以ChatGPT为代表的生成式人工智能(AIGC)的出现,引发了人工智能技术和版权保护之间的紧张关系。在模型训练端,AIGC大规模使用作品,传统版权许可制度失灵;在内容生成端,AIGC的创作自主性提升,其生成内容的“作品”性质和权利归属不明。为规范AIGC技术发展,法国国会提出法案,要求AIGC模型训练一律须取得版权许可,并增设版税制度;将AIGC生成内容的版权归属于原始作品的版权人,并设立标识义务。该法案采取强版权保护和弱技术激励的思路,是法国本土人工智能产业和文化产业实力对比下的无奈之举,对AIGC技术发展有一定的负面影响。当前我国正处于提升人工智能技术国际竞争力的关键阶段,走版权保护和技术发展的平衡道路更具现实可行性。对于AIGC的模型训练,可以采用“法定许可+著作权集体管理”的作品利用方式;对于AIGC的生成内容,在满足“人类最低独创性贡献标准+AIGC生成内容标识义务”后,可将其视为作品进行保护,并将版权归属于AIGC的使用者。
AIGCs such as ChatGPT have triggered the tension between technology and copyright protection. On the model training side, AIGC uses works on a large scale, and the traditional copyright licensing system fails; on the content generation side, AIGC's creative autonomy increases, and the nature of "works" and the attribution of rights of its generated products are unclear. In order to regulate the development of AIGC, the French Parliament has proposed a bill that requires all AIGC model training to obtain copyright licenses and adds a royalty system; attributes the copyright of AIGC-generated works to the copyright holder of the original work and establishes a marking obligation. The bill is a helpless compromise concerning the opposing ways of strong copyright protection and weak technology incentives, which is due to the contrasting gap between the strengths of France's local AI industry and cultural industry. At present, China is in the stage of improving the international competitiveness of AI technology, and it is more realistic and feasible to take the balanced path of copyright protection and technology development. For the model training of AIGC, the utilization of works of "statutory license + collective management of copyright" can be adopted; for the content generated by AIGC, under the premise of meeting the "minimum standard of human originality contribution + the obligation of marking generated by AIGC", it can be protected as a work of art, and the copyright will be protected as a copyrighted work. For the content generated by AIGC, it can be protected as a work of art and the copyright can be attributed to the user of AIGC.