Home > Article > Technology peripherals > USPTO: Adopting a human-centered approach to AI innovation
Like many insight-driven organizations, the United States Patent and Trademark Office (USPTO) uses technologies such as data analytics, artificial intelligence, and machine learning to improve its operational efficiency and performance, as well as improve its systems and process quality.
Artificial intelligence and machine learning algorithms are critical to the work of the USPTO, but at the same time, the government agency’s guiding principles are in developing and using these technologies to improve and expand programs. time, adopting a people-centered approach. Jamie Holcombe, chief information officer of the U.S. Patent and Trademark Office, noted that artificial intelligence and machine learning tools help augment the capabilities of human experts and enhance their ingenuity in their work. At this point, such tools are slightly different from the human mind. Or the reasoning ability is unmatched.
Jamie Holcombe, Chief Information Officer of the United States Patent and Trademark Office
In order to further supplement and improve this technology, the United States Patent and Trademark Office uses passive and active capture methods to Input from thousands of experienced employees trains and improves AI-driven models to ensure the technology achieves the desired results. The U.S. Patent and Trademark Office has granted more than 11 million patents since its inception and has more than 12,000 employees, including engineers, attorneys, analysts and computer experts. Not only that, ongoing feedback from frontline patent examiners is used to improve AI/ML models to drive new product development and support activity in two key areas: patent search and classification.
Holcombe noted that conducting a comprehensive patent search can be challenging given the current explosion of data volumes and potential sources of “prior art.” To address this challenge, the USPTO’s technology team is using artificial intelligence in a new patent search tool to help examiners find the most relevant sources they need when reviewing applications. This is important because the USPTO receives more than 600,000 applications each year, and the average application contains approximately 20 pages of text and images, or approximately 10,000 descriptive words. The USPTO IT department also developed and deployed a classification tool that can identify and match classification symbols associated with the invention from more than 250,000 possible classes.
In both cases, the model is developed and continuously enhanced by feedback provided by human experts who use humans to judge whether something is new or novel, and then apply Legal, factual and professional knowledge to make decisions.
Getting a steady stream of feedback from examiner experts and others can be an advantage, but it is not what the USPTO wants to determine innovation and new channels of global expertise, the only way to help solve important challenges and scale artificial intelligence. Earlier this year, the USPTO turned to the AI research community and Google Kaggle. Kaggle is a technical and social platform for data scientists and others to exchange thoughts and ideas. Kaggle holds a global coding competition every March, offering a prize of US$25,000, and calling on artificial intelligence researchers and data scientists to write code. to evaluate the semantic similarity of phrases.
This year the competition received 42,900 entries before closing on June 30, involving more than 1,800 global teams working together to leverage publicly available proprietary data sources. Holcombe explained that the goal of the competition is to promote the use of AI to help agencies and the patent community better understand patent language. "The result is not only better phrase algorithms for patent searches, but also having winning models adopted into the public domain," he said.
The USPTO also leverages other public information resources, such as Golden, a free “Wiki-style” artificial intelligence/machine learning-driven platform launched in 2019, can match topics with relevant and available data through web searches and integrate them into an information flow, running behind After the AI algorithm is launched, relevant data can be continuously added, and anyone can search for information about the company, its patents and funding sources (such as venture capital).
Although we see a lot of technical columns about technology integration, in view of the diversity and complexity of human nature , taking a “human-centered” approach to developing artificial intelligence and machine learning can be challenging. Therefore, the USPTO, under the leadership of Holcombe, developed guidelines from pilot to prototype to production, summarized as A, B, and C guidelines:
The above is the detailed content of USPTO: Adopting a human-centered approach to AI innovation. For more information, please follow other related articles on the PHP Chinese website!