Abstract as per original application
(English/Chinese): |
The patent system is economically integral (Dam, 1994; Khan, 2005; Kitch, 1977). Yet, despite its importance, it faces myriad challenges and is described by some innovation scholars as “broken” (Jaffe & Lerner, 2011) or “failing” (M. D. Frakes & Wasserman, 2019). Many of these challenges arise because of the sheer amount of information that must be analyzed to examine patent applications, and the difficulty that applicants and patent examiners have in dealing with this vast and expanding information universe as they determine whether a claimed invention is patentable. Artificial intelligence (AI) excels at dealing with large sets of data and offers the prospect for efficiency gains in the patent system if implemented appropriately. This project addresses two linked sets of questions: how might this implementation take place and what are the potential benefits and risks that automated administration of the patent system might bring?
In this project’s first stage I will engage with two topics: why policymakers should consider adopting increased use of AI to administer the patent system, and how best to implement it to further the system’s goals. This stage of the project will detail and summarizes the challenges facing the patent system and explore how machine learning and AI might help address those challenges. It will then take a theoretically, and empirically driven perspective on how to go about engineering technologies to improve the patent system’s efficiency.
In the second stage I will empirically demonstrate the capacity of AI to make labour intensive patent examination tasks more efficient. This will entail producing demonstration models allowing users to explore how different methods of automating patentability classifiers affect the classification results. Comparing different approaches, such as a naïve metadata correlational approach, with more sophisticated text-based approaches, will help reveal the tradeoffs inherent in engineering legal decision-making aides.
In the final stage, I will build on the first two by critically examining the effects of automated decision-making aides on the innovation system. The adoption of technologically-aided decision-making technologies is not a new phenomenon, but recent improvements in the abilities of machine learning techniques have enabled technologies that could fundamentally change many elements of legal and administrative decision-making. This final stage will explore these possible futures and discuss what we can learn from the patent system’s adoption of artificial intelligence technologies to aide in patent examination.
專利制度在經濟上是不可或缺的（Dam，1994；Khan，2005；Kitch，1977）。然而，儘管它很重要，但它面臨著無數的挑戰，一些創新學者將其描述為「破碎的」（Jaffe & Lerner，2011）或「失敗的」（M. D. Frakes & Wasserman，2019）。其中許多挑戰的出現是因為審查專利申請時必須分析大量信息，以及申請人和專利審查員在確定所要求保護的發明是否具有專利性時在處理這一龐大且不斷擴大的信息世界時遇到的困難。人工智慧 (AI) 擅長處理大量數據，如果實施得當，有望提高專利制度的效率。該計畫解決了兩組相互關聯的問題：如何實施以及專利制度的自動化管理可能帶來哪些潛在的好處和風險？
在這個計畫的第一階段，我將討論兩個主題：為什麼政策制定者應該考慮更多地使用人工智慧來管理專利制度，以及如何最好地實施它以進一步實現該制度的目標。本計畫的這一階段將詳細介紹和總結專利制度面臨的挑戰，並探討機器學習和人工智慧如何幫助應對這些挑戰。然後，它將從理論和經驗驅動的角度來探討如何利用工程技術來提高專利制度的效率。
在第二階段，我將實證證明人工智慧讓勞力密集專利審查任務更有效率的能力。這將需要產生演示模型，使用戶能夠探索自動化專利性分類器的不同方法如何影響分類結果。將不同的方法（例如簡單的元資料相關方法）與更複雜的基於文字的方法進行比較，將有助於揭示工程法律決策助理固有的權衡。
在最後階段，我將在前兩個階段的基礎上，批判性地研究自動化決策助理對創新系統的影響。採用技術輔助決策技術並不是一個新現象，但最近機器學習技術能力的提高使得技術能夠從根本上改變法律和行政決策的許多要素。最後階段將探索這些可能的未來，並討論我們可以從專利制度採用人工智慧技術來輔助專利審查中學到什麼。
|
