Is the cost of AI a barrier to IP offices?
Artificial intelligence (AI) can significantly improve the day-to-day operations of an IP office, but policymakers need to invest now to secure its full potential, said leading data experts this week.
Speaking at a panel hosted by WIPR Patents Live, Alexander Klenner-Bajaja, head of data science at the European Patent Office (EPO), said better transparency was key to the success of AI: “There is a lot of mistrust, you always hear this ‘black box’ thing about AI. I argue we need to make the human capable of following the reasoning of the machine.”
“You don’t need to be a fully trained engineer to understand what’s going on. [In the case of patent classification], you can easily calculate which terms contributed most to a machine’s decision and you can visualise it,” Klenner-Bajaja said.
“I would like to make all of that very transparent. Then we can say that the AI has been trained by a human expert and we can show exactly how it came to its decision.”
The panel, which featured contributors from three IP offices, discussed the variance in how AI is applied in different countries depending on needs and resources. Jens Petter Sollie, project manager at the Norwegian Industrial Property Office, said AI had helped improve the quality of trademark searches “a lot, but the cost is much higher than the savings”.
“We have gotten it very cheap so far but we don’t know if that’s going to continue. If we want to buy it in the market, it’s too expensive for us. This is one of the immediate threats for the small IP offices,” Sollie said.
While NIPO has worked with a supplier to incorporate AI products into their daily operations, bigger offices are developing their own in-house systems.
Jamie Holcombe, chief information officer at the US Patent and Trademark Office, said he hoped AI could bring greater efficiency to the federal government.
“In the government, everyone talks about better, but not cheaper or faster. We’re not tax-funded, we’re fee-funded. So we have an obligation to ensure applicants get the biggest bang for their buck,” Holcombe said.
AI tools could help “save a lot of time” in the fields of classification and search while maintaining or even improving quality, he added.
In Klenner-Bajaja’s view, AI is necessary not just to cut costs but simply to deal with the mounting quantities of data IP offices have to deal with.
“Prior art is ever-growing and growing faster than ever before. To offer a complete search, we need to use AI and we need to use it today, not tomorrow. That will allow us to handle these vast amounts of data,” he said.
But AI is only as good as the datasets it’s trained on, and there’s a base level of investment required in order to secure results. Holcombe stressed the importance of not “throwing good money after bad” in the development of expensive AI tools.
“If you fail, make sure you don’t repeat the same mistakes. There’s a certain amount of table stakes that you need in order to invest in a good AI programme. One of them is the golden dataset—if you start the training on something that’s less than ideal, and you’re not cleaning that up in the beginning, your neural networks are trained on something that’s less than optimal. That means you’ll get bad results in the end.”
WIPR Patents Live will include weekly broadcasts from some of the best speakers in the technology sector in the form of virtual panel discussions, roundtables, webinars and presentations.
Did you enjoy reading this story? Sign up to our free daily newsletters and get stories sent like this straight to your inbox
Today’s top stories
Amazon under the spotlight in latest ‘Notorious Markets’ report
