This article is more than 1 year old

AI-friendly patent law needed 'as a matter of national security', ex-USPTO boss says

Neural network training and whatnot didn't occur to 18th century founders

America urgently needs to rewrite its patent laws to recognize modern artificial intelligence technologies, business and IP leaders have said.

This sentiment emerged from a series of hearings organized by the US Chamber of Commerce, during which experts from academia, industry, and government were invited to speak. The meetings, held last month, raised important questions plaguing the development of state-of-the-art AI models: should AI algorithms be patentable? And, separately, should these systems be granted patent rights for inventions they help create?

Today's IP laws are outdated, it was argued. The rules dictating what types of innovations can be patented have stayed largely untouched since the historic Patent Act of 1793. Although the law is broad and states "any new and useful art, machine, manufacture or composition of matter, or any new and useful improvement on any art, machine, manufacture or composition of matter" is potentially patentable, there other conditions that make it difficult to patent things like machine-learning models. 

Patents are only useful if they provide clear scientific and economic benefits to the country, the group argues. It's why the Patent Act states that descriptions of the inventions should "enable any person skilled in the art or science, of which it is a branch, or with which it is most nearly connected, to make, compound, and use the same." That means someone suitably skilled should be able to take a patent text and diagrams, understand what's going on, and reproduce the technology themselves.

But take a system with a trained neural network. That collection of weights and values that mysteriously turns input data into output predictions is opaque and hard to interpret: experts often don't quite know why a model behaves the way it does, which makes explaining its inner workings in a patent difficult.

Well, OK, let's just say the patent explains how to train the neural network to produce the same results, thus allowing the invention to be recreated. But reproducibility is notoriously difficult in machine learning. You need access to the training data and other settings to recreate it. That becomes problematic if the data is medical or personal info, or proprietary, because it would need to be made public as part of the patent filing, and not all the necessary settings and tweaks may be disclosed in an application.

Patent examiners, therefore, may struggle with patent applications of AI technology, and reject submissions, if they find the text is confusing, or not interpretable or reproducible. Thus, changes are needed in the law to allow machine-learning systems to be accepted as novel inventions, it was argued. And being able to patent and protect these inventions encourages businesses to build commercial products, we're further told. Everyone gets to see the progression of tech and science, and inventors are granted rights to their specific part of it.

It is absolutely crucial, and it is a matter of immediate national security

"The patent code that [our founders] put in place was fantastic, however they did not anticipate DNA processing, artificial intelligence, cryptography, software code, and all of the modern technologies of the next industrial revolution," Andrei Iancu, former Under Secretary of Commerce for Intellectual Property and ex-Director of the United States Patent and Trademark Office (USPTO), said in a Chamber of Commerce statement on Monday.

Rejecting AI patents, however, we're told, will keep knowledge of the latest commercial applications of the technology from the public and hamper innovation.

"So, to say that the patent system, at least from that perspective, needs to modernize is an understatement. It is absolutely crucial, and it is a matter of immediate national security," Iancu added.

The chamber noted China has surpassed the US in the number of international patent filings in 2019 and in 2020. If America is to hold a leadership position in AI, its leaders need to treat IP, such as machine learning breakthroughs, as a national asset, Brian Drake, federal chief technology officer at Accrete AI Government, a company focused on building enterprise-level AI applications, asserted. 

Because for one thing, he said, rival nations are pouring all their energies into developing machine-learning technology to use against the United States of America.

"I'm talking about all the instruments of national power from our adversaries being directed at all of our national security instruments and economic power centers. That means their intelligence apparatuses, that means their direct and indirect funding apparatuses, that means their commercial military integration activities. All of those are being directed toward artificial intelligence. And make no mistake, it is about winning the future war," Drake said.

Most experts agree AI algorithms should be patentable, but whether patent authorship or ownership rights should be given to machines that produce technologies, however, is debatable. Current IP laws do not recognize non-human entities as inventors, meaning machine-learning systems cannot be recognized as such.

Stephen Thaler, founder of Imagination Engines, a company in Missouri, who applied in 2019 for two US patents which listed his machine named DABUS as the inventor, found this out the hard way when his applications were rejected by the US Patent and Trademark Office.

Thaler believes there is good reason to give machines at least authorship rights, as it would discourage humans from stealing computers' ideas and profiting from them – the originator would be on record in the patent office – he previously told The Register. But it's not clear that there is any practical use in recognizing software as inventors yet, considering they have no agency or capabilities to sue for infringement unlike humans. 

"To summarize, we cannot sustain innovation around AI without robust and reliable IP rights, which are essential to the prosperity of our innovative nation," Christian Hannon, a patent attorney serving in the Office of Policy and International Affairs at USPTO, said. "To grow our economy and stay globally competitive, we must promote invention and patenting more than ever."

The US Chamber of Commerce, one of the largest largest lobbying organizations in America, is planning to publish later this year a final report from its hearings, issuing recommendations for policy changes the US government can enact. ®

More about


Send us news

Other stories you might like