Patent-AI-ble Inventions

However, whilst the ubiquitous question of “will AI take my job?” has had competition for the headlines with the rapid take up and development of Chat GPT and other generative AI systems, developments of applied AI in areas as diverse as cybersecurity and personalised medicine have been racing forward just as fast. These have been led, in many cases, by Cambridge-based companies and researchers.

AI has also managed to buck the general trend of a reduction in available investment capital in 2023, particularly for start-up and scale-up companies. Whilst the exact numbers quoted vary widely with the sources, there is clearly a sustained appetite for backing AI innovation.

Intellectual property (IP), and particularly patents or patent applications, can play an important part in attracting investment in an early-stage company.

AI innovations can be patented if they meet the right conditions. Developments in core AI (i.e. the functioning of AI systems themselves) which, for example, cause computers to perform AI processing quicker or more efficiently, are generally patentable, although there can be issues with quantifying the improvements.

The patentability of applied AI innovation generally depends on two factors.

First, is the field of application sufficiently “technical” to be patentable? In simple terms, this means that if the outcome of the AI system is a useful determination or result in a scientific or engineering field (including medical diagnosis or prognosis), then a novel AI approach is suitable for patenting.

In contrast, uses of AI in solely business-related fields, such as finance and insurance, or for the display of information (e.g. in advertising) are generally unlikely to be suitable. However, whilst the ends of the spectrum may be clear, there is a large grey area in the middle which will benefit from expert consideration of the individual situation.

The second factor is whether the AI system is doing something both new and inventive, in particular taking account of what was already being done by human intelligence. Thus, automating existing processes using well-known AI capabilities such as image processing may not be inventive, but if the AI system is able to detect and utilise new features (or new combinations of features) which were not previously considered relevant for a determination, then this is a positive indication of inventiveness.

However, the decision on whether or not to file a patent application for an AI innovation can be more complex than in other fields. Because many AI systems can operate essentially as “black boxes”, the alternative strategy of keeping the details of an innovative approach secret should always be seriously considered.

Deciding between the approaches is complex and consideration will need to be given to factors such as the reproducibility of the algorithms (and the time and effort required to do so), whether third parties will be able to access suitable training data, and whether future disclosure of some or all of the internal workings of the system is likely to be desirable (or even required) for example for regulatory purposes, marketing or customer buy-in.

Finally, it’s worth noting that, whilst patenting AI is largely “in”, AI inventors are definitely “out”. The extensive efforts of the DABUS team to have an AI system recognised as an inventor reached the end of the line with the Supreme Court rule in November that inventors had to be real people. This hopefully means that AI won’t be replacing Cambridge inventors (or patent attorneys…) any time soon!