Can My Computer Own a Patent?

In recent years, interest has been growing in the development of artificial intelligence (AI) and the potential use of AI in different technologies for improving our lives. One example of high interest is the application of AI in self-driving cars. Other areas of technology such as medicine, where AI might be used in automated diagnosis of disease, are also of interest. However, recent advances in the AI field has led to a particular question - if AI development becomes advanced enough that some form of AI might itself develop a new invention, how would the patent system deal with this?

AI vs Machine learning

However, before dealing with this question, it's worth taking a step back to consider what AI actually is. The portrayal of AI in films and other media has led to a possible disconnect between what many people consider to be an AI compared to the current state of technology. Rather than the thinking, interactive, computer intelligences often portrayed in media, the "AI" presently being developed to solve problems is quite different. Modern AI systems use computer algorithms that in many ways don't differ from more straightforward decision making algorithms used in the past.

A key distinction behind some modern AI technology is that the algorithm used has been trained and modified in an automated way to be good at performing the required task, such that the creators of the system may no longer fully understand how the algorithm arrives at a correct result. The difference between these systems and what someone not familiar with the field might consider an AI has led many researchers and developers to prefer the term "machine learning" to describe the field as a whole.

So what's the problem for the patent system?

Before considering the issue, the first point is that many of the developments involved in creating a trained machine learning algorithm used to solve a particular problem (such as image recognition for self-driving cars) are inventions that for which, with some caveats, a patent can be obtained. However, developments in the machine learning/AI field present a potential problem for the patent system if an algorithm was made that in some way was capable of itself producing patentable inventions.

The problem lies with who has the right to apply for a patent to such an idea. To deal with, for example, disputes around ownership, the UK Patents Act sets out that the right to apply for a patent lies with the inventor, who is "the actual deviser of the invention". Various pieces of case law have added more context to this definition and presently it is a requirement for an inventor to be a human. A similar requirement exists around the world in other major patent systems. So, how should the patent system respond if someone trains an "AI" algorithm to produce inventions?

AI as a tool for the inventor

This question is no longer completely theoretical. The DABUS "creativity machine" is an AI system trained using several "knowledge areas" that has produced inventions for which the creator of DABUS has applied for several patent applications, naming the AI system itself as the inventor. While the merit of the inventions themselves has not been considered, various patent offices around the world have rejected these applications on the grounds that a patent requires a human inventor. The offices have given the opportunity for the creator of DABUS to be named as the inventor for the applications, but the creator maintains that DABUS itself should be named.

The approach of the various offices follows the line of thinking that modern trained algorithms should be considered as tools that might assist a human inventor, similar to computer modelling software. For example, a turbine engineer might use software to model the flow of air around a turbine blade, and though these experiments might arrive a new, superior shape of turbine blade. In these circumstances, the software is a tool that assisted the inventor, but no one would argue that the modelling software should own the resulting patent. Current machine learning algorithms presently seem to fall in a similar category as a tool to assist a human inventor, rather than an invention creator in their own right.

The future

But what about future developments? What if AI/machine learning systems become more sophisticated to the point where they begin to approach the entities in sci-fi films? Here, it is likely that the patent system can rely on more general law having to tackle this problem first, due to the laws regarding ownership of property. Presently, property can only be owned by a human (a piece of software cannot own a house). Patents are also property, which leads to a problem if an AI is considered an inventor as it is not possible for the AI to own the resulting patent under the laws surrounding property rights. When (if?) AI systems develop further, much more fundamental questions regarding the law will arise and it is likely patent law can simply follow the trend in these areas.