ChatGPT, an artificial intelligence to generate text, images, and music, has garnered global attention since its launch. In March, OpenAI, the company behind ChatGPT, announced the release of a more powerful version of the language model, GPT-4. Search engine company Baidu has also released its similar product. These conversational AI products are expected to be widely used in various fields of social life. With their strong language analysis and content output capabilities, generating text in accordance with user requirements, and large output efficiency, with content that is more normative and grammatically correct, these products are facing numerous copyright controversies that call for immediate attention.

The dispute revolves around whether AI-generated content (AIGC) such as ChatGPT's output qualifies as a work in the sense of copyright law, given that it is generated under the application of algorithms, rules, and templates without human creative input.

In China, there are three different schools of thought on this issue. The first holds that AIGC is a product of algorithms, rules, and templates and does not reflect the spirit of human creativity, and thus cannot be regarded as a work. The second school of thought argues that both the content and form of AIGC are the result of human intervention, reflecting human will or arrangement, and thus possess copyrightability. The third school of thought maintains that the development of machine learning technology will inevitably impact copyright systems, and thus it is necessary to adjust the standards for determining originality or the copyright ownership system to provide comprehensive protection for AIGC. However, there is no consensus yet on whether ChatGPT's output should be subject to copyright protection in China.

It is worth noting that the US Copyright Office recently issued a rule stating that content generated by AI technologies such as ChatGPT and the AI painting chatbot Midjourney is not protected by copyright law, but images created using software such as Photoshop are protectable. The difference lies in whether human creative contribution is involved in the production process.

Under current laws, the issue of whether ChatGPT's output is copyrightable needs to focus on whether it reflects human creative activity and whether that activity can be evaluated as a creative act under copyright law. Claimants must provide evidence of its originality. However, human training and optimization of the ChatGPT model should not be considered as creative work or a factor in determining copyrightability.

Access to data is essential for training and optimizing ChatGPT and similar products. This involves the use of massive amounts of data, including digital works, in the process of data input, training, and optimization. If the developers of ChatGPT and similar products do not obtain individual authorizations from copyright holders, they may face various risks of copyright infringement.

The first risk is the copyright infringement related to reproduction rights. From the perspective of machine learning, ChatGPT and similar products require the collection of massive amounts of input data, forming data copies in natural language processing systems, and repeatedly training the data based on this. If the input data contains digital works or data from protected databases that have not been authorized by the copyright holders, this behavior may constitute a violation of reproduction rights.

There is also the risk of copyright infringement related to the right of information dissemination. ChatGPT's content is generated in an open, public, and interactive network environment. If the main text generated by ChatGPT comes entirely from the original expression of one or several pre-existing works, the text may infringe on the original author's right of information dissemination.

ChatGPT and similar products are facing potential copyright infringement risks due to their use of training data, including copying, translating, adapting, and compiling existing works. This could lead to generated content that incorporates the basic expression of earlier works, constituting derivative works that violate the rights of the original creators. However, determining the legal status of such AI-generated content could stifle innovation and lead to massive lawsuits.

To balance the need to promote industry development and prevent legal risks, it is suggested that "text and data mining" be classified as a specific type of reasonable use of works. This is because traditional copyright licensing models cannot meet the demand for the scale-based use of digital works in the AI era, and AI's knowledge enhancement and original work value are mutually independent. Therefore, AI's use of existing digital works for deep learning does not harm the legitimate rights and interests of copyright owners.

While the emergence and commercial application of ChatGPT and similar products will lead to changes in related business models and improve economic and social benefits, legal risks should also be taken into account. To address this, a comprehensive intellectual property risk prevention and control mechanism should be established to facilitate the healthy and sustainable development of the AI industry.