan" to provide authors with "AI assistants" and other related auxiliary writing functions...
For platforms, AI may reconstruct the relationship between writing and reading, and is seen by web novel platforms as a ticket to the next era. Therefore, how to help creators use AI faster and better has become a question that various platforms and technology providers need to consider.
One major reason why AI continues to be stuck in industrial reform is the authors' resistance to AI creation.
Just as early adopters in innovation diffusion are often those closer to the technology, it's quite interesting that science fiction authors were the first to accept AI-assisted creation. Therefore, in the process of promoting AI writing, platforms can start pilot programs from science fiction categories and top authors, based on different groups' acceptance of new technology.
Secondly, AI tools have a usage threshold, and individual usage effects will affect the acceptance of AI. Generally speaking, the richer an individual's knowledge reserve and the more complete their thinking, the more useful AI will seem. After all, AI cannot create out of thin air. When the creator's prompt is unclear, AI is just a vast information repository, unable to extract characters and patches from massive information to create exquisite novels or artworks. This requires platforms to collaborate with creators to continuously improve the usability of tools, allowing creators to truly feel the liberation of productivity brought by AI.
In addition, the recent Tomato protocol incident has also brought issues such as AIGC legislation and AI data norms to the forefront.
Although most people's attention is focused on the web novel platforms themselves, in reality, due to the prevalence of pirated books, most creative content can be technically crawled outside of web novel platforms. This is an underwater crisis faced by creators outside of web novel platforms.
In the context of a large number of articles and works not being protected, how to define AI generation and AI plagiarism, and how to protect the copyrights of AI creators and ordinary creators are all issues that need to be addressed by relevant rules.
From the perspective of data value and business models, whether fees need to be paid for AI to use works that crystallize creators' labor, or whether this should be exchanged for the right to beta test AI creation tools, or whether to limit the scope of data that can be used to train AI... These issues all need consensus between platforms and creators.
Overseas, platforms like Reddit have started charging data usage fees to companies that train AI using their API. For example, Reddit announced it will charge data usage fees to companies including Microsoft, Google, and OpenAI.
In the process of promoting AI reform, platforms need to find a balance between protecting creators' rights and AI development. This not only concerns the development of the platform itself but also the future of the entire industry.