Protecting inventions which use Machine Learning and Artificial Intelligence – Lexology

Protecting inventions which use Machine Learning and Artificial Intelligence

There has been a lot of talk recently about the DABUS family of patent applications where DABUS, an artificial intelligence (AI), was named as an inventor. This has prompted a lot of discussion around whether an inventor must be a human being and there is no doubt that this discussion will continue as AI finds its way into more and more aspects of our lives.

However, one of the other parts of the discussion around AI in patents is around the patentability of inventions which apply machine learning (ML) and AI based concepts to the solution of technical problems.

Why consider patent protection?

Patents protect technical innovations and technical solutions to problems. They can offer broad legal protection for the technical concept you develop, albeit in exchange for disclosure of the invention.

Here in the UK, a patent can give you the right to prevent others from exploiting your invention and can help you to mark out legal exclusivity around a patented product.

Can I not just keep the invention a secret?

It is an option to utilise the invention as a trade secret, but the protection of the trade secret involves considerable effort to implement the technical and administrative environment which will enable the trade secret to stay as a secret. This can include changing your physical workplace to confine certain environments where trade secret-protected inventions are being used. This can also include implementing technical measures to inhibit access to trade secrets from unauthorised individuals. Such technical measures are particularly important for AI and ML-focused inventions as they are often embodied in computer program code which can simply be transferred from one computer to another

What is perhaps more pertinent is that if your AI or ML-enabled concept is to be implemented in association with hardware which is to be sold publicly, then this will by definition negate the value of the concept as a trade secret as it will become publicly available. It may require decompilation or reverse engineering to access the code, but this does not mean that the code is secret.

There may be additional know-how associated with your invention which is worth protecting as a trade secret but as part of a suite of IP rights (including patents) which are focused on protecting your invention.

How much information does the patent application require?

All patent applications are drafted for the skilled person who in this context would be somebody skilled in the techniques of ML and AI, although not necessarily an expert. That is to say, it needs to be enough information to enable such a person to put the invention into effect.

This should include technical information about features which provide an advantage over previous systems and clear identification of advantageous features and why they are advantageous. This will give your Patent Attorney the best possible chance of framing the invention in a way which convinces patent offices around the world to grant a patent.

It is also advisable to include disclosure of at least one set of training data and details of how it has been trained.

In the context of AI and ML it is particularly important to draw attention to technically advantageous features as some patent offices will need a lot of convincing to grant patents for these inventions. It is particularly useful to draw attention to features which solve technical problems or are motivated by technical considerations rather than economic or commercial considerations.

The EPO have stressed that patents will be granted when ML or AI based inventions are limited to a specific technical application or required a specific technical implementation which are directed to a technical purpose. These advantages and details of implementation will enable a patent attorney skilled in drafting patent applications for ML/AI to present your invention in the best light as possible from the perspective of the EPO or the UKIPO as they will enable us to clearly set out how the invention delivers the technical application and solves the technical problem.

Our software patents team are specifically noted for their skill in drafting computer implemented inventions for the UKIPO and the EPO.

Although a lot of information is required, we do not necessarily need program code. It would help, however, to at least include a pseudocode description of the invention so that we can garner an understanding of how the invention works as a series of steps this helps with the description.

Are AI and ML not just like software, i.e. cannot be patented?

It is possible to patent software-based inventions but, like other inventions, the invention needs to solve a technical problem. This is the same with inventions which apply AI and ML.

AI and ML inventions are treated in Europe like other mathematical methods in that they are rejected as excluded from patentability if they do not solve a technical problem. It is best to illustrate this by example.

If your invention is to improve a technique which is used to analyse data such as, for example, your invention improves K-means clustering with no other benefit to a technical field, then you can expect to face considerable obstacles to obtaining a patent to protect your invention. However, if your invention applies K-means clustering to achieve a specific improvement to a specific technical system then you are likely to face less obstacle to obtaining a patent for your invention.

That is to say, when considering whether you wish to pursue patent protection for the technology you have developed then focus on what the innovation achieves in a technical field.

What if the technique has been applied elsewhere? Can I still get a patent?

Referring back to our K-means clustering example, if you see that K-means clustering has been used in sensing of rain droplets on a car window to determine the appropriate setting for the windscreen wipers, then that does not necessarily mean that you cannot get a patent for K-means clustering applied to determining the likelihood of a denial of service attack on a server.

That is to say, if you are applying known technology to a new field and solving a technical problem in that field, there is an arguable case for a patentable invention.

Are there differences between Europe, US and other jurisdictions?

The approach to these inventions across jurisdictions can be different and complete consistency is difficult to guarantee. However, in drafting your patent application we would seek to make the language as flexible as possible in order to admit differing interpretations of the law across jurisdictions and to give the prosecution of your patent applications in those jurisdictions the greatest possible chance of success.

What do I do next?

If you have developed technology which applies AI or ML, then consider whether you could achieve patent protection for that invention. Contact one of our software patent experts to discuss the invention and your options.

It is also useful to note that having a pending patent application can be a useful deterrent for competitors and the uncertainty created for third parties by the existence of the patent application can provide you with the space in the market to establish your exclusivity, develop your customer base and build your brand.

See original here:
Protecting inventions which use Machine Learning and Artificial Intelligence - Lexology