Technology keeps creating challenges for intellectual property law. The infamous case of the "monkey selfie" challenged the notion of not just who owns a piece of intellectual property, but what constitutes a "who" in the first place. Last decade's semi-sentient monkey is giving way to a new "who": artificial intelligence. The rapid rise of AI has forced the legal field to ask difficult questions about whether an AI can hold a patent at all, how existing IP and patent laws can address the unique challenges that AI presents, and what challenges remain. The answers to these questions are not trivial; stakeholders have poured billions upon billions of dollars into researching and developing AI technologies and AI-powered products and services across academia, government, and industry.
"Even if artificial intelligence is only a part of a larger solution, we must arm the gatekeepers of patent rights with better tools so they can better carry out the goals of the patent system." About a month ago, Steve Brachmann authored an article concerned with a brief given to Capitol Hill staff by Professors Frakes and Wasserman. The article highlighted fundamental, as well as practical, problems with Professors Frakes' and Wasserman's proposal (i.e. Frakes and Wasserman cite sources saying that the U.S. Patent and Trademark Office (USPTO) is granting too many bad patents which "unnecessarily drain consumer welfare, stunt productive research, and unreasonably extract rents from innovators." "On average, a U.S. patent examiner spends only eighteen hours reviewing an application, which includes reading the application, searching for prior art, comparing the prior art with the application, writing a rejection, responding to the patent applicant's arguments, and often conducting an interview with the applicant's attorney. If examiners are not given enough time to evaluate applications, they may not be able to reject applications by identifying and articulating justifications with appropriate underlying legal validity."
AI (Artificial Intelligence) is a revolutionary technology, which is being a part of everything we do in our life on a daily basis. If we talk about a few decades ago, there were tasks that could only be done by humans, such as playing chessboard, doing research work etc, but nowadays, AI-based Machines are capable of doing these tasks. And soon, AI will be developed to the level that it will change the way we work, study and communicate. Moreover, if we will talk about patent offices worldwide, every year the number of filings of the patent applications in patent offices is increasing exponentially and the patent examiners have limited time to work on a patent application. So due to these reasons, the quality of patents is getting compromised.
There has been a surge in applications of machine learning over the last few years as companies look for ways to leverage big data in their products and services. That has corresponded with a big increase in another type of machine learning application – i.e. those sent to the United States Patent and Trademark Office for protection. But the ramifications of the machine learning-patent uptick are not yet clear. Statistical and anecdotal evidence suggests we're in the midst of major upswing in patent protection requests for machine learning inventions. While hard numbers can be tough to come by due to intricacies of the USPTO process (and the fact that it will conceal applications upon request), several researchers have identified what they see as a surge in interest in protecting machine learning products over the past several years.
In this work we focus on fine-tuning a pre-trained BERT model and applying it to patent classification. When applied to large datasets of over two millions patents, our approach outperforms the state of the art by an approach using CNN with word embeddings. In addition, we focus on patent claims without other parts in patent documents. Our contributions include: (1) a new state-of-the-art method based on pre-trained BERT model and fine-tuning for patent classification, (2) a large dataset USPTO-3M at the CPC subclass level with SQL statements that can be used by future researchers, (3) showing that patent claims alone are sufficient for classification task, in contrast to conventional wisdom.