論專利法對人工智慧之保護──歐美實務之觀點

How to Protect Artificial Intelligence under Patent Law?-European and American Perspectives

本文分析歐洲與美國實務界的觀點，探討專利法對於人工智慧的保護。首先，未來歐洲人工智慧專利的申請與審查的核心問題，仍會圍繞著技術貢獻之有無的相關議題。由於「核心人工智慧」實際上即為數學演算法本身，而數學演算法本身乃是歐洲專利公約所界定的純粹數學方法，因此「核心人工智慧」最有可能被認為是欠缺技術特性的人工智慧類型。不難理解的是，近年來在歐洲所有人工智慧科技的專利申請，幾乎均主張其具有技術特性。單憑此種主張，或許可能說服歐洲專利局將「訓練模型或機械學習」和「人工智慧工具」認定為具有技術特性發明，但此種主張，卻無法解決上述「核心人工智慧」事實上即為純粹數學方法的問題。至於未來相關的專利申請工作，是否僅為一場竭力展現技術貢獻的申請競賽，則值得各界密切的觀察。其次，在大西洋另一端的美國，人工智慧專利的申請人倘若未就「核心人工智慧」加以「包裝」，則其專利申請案幾乎確定會遭美國專利商標局核駁。然而，美國專利商標局未來應不會輕易地將AI科技成果認定為美國專利法第101條所界定的欠缺適格性專利客體，而會讓人工智慧科技成果的新穎性、非顯而易知性及其技術揭露的程度，獲得更多被檢視的機會。

This Article analyses the protection of artificial intelligence under patent law from European and American perspectives. First, the key issues regarding examination of AI patent applications concern the requirement of “technical effort”. “Core AI” are mathematical algorisms, which fall within the concept of “mathematical methods as such” in the European Patent Convention. The European Patent Office would presumably hold that “Core AI” is not patentable as it lacks technical effects. Recently, patent applicants in Europe often claim that their AI-related technologies have made “technical contribution”. The European Patent Office might accept this claim, if the technologies are “Trained models/machine learning” or “AI tools”. Nonetheless, it would reject a patent application where the technology concerned is “Core AI”. “Core AI” is considered algorisms, which are not patentable under the European Patent Convention. The future applications in AI-related fields might become “patent drafting games”. Only the skilled patent attorneys demonstrating clever drafting maneuvers could win the games. Second, the United States Patent and Trademark Office would probably reject a patent application if the technology concerned is “Core AI”. In the examination of an AI patent application under Section 101 of the US Patent Act, it seems that the Patent and Trademark Office would not be overly strict. The Office would examine for patentability looking at whether the AI-related technology is new, non-obvious, and well-disclosed.