What's infringement got to do with it?
Parminder Kaur Lally
IP Strategist | Mentor to Startups | Patent Attorney | AI Geek
“That’s not my invention!”?
Inventors are sometimes surprised when they read a patent application that has been written for their invention. The surprise can arise because their invention has been disassembled and reassembled by a patent attorney.?The end result may be a set of claims that an inventor feels doesn’t look like their invention.?Setting aside the times when we have genuinely misunderstood their invention, often this happens for patent law reasons.?For example, good patent attorneys will try to write about your invention in broader terms.?So, instead of saying, for example, that data is transmitted via Bluetooth, the claims may refer to “communication means” and the description may include a list of possible communication means, such as Bluetooth, WiFi, wired and wireless communication techniques, and so on.?Another reason the claims may not appear to map directly onto your invention is because a good patent attorney has kept patent infringement in mind while writing about your invention.?
In this edition of the brAIn blog, I’m looking at the claims part of a patent application again.?Previously, I explained how the claims need to define an invention in concrete, technical terms.?This time, I’ll explain how the claims are defined with one eye on how someone might infringe your patent.?
How and why are the claims relevant to patent infringement?
The claims section of a patent application is where we define the subject matter for which we are trying to obtain legal protection.?If you obtain a patent for your invention, you can use it to stop others from committing certain acts with respect to the patented invention.?In the UK, for example, you can stop others from making, using, selling, offering to provide/sell, importing, and keeping/storing your patented invention without your permission.?It is the claims section which defines the scope of the patented invention, i.e. which acts a third party can and cannot perform without your permission as the owner of the patent.?
Patents are also territorial rights.?This means that you can only enforce your patent against a third party in a country where the third party is acting and where your patent is in force.?So, if you only have a patent in the UK but a third party is selling a product that looks like your patented invention in the US, you cannot enforce your UK patent against the third party in the US.?This means your patent attorney will ask you where you intend to commercialise your invention, manufacture your invention and where your competitors are based, as this will help to determine a patent filing strategy.
With respect to software and AI inventions, method claims are common because we are trying to protect the key steps performed by an algorithm or code.?A good patent attorney will think about where each step of the method is being performed, and when.?I’m going to explain this by using some examples to illustrate some common issues that can arise.
Example 1 – training and inference claims
Consider this example method claim:
A computer-implemented method for recognising objects in an image, the method comprising:
generating a trained ML model to analyse an image and recognise objects;
receiving an image for analysis;
using the trained ML model to recognise objects in the received image.
In this example claim, there are three steps.?Two of these steps relate to processing an image using a trained model, and one of these steps relates to training the model in the first place.?
Who would infringe this claim?
In some cases, perhaps the party that trains an ML model also analyses new images using the trained model.?In which case, this claim could be used to stop that party.?
However, in other cases, the party that trains an ML model may supply the trained model to others for use.?This often happens when companies, for example, provide trained models to their customers for use.?In this case, neither the company nor the customers directly infringe the claim because neither performs all three steps of the claim.?This means that no party directly infringes the claims and the patent may be more difficult to enforce.?
When would this claim be infringed?
The claim requires that each time the method is performed to analyse an image, the trained ML model is generated, i.e. the training of the ML model occurs. ?The claim wording means that infringement would only occur if all three steps were performed when an image is analysed.?As models are not trained every single time they are used, it is unlikely this claim would ever be infringed by someone performing image analysis using an already trained model.?
This is why a good patent attorney will divide an AI invention into the steps performed during training, and the steps performed during inference, and will write separate claims for the training and inference stages.
领英推荐
Example 2 – client-server / transmitter-receiver type claims
Consider this example method claim:
A computer-implemented method for analysing images using a trained ML model, the method comprising:
capturing, at a user device, an image;
?????transmitting the image to a server for analysis;
?????analysing the received image using the trained ML model;
?????transmitting a result from the analysis to the user device.
In this example claim, there are four steps.?The first two of these steps are performed by a user device, and the second two of these steps are performed by a server.?
Who would infringe this claim?
As steps of the method are performed by both a user device and a server, neither the owner/user of the user device nor the owner/operator of the server directly infringes the claim.?
A good patent attorney will divide a software/AI invention into the steps performed by each component/device, and will write separate claims for the methods performed by each component.
Where would this claim be infringed??
In many cases, the user device and server may not be located in the same jurisdiction.?This may mean that all of the steps of the method are not performed in the same country, making patent enforcement difficult via a single patent in one jurisdiction.?This is another reason why a patent attorney will divide the invention into the methods performed by each component, which allows the methods to be separately protected in countries of commercial relevance (e.g. the country where the server is located).
Divided Infringement
I’ve mentioned that it can be more difficult to enforce a patent if a single party does not perform all of the steps of a method. In the US, divided infringement is a form of patent infringement liability that occurs when multiple actors are involved in carrying out infringement of a method claim, and no single accused infringer performs all of the steps of a method.?In 2015, the US courts decided that a single party can be responsible for another party’s performance of method steps under certain circumstances:
This doctrine of divided infringement was articulated in the Akamai Technologies, Inc. v. Limelight Networks, Inc. decision of the Federal Circuit.?This means that in some cases, the actions performed by, for example, a user, may actually be attributed to another party.?However, it is better practice to write claims which are separately directed to each party’s activities.?
Key Take-Aways
In summary:
As always, if you want to know more or want to know whether your software or AI invention could be protected using patents, please get in touch with me via email ([email protected]) and I would be happy to have a confidential chat with you!