This article is the last in a series of HLK articles on patents for artificial intelligence. Here are our top tips to ensure you get the most out of Europe’s pool of AI patent applications.
Is it worth filing a patent application at all?
If the invention of AI is “hidden” and not easy to detect, it is tempting to think that it is not commercially viable to file patents for such applications at all. but, As we wrote beforeThe upcoming transparency regulation of AI and standardization efforts means that in many industries hidden AI will be pushed into the open. In fact, due to the standardization efforts in AI, some of these things may end up being among the most valuable patents in the portfolio.
For example, it would likely make commercial sense to submit applications for inventions that would be used in the UK public sector, used in safety-critical systems, or in healthcare, among other things. These types of fields offer the greatest chance of obtaining standard core patents. Moreover, even if your competitors are not compelled to be transparent by law, in these areas they may also effectively be forced to disclose their methods for commercial reasons which will also expose the technology. Afterall, who would want the mysterious “black box” as a doctor or pilot?
Determine what type of AI invention you have and whether it is excluded from the subject eligibility
If you decide that you want to pursue patent protection, be aware of how the invention is likely to be presented by the European Patent Office and whether it falls into one of the excluded subject categories. As we wrote beforethe EPO generally consider Applied AI inventions, where AI is applied to a technical problem (eg image classification, communication, etc.) better than Core-AI inventions where the inventor has improved the AI field itself (eg types new models and methods of training or preprocessing of data).
If you conclude that your invention is a Core-AI invention, don’t despair! But take the extra time to think about how your organization might want to monetize the invention. If there are particular licensing opportunities, or products you are considering that will incorporate the invention, use cases should be added to the patent application describing how the invention will be applied to these particular technology areas. Use cases can be used as a basis for narrowing the scope of claims in prosecution, in a way that still results in a commercially useful scope.
Put the right level of detail into the app
The European Patent Office grants a patent application only if the application is ‘sufficient’. In Europe, sufficiency is a legal requirement that a patent application must contain sufficient details for a person skilled in the art to be able to re-create the invention. Thus, if an applicant wants to ensure that their investment portfolios do not conflict with this requirement, care must be taken to ensure that applications contain the correct level of detail.
To determine the level of detail required in patent applications for AI inventions, the European Patent Office has provided two main sources of guidance:
Case law in this area, such as T0161/18, indicates that the EPO considers a high level of detail required for AI applications. T0161/18 suggests that detailed information about the training data, such as an example of a training data set, or detailed information on how to compile this training data set is necessary for the application to be sufficient. This level of detail places a huge burden on applicants.
However, it should be noted that the adequacy of disclosure is assessed at the date of filing the application. The application in T0161/18 has a priority date from 2005, when commercial use of AI was in its infancy, so perhaps it could be argued that in 2005, it would have been an undue burden for a skilled person to have to compile a training data set on Limited information basis.
It seems to me much more difficult to argue that the same case brought today is insufficient, however. With AI ubiquitous and easy access to open-source development tools, a developer working in this field might easily be able to compile a training dataset and train different test models, almost in real time. Thus, we may expect the sufficiency criterion to change over time.
As another source of guidance, the EPO previously noted that applications describing machine learning models can be given the benefit of the doubt as to adequacy if it is readily apparent that the inputs and outputs of the model causally related. The example given was that less information might be needed in the description of the application A neural network trained to detect faces in images obtained using an infrared cameracompared to the description of the application Predict IQ from fingerprints.
This makes sense, because if a causal link is not known, or at least plausible, between two types of data, then trying to patent an ML model that can transform one type of data into the other is not much different than trying to patent a proverb perpetual motion machine It’s not clear how that would work either.
As a general guideline, if there is any doubt about how the inputs and outputs of a form are related, more information should be added to the application in order to provide enough detail for someone else to recreate the claimed form.
We go further to suggest that this applies more broadly, for example, to Core-AI inventions where the inventor has made an improvement on some type of model (eg invented a new type of neural network, or a new method of training) often What is not obvious (on an intuitive or at least normal level) Why The particular change made works the way it does. New architectural models, for example, can be opaque in terms of How Special configurations achieve the effects they do.
In such cases, if the patent application is granted, it is best that you provide in the application more information than you may consider necessary.
What information can be provided to improve grant opportunities?
A detailed example should be provided, including, for example, a fully described architecture for a given model type, parameter setting, and training data that can be used. You can refer to an open source data set or refer to an appropriate data set known in academia.
If a causal link is not readily apparent, any information supporting an association may be added. Academic papers can be a reference source here as well.
Experimental data can also be added to demonstrate that the claimed effect can indeed be achieved. In the case of an improved model, comparison data can be added to an application, showing, for example, an improvement over a prior art example trained in the same way on the same data.
In short, it is possible to obtain commercially useful patents for AI applications in Europe. It can serve as a balance between choosing which inventions can be discovered, and also which can be patented before the European Patent Office. When you find an app that strikes this balance, be sure to enter enough information to give your app the best chance of achieving a good result.