How IP Offices Use Artificial Intelligence

Artificial intelligence (AI) has become a source of intrigue, and its impact on society much debated. While intellectual property offices around the world are wrestling with how to handle AI-related, or even AI-created, intellectual property (IP), many are beginning to use the technology to facilitate their own work.

Just last month, the World Intellectual Property Organization (WIPO) announced the launch of a new image-search technology powered by AI. It will be implemented through 45 trademark offices worldwide (covering almost 38 million marks) as part of WIPO’s Global Brand Database. Unlike previous iterations of AI-powered image search, which mainly rely on analysis of shape or color to determine similarity, the new tool is able to identify combinations of concepts to do the same.

Announcing the new initiative, WIPO Director General Francis Gurry called on IP offices not currently participating to consider joining, underlining that “a larger data pool means better AI results.”

Communication plus collaboration between IP offices is likely to be crucial for AI-based solutions to be implemented effectively, says Christophe Mazenc, Director of the Global Databases Division at WIPO. He says current collaborations with national offices provide WIPO with “diverse data” that is compiled, prepared, and cleaned for building AI tools.

“If these tools bring some benefits, they can be shared with collaborating offices to help improve their operations,” he explains. “We are also envisaging transferring knowledge about building, maintaining, and using AI tools.”

As well as trademark search, WIPO uses AI to enhance other areas of its work, says Mr. Mazenc. The Madrid System has been using machine translation to speed up the work of translators for goods and services texts, reducing translation outsourcing costs. As the AI technology is open source, WIPO can build on and customize it to fit the organization’s needs, Mr. Mazenc says.

Trademark administration procedures involve a number of repetitive tasks. With time, Mr. Mazenc points out, such tasks will become more efficient due to advances in AI technology—with more reliance on the technology than on humans.

However, those who worry that AI will supplant humans need not be too alarmed for now, at least when it comes to IP. “AI has benefits, but it’s not perfect,” Mr. Mazenc suggests.

AI outputs still need to be validated by humans. In addition, people will continue to have an important role to play to ensure quality control. AI tools often suffer from a “black box” effect that leads to difficulties in their quality assessment and in their adoption by end-users, he adds.

Around the World

While WIPO has certainly been a pioneer in using AI for trademark work, many national offices have also begun using, or exploring the use of, the technology.

In January, the Trademark Office of China’s National Institute of Intellectual Property (CNIPA) announced the launch of a new smart device search function. In line with a national reform goal to improve the efficiency and quality of trademark examination, and reflecting a continuing acceleration in the number of trademark applications in the country, the new tool uses AI for image searches.

In a statement, CNIPA said that, this has increased productivity and decreased the number of trademarks that an application needs to be checked against from the tens of thousands to around 5,000. In addition, CNIPA said it is looking to extend its use of AI—and to become a standard setter in the field, enabling the Office to increase its voice in international cooperation..

While China’s solution was developed in partnership between CNIPA and Beijing Huazhi Jingwei Company (China), Shanghai Yitu Company (China), and Tencent (China), IP offices in other countries are looking to off-the-shelf solutions.

The Norwegian Industrial Property Office (NIPO) uses basic AI for its trademark search function, and currently uses “lower degree AI” in its digital case management system, according to Jens Petter Sollie, IPR Systems Manager at NIPO. This lower-degree AI involves pattern recognition to assist with hit sorting for trademark comparison. “It gives value,” Mr. Sollie says, “but just as a complement.”

However, NIPO is currently exploring the next generation of AI-based software, and hopes to upgrade soon. NIPO uses software called Acsepto, produced by Sword Group (United Kingdom); it is also used by several IP offices worldwide (including in Croatia, New Zealand, and South Africa). The latest version of the software employs image search and classification using “neural network AI,” which is trained on previous examiner activity to enable more accurate results, according to Mr. Sollie.

He says that ultimately, the biggest beneficiaries will be users of the system; in time, they will be able to make free searches, which in turn should improve the quality of filings. The aim is to roll out the software to cover trademarks initially, and then adapt it for designs and patents too.

“In time AI will improve the quality of decisions, both by enhancing the capacity of the Office and the quality of search results by officer,” Mr. Sollie says, “but perhaps most importantly, [it will improve] the quality of filings by the customer and their understanding of quality levels.”

In Chile, the Instituto Nacional de Propiedad Industrial (INAPI) has been using AI in trademark examinations for over a year now. The technology allows INAPI to compare image trademarks with “incredible accuracy,” finding all relevant images within its database that contain some kind of similarity, explains Lorena Mansilla Inostroza, INAPI’s Acting Deputy Director of Trademarks, Geographical Indications and Appellations of Origin.

The tool was developed in conjunction with the University of Chile Physics and Mathematical Sciences Faculty, which owns the associated IP rights. Ms. Mansilla Inostroza explains that although INAPI has started rolling out the technology, it is still being tested and refined. Once it is fully developed, “I believe its possibilities will be very broad,” she suggests.

What are the benefits? Among the many advantages, Ms. Mansilla Inostroza says AI allows INAPI to improve processing times and increase the accuracy of searches.

“Specifically regarding image comparison, the great advantage is that it makes it an objective, science-based exercise, diminishing our margin of error and giving us a solid ground for refusal in certain cases. It also makes the process quick and reliable,” she says.

Therefore, Ms. Mansilla Inostroza adds, “Our expectations for the future are very high.” INAPI hopes to use AI in other stages of the trademark examination process, particularly within formal examination of goods and services. “Being able to conduct this phase automatically would be very useful in order to save time and speed up the registration process,” she explains.

Some of these benefits are appreciated by the Japan Patent Office (JPO), which has not yet implemented AI in trademark examinations, but is exploring how it might do so in the future. Satoshi Noguchi, Director General, Trademark and Customer Relations Department, says the Office is expecting AI to relieve trademark examiners’ workloads and improve examination quality and efficiency.

“The number of trademark applications is increasing in recent years, so if we can use AI, we hope it can lead to providing applicants earlier notices about examination results,” he explains.

Use Studies

In a two-part study, the JPO is first examining whether it can use AI for image searches. Mr. Noguchi explains that when searching for prior figurative marks, examiners “must make significant efforts” to ensure that their search queries cover a large number of databases and check all of the search results. Quite simply, this creates a vast amount of work for examiners, he says, so the ultimate goal is to reduce their workloads.

In fiscal year 2017, the JPO verified the search accuracy of image searches through the use of AI and identified issues that need to be resolved. In fiscal year 2018, the JPO continued its work to increase search accuracy and clarify the scope of image searches that are capable of supporting figurative trademark examinations. It also began conducting trials on AI-based image searches by examiners, including the task of assigning search codes to figurative elements of the filed mark, Mr. Noguchi says.

The second phase of the JPO’s study relates to the examination of similarity of goods and/or services. Throughout fiscal years 2017 and 2018, the JPO studied the use of AI to support the assignment of similar group codes, which helps examiners to ascertain the scope of similarity between goods and/or services, to designated goods and/or services, says Mr. Noguchi.

To conduct the study, the JPO developed AI that could learn the lists of goods and services in its Examination Guidelines for Similar Goods and Services, as well as past examination results.. The assignment of similar group codes is one of the most time-consuming tasks for trademark examiners, Mr. Noguchi explains.

“Therefore, the JPO believes that if AI-based systems can be used to do this, it will improve examination quality and speed up the examination process,” he says.

One of the main limitations is that, as things stand, AI is not 100 percent accurate, so there is still a need for human input. Mr. Noguchi gives the example of decisions on trademark registrability, a requirement that is based on consumer recognition. “As of today, it is technologically difficult for AI to make decisions about trademark registrability; instead, we think that AI can become a means of supporting examiners’ work,” he suggests.

While there’s certainly a long way to go before AI can revolutionize the work of IP offices, the speed of advancement suggests it would be foolhardy to predict that it won’t. In the meantime, the technology is already enhancing IP offices’ work, and perhaps more significantly, is proving more beneficial for users of the systems.

That trend is likely to continue as more offices adopt the technology, and specifically as initiatives such as WIPO’s gain traction, since the strength of an AI solution is predicated to a large degree on the quantity and quality of available data. The extent to which IP offices coordinate their efforts in this area is likely to have widespread ramifications for the future of the industry. In any event, the days of trademark examiners manually trawling through thousands of applications are, if not over, then on the way out.