人工智能深度学习中的著作权侵权认定
Identification of Copyright Infringement in Deep Learning of Artificial Intelligence

人工智能技术的发展引发了新的法律问题。作为一种全新的著作创作模式，人工智能深度学习所造成的侵权较之于一般著作权侵权更为复杂。人工智能深度学习的产出品具有独创性，传统著作权侵权中所适用的“实质性相似 + 接触”原则，不能充分回应人工智能深度学习的侵权问题。主要原因一是在于人工智能独特的数据获取方式满足“接触”这一要件；二是由于写作型人工智能在深度学习中具有特殊构造，认定“实质性相似”的三种方法均难以用于认定输出作品与源作品之间存在的关系。对此，在当前的技术水平下，可以考虑利用合理抗辩来平衡著作权原权利人与人工智能创造者之间的法益。这就要求：一是明确“合理使用”标准，二是延展“法定许可”标准，三是要对前述两类标准合理选择适用。
The development of artificial intelligence technology has caused new legal problems. As a new mode of copyright creation, the infringement caused by the deep learning of artificial intelligence is more complex than the general copyright infringement. The products of AI deep learning are original. The principle of “substantial similarity + contact” applied in traditional copyright infringement can not fully respond to the infringement of AI deep learning. The first reason is that the unique data acquisition method of artificial intelligence meets the requirement of “contact”; Secondly, due to the special structure of writing AI in deep learning, the three methods of identifying “substantial similarity” are difficult to identify the relationship between output works and source works. In this re-gard, under the current technical level, we can consider using reasonable defense to balance the legal interests between the original copyright owner and the creator of artificial intelligence. This requires: firstly, to clarify the “reasonable use” standard; secondly, to extend the “legal license” standard; and thirdly, to reasonably select and apply the above two types of standards.