Focusing on Value: Viewing AI Through an IP Lens
October 5, 2020
By Denis Keseris
In the last decade, Artificial Intelligence (AI) has exploded into the business world. The increasing use of AI to solve real-world problems has resulted in an increase in commercial competition and a desire by founders (and their investors) to convert the intangible value of their technology into business value. These trends are evidenced by the exponential growth of AI-related patent applications filed throughout the world1. Here at home, with no shortage of innovation2, and recent court decisions3 removing hurdles to the patenting of AI-related inventions, these trends look set to accelerate.
In an area of such fast-paced developments, it is easy to rush into the exercise of acquiring patents, particularly when strongly encouraged by investors, board members and strategic business advisors. Before doing so, however, it is worth taking a few moments to extract ourselves from the complex process of this endeavour, and to refocus our minds squarely on its purpose. It is worth reflecting on what it is exactly we want to protect, and why.
In other words, when considering how best to go about protecting your AI, it is most efficient and effective to start with the actual value your AI will bring to your customers/users, and work backwards. At the end of the day, shrewd competitors will care more about going after your customers than they will about replicating your technology!
Protecting the underlying value proposition of an AI-related invention can be tricky, to say the least, but a key step is understanding which type of AI invention one might be dealing with, from a patent perspective. This is because the focus and scope of any resulting patent application will depend on the type of AI invention involved.
That being said, in assessing any of the types set out below, it is crucially important to resist starting with the details of how the invention does what it does (i.e. your tech’s process) - though this will usually need to be disclosed to a greater or lesser extent in your patent application - and instead try to focus initially on the value proposition provided to customers/users by your invention (i.e. your tech’s purpose). Patents are tools designed to be used in specific commercial contexts, and should therefore be created with those same commercial contexts in mind.
Implementations of AI
Implementations of AI are inventions that have an inventive concept (i.e. the crux of the invention) related to an AI itself. Typically, the value proposition of these inventions is a better AI and, in many cases, the actual product and/or service relates to selling that better AI (or access to it). An example of this is Google’s System and Method for Addressing Overfitting in a Neural Network (US9406017B2), and in particular what value that invention contributes to the Google Cloud AI platform.
The technical features of this type of invention include, for example, new models, pre/post processing of data, and mapping of data, etc. For this type of invention, it is important to start off by defining the invention in terms of the effects that any novel characteristic(s) have on a resulting AI, and then proceeding to add the minimum number of technical constraints required to arrive at those effects.
AI-assisted products use AI to improve their decision-making abilities. AI-assisted products typically result in smarter products, as opposed to better AIs. A smart robotic system for delivering packages within a warehouse is a good example. Such a system could include a number of sensors (i.e. producing inputs) and actuators (i.e. using outputs) that relate to each other through an AI. Unsurprisingly, users of the resulting product (e.g. warehouse owners and operators) are likely to care little about linear regression models and support vector machines, and more about floor-space utilization and labour efficiency. As such, it is important to define the invention in terms of its components (e.g. its sensors and actuators), and specifically how those components interact to increase floor-space utilization and labour efficiency.
In a few cases, these inventions will require specific implementations of AI to function optimally, and there may therefore be some overlap with the category described above. Most AI-assisted products, however, can be implemented using different AIs, and even different types of AIs. It is often therefore better to avoid defining the invention in terms of a specific implementation of AI.
Indeed, if an AI-assisted product is commercialized without needing to explain the inner workings of the AI to customers/users, it may be advantageous to describe the AI as a “black box” in order to avoid unduly narrow patent protection. In such cases, any elements that are considered to be part of the “black box” may not be disclosed in a patent application, and could be well suited for trade secret protection, provided they are well documented and reasonable steps are taken to maintain secrecy.
AI-implemented method inventions relate to the application of AI to a given automated method. With this type of invention, the value provided to customers comes from “smarter” automation or, in some cases, the automation of methods that would not have previously been possible because of their complexity. For example, an AI-implemented method of diagnosing heart disease by analysing echocardiogram data would fit into this category.
From a patent perspective, AI-implemented methods fit somewhere between implementations of AI and AI-assisted products. The degree to which the technical characteristics of a specific AI will need to form part of the definition of an AI-implemented method invention depends on how much of the ultimate value is predicated on those technical characteristics. In terms of the diagnosis system example given above, that would mean defining the invention in terms of the AI’s inputs and outputs, as well as how they might relate to each other to produce specific results. Only the characteristics of the AI that are necessary for producing the specific results should form part of the definition of the invention, with other characteristics possibly being well suited for trade secret protection.
AI-generated products are new products that have arguably been “invented” by AIs. A new bicycle having a shape designed by a shape optimization AI to maximize specific aerodynamic properties would be an example of such a product. While this type of invention is still rare, its impact on the patent system has already been felt. In 2018, a number of patent applications were filed around the world naming “DABUS” as the inventor. DABUS is an artificial intelligence system that purportedly designed a new beverage container and new fractal light signals.
The difficulty with this category is not so much with identifying the technical features that contribute to the value proposition of the invention, but rather identifying the actual inventor, and thus who should own any resulting patent rights. Some patent offices, including the United States Patent and Trademark Office and the European Patent Office, have issued decisions rejecting the DABUS patent applications for failure to name a “natural person” as an inventor. These inventions therefore need to be considered carefully in light of evolving inventorship issues.
In conclusion, while the complexity of AI as a technology can make the process of seeking protection complicated, it should not distract from the purpose of seeking protection. Business people should be aiming to protect (as far as possible) the value that an invention brings to the market, so that the commercial exclusivity afforded by that same protection can be used to enhance business value.
Content shared on Bereskin & Parr’s website is for information purposes only. It should not be taken as legal or professional advice. To obtain such advice, please contact a Bereskin & Parr LLP professional. We will be pleased to help you.