From self-driving vehicles and autonomous drones, to virtual doctors and automated personal assistants, AI is expected to fundamentally disrupt the way that people live, work and interact with each other.
AI is increasingly the key to significant innovations across almost all segments of society, manifesting itself in vastly different applications. For example, in July 2019:
- the first human vaccine that was wholly created by an AI program began clinical trials in the United States. The AI program, known as SAM (Search Algorithm for Ligands), was designed by researchers at Flinders University in Adelaide; and
There are vast opportunities for businesses operating in industries where AI has become more prevalent. However, with these opportunities come significant challenges. Ashurst's series of articles on AI will consider these issues from an IP perspective.
This first article in the "AI: Understanding the IP" series, provides a brief introduction to the protection of IP in AI. A key question for businesses to consider when dealing with AI is what kind of IP protection is most appropriate to protect the AI technology. While we consider patent, copyright and trade secret protection below, the answer to this question will vary depending on the type of AI and the intended strategy of the business in using and promoting their AI on the market.
Patents
According to the World Intellectual Property Organisation (WIPO), since the emergence of primitive AI in the 1950s, researchers have published over 1.6 million AI-related scientific publications and filed patent applications for nearly 340,000 AI-related inventions. Over half of these inventions have been published since 2013.
This recent AI patent boom is indicative of the powerful role patents play in the AI space. In Australia, patents provide the inventor with the exclusive right to "exploit" the relevant AI technology for a limited term (up to 20 years), in exchange for public disclosure of how the invention works.
Patent protection is highly valued among AI developers as patents give the patent owner a monopoly. Therefore, any competitors who are found to have exploited the patented AI technology will be ordered to cease the infringing conduct even if they had no knowledge of the patent or patented technology at the time of developing their technology (although lack of knowledge may be relevant to whether the Court also awards damages or an account of profits).
This is particularly relevant in the AI context, given a substantial number of AI patents are being filed for variations of similar AI techniques and functional applications. For example, WIPO estimates that approximately 40% of all AI-related patents disclose machine learning (including deep learning and neural networks) as the patented technique, and approximately half of all AI-related patents disclose computer vision (including image recognition) as the functional application. With AI technology being concentrated in these fields, it is conceivable that inventors may independently develop overlapping technologies such that the first inventor to obtain patent protection will obtain a valuable position over their competitors using technology which falls within the scope of their patented invention.
However, patents can be expensive and take a long time to proceed to grant. Further, patents may not be granted for AI technology or, if they are granted, may be subject to validity challenges. There is an ongoing and unsettled debate in Australia regarding the patentability of computer implemented inventions. In fact, this issue was recently considered by an enlarged five-judge bench of the Full Federal Court of Australia (see our article "Full Court declines to stipulate boundaries for patentable computer-implemented inventions" here). We will address the patentability question in more detail later in the series.
Copyright
In Australia, computer programs are protected by copyright as literary works. Copyright protection arises automatically upon creation (making it much cheaper than a patent) and extends to the expression of the source code embodied in the software that underlies the AI technology. In contrast to patents, a copyright term is much longer (the author's life plus 70 years).
Importantly, copyright does not protect ideas or the functional aspects of software. Further, copyright only prevents copying. That is, an infringer has to have copied the whole or a substantial part of the source code to infringe.
Using the AI poker bot Pluribus as an example, copyright would not protect the algorithm or method designed by the research team that designates the poker strategy employed by Pluribus. Copyright protection would, however, extend to the source code that embodies the expression of this strategy in the software. Accordingly, copyright protection alone would not prevent another person from designing a different AI bot using the same algorithm as Pluribus, provided they employed different coding to implement the algorithm.
Trade secrets
Trade secrets can be a highly effective means of protecting valuable business information. A trade secret is a technical term describing a category of information that has a quality of intrinsic confidentiality, the secrecy of which is maintained by its owner. Trade secrets are enforced through actions for breach of confidence or, if a non-disclosure agreement or clause exists, breach of contract.
Like copyright, no application or registration is required to obtain trade secret protection; rather, trade secret protection arises automatically provided that the trade secret owner can demonstrate that the information has not been widely disseminated and, if it were disclosed to a competitor, it would be liable to cause real (or significant) harm to the owner of the secret. Trade secret protection can theoretically last forever, as long as secrecy is maintained and the information is not publicly known.
Provided that the technology cannot be reverse engineered, AI technology may be suited to trade secret protection. It is often the case that the most competitively valuable information in a computer implemented product is the algorithm. Consumers of the product interact only with the AI interface and will typically not have access to the algorithm. This means the algorithm could be protected as a trade secret provided the appropriate security measures were in place. In fact, Google's search algorithm is a famous trade secret.
The key advantage that trade secret protection provides over patents and copyright is that trade secrets can protect a broader range of information (including business methods, inventions, and even original ideas in certain circumstances) on the condition that the information is kept secret.
Businesses intending to rely on trade secret protection should put strict measures in place to prevent the secret being inadvertently or deliberately disclosed. However, even with reasonable security measures in place, there are risks in relying on trade secret protection to safeguard AI technology.
Notably, trade secret protection can be instantaneously and irreversibly lost if the secret is disclosed publicly. Although the business may be able to pursue an action for breach of confidence or breach of contract against the party that disclosed the secret, it would be next to impossible to prevent the secret from circulating online once it has leaked. Further, even if the trade secret remains in place, the secret holder has no recourse if a competitor independently develops the same AI technology that is protected by the secret.
Conclusion
Patents, copyright and trade secrets are all viable means of protecting AI technology. However, the right approach is dependent on many factors including: the type of AI to be protected; the likely lifespan of the technology; the value of the AI; and its importance to the business. As we progress through the AI and IP series, we will look at the various IP challenges presented by AI in more detail. So stay tuned…
Authors: Nina Fitzgerald, Partner; Eoin Martyn, Senior Associate; and Andrew McClenahan, Lawyer.
