Patenting Artificial Intelligence
It is hard to avoid the AI revolution that is taking place around us. Indeed, in a lot of areas this revolution is already mature. We have come a long way from teaching computers to play draughts and chess or tell the differences between cats and dogs and it often seems like your smartphone knows more about what where you are likely to go next (and what the weather, traffic and COVID infection rates are there) than you do.
This "revolution" is not just evident in the high-tech, computer-focused fields that are traditionally associated with AI. The latest breakthrough in biological or pharmaceutical research is increasingly likely to have resulted from a "random forest" or an SVM as from a well-plate or wet assay. Computer scientists are in demand not only to create game worlds in the latest MMORPG but to sift, organise and exploit the vast data resources available from genome projects, customer loyalty cards and social media.
From an IP perspective, AI poses its own challenges (and not just in the widely-reported recent debates about whether a computer can be an "inventor" or not). Contrary to popular opinion, AI algorithms and techniques can be patentable, albeit often not in themselves, but certainly in their application. The key question is whether the AI technique is being used to solve a "technical problem" which, in broad terms, means something external to the computer itself, and related to the fields of science and engineering rather than business and commerce. Examples of technical problems for which AI solutions have been successfully patented include autonomous driving applications, such as threat detection and analysis and vehicle navigation, cyber security, personalised medicine and drug discovery, to name but a few.
However, just because an AI technique can be patented does not always mean that it is a company's commercial interest to apply for a patent. The very nature of AI is that many such techniques operate entirely within their own "black box" and whilst their results may be evident to all who encounter them, their inner workings remain hidden to all but those who programmed and, where appropriate, trained them. Patenting necessarily involves publication and disclosure of many of these inner workings to the world at large.
Therefore there is frequently an important choice to be made between keeping the details of the operation of the algorithms and their inputs secret, thereby reducing the information available to those seeking to compete, and seeking patent protection which, if successful, could exclude any third party development of competing solutions, not just attempted copies.
Unsurprisingly, there is no "one-size-fits-all" answer to this question, with a multitude of factors applicable in each case. Typically, this includes the extent to which it is necessary, whether due to regulatory requirements or customer comfort or otherwise, to disclose some information about the process, the ease with which competitors could develop a similar system from the publicly-available information and the risks of knowledge transfer as employees (and indeed founders) move on to new ventures.