Although there are often misconceptions about it, AI has the potential to be protected like all other areas of technology innovation. There is often the mistaken understanding that technology like AI, which is implemented using computers, cannot be patented. Whilst there are practical challenges to obtaining patent protection, patent offices around the world regularly grant patents on AI inventions.
To be patentable, all inventions must be new and provide some non-obvious contribution to what is already known and the same rules apply for AI technology. There are other requirements which exclude some inventions which are considered to be unsuitable for patent protection or are covered by other areas of IP. These tend to be the more abstract or non-technical areas such as presentation of information, mental acts, recipes, rules for games and so on. Patents are primarily directed at protecting technology but it can mean certain types of the invention are harder to protect with patents.
AI is inherently implemented in technology but it also inevitably branches out into other areas such as display of information, automation of tasks previously carried out by humans and so on. These areas can fall foul of the exclusions to patentability if the invention for which protection is sought is solely in these areas. It is therefore important to focus patent protection on what is clever about the underlying technology and focus less on what might seem to be the improved user-facing features of the product it is implemented in. For example, a product may be a complete social media platform with lots of nice features that elegantly present messages exchanged by users. However, the protectable technology may be a small part of the underlying system, for example in the way the messages are encrypted when they are being transmitted.
AI covers a broad spectrum of technology that typically aims to replicate things that humans have previously only been able to do, such as driving a car. AI often makes use of machine learning where a system observes real-world systems and perhaps experiments or simulates scenarios to determine what are the best actions in certain circumstances, such as a pedestrian stepping out in front of a car. Having analysed vast amounts of data, a model may be developed to provide a system for providing desirable outputs (swerving the vehicle) from a complex set of inputs (cameras, sensors etc.). The precise structure of this model is often extremely complex based on a large number of inputs and may make it impossible to define the relationship between the inputs and the outputs, which can make such systems difficult to protect in a patent.
Therefore when seeking to protect an AI system, it can be important to look beyond the user interface and user functionality and inside the box to identify the techniques and processes that are demonstrably technical and which can be legally defined. There may be many such features and it is then a matter of identifying those which offer the greatest advantage to the product or put another way, without which a competing product would be significantly inferior.