論機器學習之著作權困境與應對

A Research on the Copyright Difficulties and Response of Machine Learning

人工智慧技術之關鍵，係以資料訓練演算法為特徵的機器學習，需要蒐集、處理並輸入巨量資料供演算法訓練，因而一開始即可能涉及大量重製、改作或編輯他人之著作。人工智慧對著作資料之利用行為，並不必然可以援引現行著作權法之限制規定而免責，使得企業為開展人工智慧研發與應用，需要對大量著作申請授權並支付費用，這不僅造成實踐之困難，亦嚴重阻礙人工智慧領域之科技發展。反觀美國、日本與歐盟等國家或地區已經提供合法化解決方案之背景下，為消除人工智慧發展之著作權障礙並提供國際競爭優勢，有必要制定合法化利用之特別規定。相比於透過著作權限縮、合理使用或強制授權制度，法定授權更有利於降低交易成本，衡平各方之利益，保障著作權人利益的同時，為人工智慧科技之應用與發展提供良好法制環境。故而，本文認為應就機器學習制定法定授權制度，並客觀界定具體適用情形，完善相應配套措施。

The development and application of artificial intelligence is a hot task in the high speed of the era of science and technology. Machine learning is the key to artificial intelligence, which is characterized by data training algorithms. It requires to collect, process and input large amounts of data for algorithm training. Therefore, extensive copying, reworking or editing of other people’s work may inevitably be involved in the very beginning. However, the use of works by artificial intelligence would not necessarily exempt the limitation of the current copyright law, which makes enterprises need to apply for authorization and pay fees for a large number of works. It not only causes practical difficulties, but also seriously hinders the development of science and technology in the field of artificial intelligence. According to the present situation that the United States, Japan, the European Union and other countries or regions in the world have already provided legalized solutions for the problems above, it is necessary for Taiwan to formulate the special rules for the utilization of legalization in order to eliminate the copyright obstacles to the development of artificial intelligence and provide international competitive advantages. Consequently, this paper analyzed the business model as well as proposing four kinds of coping styles such as the restriction of the copyright, the fair use, compulsory license system, and statute license system. In result, compared with the restriction of the copyright, fair use or compulsory license system, the statute license system is more conducive to reducing transaction costs, balancing the interests of all parties, protecting the interests of copyright owners, and providing a good environment for the application and development of artificial intelligence technology. Therefore, this paper argues that the statute license system should be applied to machine learning, and the specific applicable situation should be scientifically defined to improve the corresponding supporting measures.