FROM SEARCH ENGINES TO SOCIAL MEDIA FEEDS TO SMART ASSISTANTS, artificial intelligence (AI) is already integral to modern life.

Building upon the foundations of the internet and the digital age, and rapid advances in technology, AI is propelling us into the next digital frontier.

AI describes any computer-aided technology that simulates human intelligence, and broadly covers machine learning, deep learning, and neural networks. Its application is extensive and spans both personal and professional contexts.

In gauging the economic potential for AI between now and 2030, PwC found that global GDP could be 14% higher in 2030 than it was in 2017 as a result of AI—the equivalent of $15.7 trillion—making it the “biggest commercial opportunity in today’s fast changing economy”.

And the McKinsey Global Institute estimated that around 70% of companies will be adopting some form of AI by 2030.

AI is an increasingly important asset, then, for businesses, organisations and inventors, but this presents both an opportunity and a dilemma in terms of how to protect it.

Before looking at the intellectual property implications of AI, though, it’s important to consider definitions. If a trade secret is information that a business wants to protect because it has particular value to and is unique to that business, what would an AI trade secret look like?

SECRET SAUCE: WHAT IS AI?

James Tumbridge, partner and barrister at Venner Shipley in London, stresses the importance of defining “true AI” as opposed to software.

“Quite often people call me and say they’ve got AI. And when I look at it, it’s not AI. When it is true AI and it’s actually learning and developing to perform a function, those who created it and own it might [then] wish to claim that what it’s learning to do is confidential, possibly a trade secret.”

But he points out that AI trade secrets differ from more straightforward and easily contained trade secrets such as a recipe, for example.

“The much more problematic trade secrets come up in software and AI,” he explains. “The development of products, particularly in the software space, is iterative. You don’t know which step becomes the most important until after the event.”

True trade secrets, particularly with AI, come into play with algorithms—or the “secret sauce”, according to Robert Peake, also a partner at Venner Shipley.

“Algorithms are based on source code, and the source code underpinning those algorithms is typically going to be treated by the owner or by the developers as being a trade secret,” he says.

UNIQUE CHALLENGES

Given the dynamic nature of technology, there are unique challenges in protecting AI trade secrets. This starts from within, in the realm of employee access.

“It’s important to remember that the risk to trade secrets comes not just from interaction with third parties, but also from within,” explains Peake. “Businesses need to be aware that not everyone across the business needs to have access or should have access to all of the information you have.”

And, since the tech sector is known for its high turnover of staff, there is the issue of employee mobility.

“When we’re talking about the coding of computer software, there can be a revolving door of individuals contributing to software, especially in the early days,” adds Peake. “That’s how startups tend to function.

“[But] that’s a lot of people who are coming in and getting a glimpse of what is going to become the value proposition for your business. So businesses need to be very careful about that.”

Sigrid McCawley, managing partner at Boies Schiller Flexner, provides training for her clients in trade secrets protection.

She says it’s not easy for clients to protect their trade secrets, particularly in AI—not least due to the nature of the technology and employee access issues.

“I tell all of my clients with trade secrets that protection is not necessarily easy from a corporate perspective, because you have to have very vigilant restrictions.

“In a number of areas, there has to be, for example, restrictions in access. That means the individuals who have access to the trade secret need to be restricted. And with AI that can be challenging.”

This, she says, is due to the fact that AI usually relates to software and is integrated with other systems.

“So in trade secrets, you can have multiple parts of a trade secret that you’re protecting,” she explains. “Certain of those parts can be public, certain of those are secret, and then the compilation of those becomes a trade secret.

“Often, companies are dealing with multiple pieces that make up their ultimate trade secret, each of which could be defended on its own.”

External considerations include licensing, especially with open-source software in developing AI being commonplace. And reverse engineering, a frequent practice within software development, could become easier to undertake as AI technology advances.

LEGISLATIVE LIMITATIONS

Aside from the steps you can take to defend your AI trade secrets, how can the law offer protection? From a cross-border perspective, there are as yet no globally harmonised laws on protecting AI, trade secrets, or both.

Matt Hervey, partner and head of artificial intelligence law at Gowling WLG in London, highlights that if your trade secret is misused in certain countries, you may lack adequate remedy.

“That’s also true of patents unless you’ve patented everywhere in the world where the court will give you a remedy,” he adds.

He points out that if someone misuses that patent in a country where you can’t enforce it, you can still enforce it in a country which does have patent rights.

“But a trade secret, once it’s no longer secret, you can’t put the genie back in the bottle, even if you have the legal protection, which you did have in another country.”

Peake advises that the first step to protection is keeping that secret a secret.

“There’s no getting away from the fact that actually the first line of defence is to keep it safe. And after that, [you need] to try to pull it back through enforcement.”

Julien Lacheré, a partner at BCF, says a little more clarity on the patentability of AI would certainly help the community to better draw the line between what is eligible and what is not. “I think that would bring more certainty to all stakeholders,” he adds.

Other than that, he adds, the established legal frameworks of copyright, trade secret protection and patent laws combine to provide powerful tools to properly control and protect AI technologies.

“But I am not sure that the big priority on AI is around increasing the level of protection,” he says. “I think AI brings a lot more questions and challenges, such as explainability and regulation of the use of AI— for example, in the automotive industry with autonomous vehicles.”

He adds that the medical field is “still lacking a proper legal framework to fully deploy the technology in a trustful manner to the public”.

Ilya Kalnish, also a partner at BCF, agrees. “I’m not sure if new legislation is necessarily the right answer,” he muses.

“You know, AI is a very fast-developing field, [yet] a lot of laws, especially the patent laws, were put in place well before anyone could envision computers doing what they’re doing today.

“The buckets are very well defined. What we need is clarity, and adapting these frameworks to the realities of AI today.”

NO REGULATORY INTEREST

In the EU, the Directive on the Protection of Trade Secrets 2016 aimed to harmonise member states’ diverging national laws on trade secrets.

And within the UK, in addition to common law protection, there is the Trade Secret Regulations 2018, which implemented the EU’s Directive into UK law and remains in place following Brexit.

But Tumbridge points out that there’s no particular law in the UK and Europe that deals with trade secrets in the way there are for trademarks or patents.

More pointedly, though, there’s a marked distinction between regulating AI and regulating trade secrets.

“The great concern of regulating AI has got nothing to do with trade secrets,” he says. “There’s no regulatory interest in trade secrets, because it’s not what the regulators want to talk about.”

He adds that he “doesn’t expect to see any laws or regulation tackling any of that”. “The common law general structures are already in place in the same way as any other confidential information is protected by contract, or by seeking injunctions.”

PROPOSED EU REFORM

Meanwhile, the EU is proposing sweeping reform of the regulatory framework for AI across all sectors in Europe in the shape of The AI Act, which would be the first law in the world on AI by a major regulator.

“Its aim is to ensure that artificial intelligence (AI) systems placed on the EU market and used in the Union are safe and respect existing law on fundamental rights and Union values.”

Peake is closely following developments regarding the proposed legislation, which he says is “enormous in scope”.

But he believes that more and more complex questions will arise around the ability to protect AI’s ‘secret sauce’—or trade secrets—when it comes to regulating the technology.

He adds that some of the industry feedback on the draft legislation centred on concern over the transparency and “explainability” requirements.

This feedback, explains Peake, was directed at two related regulatory proposals that would require high-risk AI to be developed in a way that would enable users to interpret a system’s output and to use it “appropriately”, and would support “effective human oversight” of the system.

The level of transparency required would allow the “human in the loop” to “fully understand the capabilities and limitations of the system”. That high threshold proposal sparked “considerable debate”.

“From a technical standpoint, we had industry saying, ‘there are instances where that is just not feasible’,” he explains.

LIFTING THE HOOD

Peake adds that industry is going to be reluctant to “lift the hood” as to how an algorithm arrived at a particular decision, because how it got there is closely tied to the value of the business and the value of the product. “They’re not going to want to share that with regulators and potentially with the public, depending on how that process unfolds.”

The other comment from industry, he explains, was that on a technical level, lifting the hood also “may not actually do very much” in that the regulator may not be in a position to really appreciate what they’re seeing.

“The only people who might be able to appreciate that are the competition—those very highly skilled people who might see how something is functioning [and see how] their competitor’s application seems to have taken off more in the marketplace.

“Of course, we’re going to see some big players creating applications which will be general purpose AI that will be used as a sort of toolkit. And it won’t necessarily be obvious to those creating the system, and then promoting and selling the system, how their end users are going to deploy it.

“Business will be creative; they will use these AI components in their own business applications. So there’s also uncertainty there. We don’t know what regulatory issues might come up.

“But you are going to end up, I think, hitting this road bump over and over again, about confidentiality around the way in which a proprietary system functions.”

LEGISLATION: A US PERSPECTIVE

In the US, while the Defend Trade Secrets Act 2016 helps companies to protect their trade secrets on a federal level, different states have varying laws on AI.

McCawley explains that protecting AI trade secrets is a very difficult space for her clients in the US to navigate, with legislation forever playing catch-up.

“The technology is ever-changing, and the law is always a little bit behind,” she says. “Our legislators are trying to respond with particular AI-specific restrictive laws, which make it complicated for companies because they may be creating a product that then is somehow going to be regulated in a way they didn’t anticipate.”

And in terms of litigation, currently, US law requires proof that a company has taken all reasonable measures to protect their trade secret if that secret has been misappropriated.

“Protecting trade secrets is the most important thing you can do,” explains McCawley. “Because if you have to deal with a misappropriation, you need to be able to prove to a court of law that you have taken all of those measures in order to receive any kind of damages.”

TRADE SECRETS V PATENTS

In protecting AI trade secrets, judging whether to use patent or trade secret protection is a key consideration.

The latest research from the World Intellectual Property Organization (WIPO) identified a surge in applications for patents relating to AI technology, with more than half of all applications since 2013 being for AI-related patents.

In reality, trade secret protection forms part of a matrix of IP protection, most often alongside patents.

In fact, Tumbridge explains that Venner Shipley protects a lot of AI via patents, rather than trade secrets.

“The truth is the patent itself gives you a very good steer, but rarely tells you exactly how to [apply] it,” he explains. “A coder would have to experiment, and that’s where know-how comes in, which we protect by contract.”

In terms of determining whether AI is better suited to patent or trade secret protection, it depends upon the business case, explains BCF’s Kalnish.

“A very simplistic example is, if the engineer [of the AI] is going to be publishing a white paper, and speaking at a conference about the technology, the company has no choice but to file a patent application because they cannot keep it a secret,” he explains. “So we literally have to dissect the business case behind each of the technologies.”

Lacheré says that they typically start exploring IP options in a ‘mining session’ with the client.

“So before we even make a decision as to whether or not we go for patents or trade secret protection, we usually need to get up to speed on the technology and understand, among the zillion lines of source code, what is the invention and the contribution of the development team and of the clients.

“Because when you work on AI projects, it’s usually many stacks of technologies that are merged. And you need to have a clear understanding of what was already known, and what the team reused from existing products, [be they] proprietary or open source technology.”

Kalnish adds that in most cases, though, companies have a choice over which type of IP protection will suit their AI.

“For example, some technology is easily reverse-engineerable. And if it is, keeping it as a trade secret becomes a virtually impossible task, from a practical perspective.”