Patenting Artificial Intelligence: Main Challenges and Key Considerations

This blog is part two of a two-part series on patents and AI technology. The first part, Exponential Increases in Artificial Intelligence Patent Filings, discussed the growing number of AI patents filed and granted, including some recent examples.

In this blog, we examine the challenges of patenting artificial intelligence (AI) and the key considerations companies should take into account when writing strong AI patent applications.

The Challenges of Patenting AI

As with most software-related patents, AI applications face a number of significant hurdles. Three common challenges are described below.

Is it a patentable object?

Mathematical algorithms are not patentable in many jurisdictions. The implementation of a mathematical algorithm using a computer does not, in itself, transform the algorithm into patentable subject matter. At one level, machine learning models are simply mathematical algorithms embedded in a computer.

Is it new and not obvious?

Machine learning models – and their use or application – should be new and/or be a non-obvious extension of previous work.

Is it sufficiently described?

Claiming that a method and system uses “all machine learning”, without further details, is unlikely to meet the threshold requirement that a patent application provide sufficient disclosure to support a claimed invention.

Things to Consider When Writing Strong AI Applications

Here are a number of key considerations for writing stronger AI applications to mitigate the aforementioned barriers:

Solve a real world problem

Demonstration How? ‘Or’ What AI solving a real-world problem is an important factor in representing patentable subject matter in many jurisdictions.

European patent practice, for example, emphasizes “applied AI”, which shows how the AI model is applied to improve or solve a specific technical problem. This makes it possible to establish that the invention makes a technical contribution and is not a simple mathematical algorithm.

To this end, demonstrate how well known machine learning models are adapted at new use cases can also show novelty.

Here are some examples of technical applications for AI:

apply AI to image processing; Where
apply AI for medical heart rate monitoring.

Improve an IT function

Even if AI is not practically applied, an AI invention may have other types of improvements to make it patentable. For example, if the improvement relates to improved computer operation, it may also demonstrate a physical result beyond a simple mathematical formula. This type of framing is more suitable for “Core AI” type inventions. That is, an invention in the actual AI model, rather than its application to a specific use case.

Here are some examples of IT improvements:

adapt an AI model to fit unique low-processing hardware applications; Where
reduce or optimize some model parameters to generate faster computational outputs while maintaining or improving predictive accuracy (e.g., reducing the size or layers of networks, generating outputs using less training data, etc.).

Surrounding material environment

Machine learning models are generally not applied in a vacuum, but are often applied in a specific environment context. For example, models are applied in contexts involving receiving physical data from sensors and processing that data. This is related to the specific use case of the model.

In particular, the description of this material environment can add a physical dimension to the claimed invention, which also helps to overcome the obstacle of patentable subject matter. Moreover, it can add a dimension of novelty to the overall concept.

The following are examples of surrounding hardware environments:

obtain ECG and other biometric data from physical sensors, attached to the patient, and integrate them into real-time heartbeat monitoring AI; Where
acquire images from a physical camera mounted on a vehicle and transmit them to an object detection AI in real time.

Post-processing model output

Even further, how is the output(s) of the model processed later can also address patentable subject matter and novelty barriers. For example, if the output is used to make a unique physical change in the real world, it may demonstrate something more than a mathematical algorithm and may also contribute to the novelty of the overall concept.

Here are examples of post-processing model outputs:

using the output of an image processing AI to perform feedback control on the motion of a vehicle; Where
using the output of a voice-recognized artificial intelligence, for example in a home assistance system, to perform feedback control on the temperature or lighting inside a home.

Description of the trained model

Despite the temptation, it is often insufficient to describe in general terms that “any” machine learning model can implement the idea. While this is generally true, the request should at least describe a possible implementation to meet disclosure requirements in many jurisdictions. Additionally, as noted above, there may also be novelty in the development and description of a new model architecture.

Here are some examples of considerations:

describe the configuration of the model layers (for example, the use and arrangement of certain types of layers);
if the model is a “plug and play” model, describe any adaptations to fit specific use cases or applications;
describing the type of input data required for the model, as well as any pre-processing of the input data (eg data labeling); and or
describing the types of output data generated.

How is the model trained?

Finally, explaining how the machine learning model is trained can also, in some cases, be an important factor in meeting the sufficient disclosure requirement. Additionally, there may still be additional novelty in how the model is trained to fit a specific use case, or otherwise how it is trained to generate a more accurately trained model.