Patenting intellectual property can be complex
BE PATIENT, BE PRUDENT AND BE PROACTIVE
Benjamin Alarie, co- founder and CEO of Blue J Legal, says the patent for its AI software for legal and accounting firms is filed and pending. But in an AI world, he’s the first to admit it is far from a simple process.
“Patents take three to five years to process. At the same time, technology is moving very, very quickly. If you wait too long to patent something that’s truly innovative, it’s likely someone else will file.”
Securing intellectual property ( IP) can be a major hurdle for startups at the best of times. But software — and in particular AI — brings its own unique challenges.
“There are specific complications compared to mechanical patents that are more straightforward,” Alarie says. “For example, there are a lot of nuances there in terms of what is patentable. If you have a relatively small startup, it may be advantageous to have some IP protection in order to defend your work, but it can be very expensive. The key is, how much do you want to bite off to protect yourself ?”
Organizations such as TD Bank Group are recognizing the challenges facing the startup community on the IP front. It has recently announced a $ 30 million investment pool that will provide patent funding and expertise to new fintech ventures, so they can focus on rapid growth and innovation. “The AI domain is exploding and there is lots of potential out there,” says Tim Hogarth, vice-president, Innovation Framework & Strategies. “This program is intended to help them move faster and get done more quickly.”
Speed is only one part of the equation, however. Guy Levi, principal with Levin Consulting Group in Wyckoff, N. J., notes that the first question on the part of investors, or anyone interested in a merger and acquisition of a startup or its technology, is whether they wholly and solely own its IP. Yet the answer is not that straightforward with AI.
“Why AI is so challenging is that patent law progresses linearly, while advances in AI move more quickly creating an accelerating return situation,” he explains. “In other words, the gap between issues created by advances in AI technology and the ability of the law to address them is getting wider. Three to five years is pretty much an eternity in the AI world and any time in that process, you could be stepping into someone else’s IP.”
One major issue is that AI startups typically use one or multiple open source software resources to build their solutions. But not all licence agreements are the same, Levi notes. “Some don’t allow you to protect your IP; others exclude you from excluding others using the source codes; while still others allow you to get a patent, but restrict you from enforcing any claim. The problem is, most companies do not know what open source they have.”
A second and intriguing challenge is that machine learning by its nature constantly writes its own code to improve itself. So who is the owner or inventor and how do you know if that code infringes on someone else’s IP?
The third conundrum is, what exactly do you protect, whether through copyright, patent or classifying it as a trade secret? Is it the source code, data set, test data sets, or other component of the process?
“These are issues we are always talking about,” Levi says.
While it may seem insurmountable, there are some steps a startup can take when considering their IP options.
Be very clear who owns the resulting intellectual property before embarking on anything. “Sometimes people get so excited about discovering things and moving the field ahead, they can get caught unawares further down the line,” Alarie says. “If you’re part of a hackathon team for example, have a common agreement amongst the team members before you start.”
Then take the time to understand the licence agreement for the open source software you are using.
Since patent rights are based on first to file, getting an early filing date is also essential, Hogarth says. “That claim date is applicable worldwide. Hence it is critical for startups to file as early as possible, which is often when money is the tightest.”
Things could get even more complicated as t he i ndustry evolves from narrow, single task AI applications such as a cancer diagnosis or parking, to general AI that will perform more than one task (e.g. improving the quality of life on Mars), Levi says.
“When we get to general AI, all bets are off.”