• Guy Regev

IP Protection in Artificial Intelligence

Artificial Intelligence (AI)—a set of processing and algorithmic technologies that enable machine-based intelligence to train and learn to simulate or augment certain aspects of human thinking and decision making—is rapidly establishing itself as the technological frontier of the future. While not a new concept or field, the recent explosions in data and cloud computing processing power, alongside the development of evermore sophisticated algorithms in the core disciplines of Neural Networks, Machine Learning, Probabilistic Reasoning, Computer Vision, and others, have finally given AI the fertile ground it needed to grow and truly take off.

While we are still at the early stages of the AI revolution, it is already disrupting various commercial sectors, and it is not an exaggeration to say that it will very likely end up transforming nearly every single aspect of our lives and our daily experiences.

As this futuristic potential becomes clearer, AI growth continues to accelerate, and related IP protection and patent filing activity is accelerating right along with it. Nearly 70% of all AI-related patents filed to date worldwide have been filed over the past 5 years. Though there are significant AI patent contributions from academic institutions (especially in the areas of distributed AI, advanced Machine Learning techniques, and robotics), the AI patenting activity is driven mainly by some of technology’s biggest heavyweight companies. IBM, Microsoft, Google, Samsung, and Tencent are the top AI IP property owners, holding approximately 10% of all issued patents. And, also much like other disrupting technologies (like AR/VR), the vast majority of worldwide patent filing activity is being dominated by the U.S and China, with the latter currently leading in overall patents issued.

It is indeed evident that Artificial Intelligence is very quickly establishing itself as a key technological field, seeing rapid growth in innovation, development, commercial applications, and IP protecting and patenting activity. It is also clear that it has immense disruptive potential in almost every sector and industry imaginable. Therefore, as happened with other transformative industries, we can confidently expect AI innovation and patenting activity to continue accelerating, followed by heavy and broad litigation battles, right? Well, for the AI industry, the answer is not so simple.

Active IP litigation involving AI is relatively low, with less than 1% of overall patents being litigated. Some of that may be attributed to the industry as a whole still being in its nascent stage. But there is more to it. It is indisputable that AI presents a unique and particular challenge to the current patent system. For one, under Alice (the current law governing subject matter patent eligibility), “abstract ideas” are not patentable, and there is a clear trend in case law showing this to be applied much more severely to software-centric ideas. This is quite an obvious problem for AI developers, as AI innovations are implemented almost entirely in software and firmware. While recent, new guidance issued by USPTO appears to promote a potentially friendlier environment for AI-related applications, there remain significant risks in seeking patent protection (In recent years, patent filing data indicates about 90% of AI-related patent applications were rejected(!)). Moreover, some AI related innovations are simply not suitable for patenting. For instance, the raw data collected while developing certain AI technologies or applications (such as autonomous driving) is not patentable by itself, nor is it patentable when combined with a conventional and well-known algorithm, regardless of how valuable the end results may be to the company developing it.

Secondly, AI implementations are often inherently hidden, comprising processes that reside and take place on the cloud, with no external visibility. Such technological implementations make patent protections much less effective as a patent owner might not ever be able to discern whether a competitor is infringing on their patented, disclosed technique.

Finally, there is the question of machine-driven infringement. While this may not be an immediate concern, it is important to note that advanced AI implementations ultimately allow computers to perform some function without being specifically programmed to do so. This means that a situation could arise in which infringement occurs unintentionally and without direct human intent. The current patent law system has no clear way of dealing with a non-human infringer or with deciding liability in cases where an unforeseen infringement occurs unintentionally a third-party user of an AI application.

Thus, for AI innovators trying to identify the best way to protect their IP, the risk/reward ratio currently offered by traditional patent protection does not seem like an attractive course of action. Understanding this, it is not surprising to see more and more AI companies gravitating towards the use of Trade Secrets.

A Trade Secret designates information that provides a competitive edge and is of significant economic value to a company, precisely because it is unknown to others. While trade secrets have been less prevalent in IP litigation discussions and analysis relative to traditional IP protections, they have always been important for protecting information essential to the competitiveness of any business – Just Coca Cola, with its secret formula, probably being the world’s most famous, and well-protected, trade secret.

There are several advantages to using trade secrets. For one, they eliminate the risk of enabling competitors and infringement due to the disclosure of the invention. They also offer more flexibility in some respects compared to traditional IP protection: there is no application or registration required, it can be utilized on information of any nature (datasets or results, as well as specific algorithms or implementations), and it can theoretically last for as long as the information is kept secret—again, just ask Coca Cola about their 135-year-old secret.

More specific to decision makers, trade secrets offer an appealing alternative to traditional IP property protection in circumstances where patentability for the information is in doubt, infringement of the innovation may be undetectable, and when overall sector innovation is so rapid that it outpaces the patent protection system itself—All core characteristics of the current AI landscape. Furthermore, with the Defend Trade Secrets Act of 2016, and a smoldering geo-political tech war centered around innovation leadership and IP protections, IP owners have federal-level options when seeking to remedy trade secret misappropriation. The convergence of all these factors and considerations makes trade secrets a most attractive option for AI innovators to pursue.

However, trade secrets do have inherent disadvantages, the most glaring of which is that it is only in effect while the information being protected is indeed a secret. Once disclosed – intentionally or not – the underlying data or innovation can no longer be considered a trade secret or protected as such. Therefore, it is prudent for companies and innovators seeking to protect their IP to try and strike a risk/reward balance and use a complementary mix of patents and trade secrets. For example, a company might turn to trade secrets to protect their data and their evolving algorithms while complementing this with patent protection for core inventions and application-level specific innovations (such as in voice, image, and gesture recognition). Some companies may even opt for a more comprehensive mix of strategies, joining specific patent pools, or participating in selective open sourcing for specific innovation areas.

As the whole industry is still rapidly evolving, so too will the underlying conditions, and with them the desirable IP protection strategy. Suppose future technologies aid in the discovery and detection of infringing AI implementations, and the patent legal framework itself is updated for clearer considerations on software-centric innovations. In that case, companies might shift back towards traditional IP protection and away from trade secrets. It is a near certainty the conditions and considerations will change and evolve—but it is foolish to try and predict how and when.

What we can do —and what we are doing in our firm, GRT—is to prepare according to the highest likelihood outcomes. AI will dominate the future of technology. It will undoubtedly be fought extremely hard for in the professional, consumer, and geopolitical arenas. And it will certainly be a vast litigation battleground. Technical experts in the building blocks of AI will be in heavy demand for both R&D and IP property protection and litigation efforts, and our cadre of experts is tailored accordingly.

Furthermore, litigators in this space must also acquire knowledge and experience that spans beyond patents and traditional IP protection. Trade secrets will have a big part to play in the court battles to come, and knowing their advantages, disadvantages, and consequences for social and geopolitical issues will be paramount.

Again, preparation is the key, as the stakes could not be higher. For, as Vladimir Putin said back in 2017, whoever becomes the leader in AI will end up being “the ruler of the world.”

45 views0 comments

Recent Posts

See All