Artificial Intelligence (AI) is able to impact almost everything in much the same way electricity did in the early 1900s by replacing steam powered machines. For example, AI can transform FinTech, healthcare, logistics, search engines, etc.
The obvious advantages of AI are that errors are reduced, repetitive one second human thought tasks are replaceable (e.g. is that a dog or cat in the photo), scalability and continuous operation. AI is also able to surpass human level capability such as quickly deriving insights from large volumes of data. The benefits to the user include more personalised service (e.g. more targeted advertising to increase sales) and feedback on user behaviour for R&D teams to develop new products/services or improve existing products/services.
The simplest type of AI being widely implemented now in many industries is through a process of supervised machine learning. Supervised machine learning requires a training data set and a known output. Businesses who possess or accumulate a lot of data are in the best position to leverage AI because they have the training data sets required. Devices that generate data (e.g. sensor data from home appliances) or services that generate data (user data like Google search queries) are the fuel for AI.
90% of the world’s data was generated since 2015. 2.5 million terabytes (Tb) of data is globally generated each day. One important enabler is that the storage cost for data has dramatically fallen from $500,000 per gigabyte in 1981 to $0.03 per gigabyte today. Another enabler is cloud computing which provides businesses with the ability to outsource data storage that is scalable and therefore eliminate a substantial upfront purchase expense and reduce ongoing maintenance costs.
Data is the new economic moat
Many applications require a lot of data. For example, 150,000 hours (10 years of audio data) of training data is required for very good speech recognition and a minimum of 15 million images are required for computer vision algorithms to recognise faces although 200 million images would increase accuracy. Therefore businesses with vast amounts of data can maintain a sustainable competitive advantage.
The modern role of data is to provide an economic moat. For example, it would be very difficult for a startup to displace Google in the search engine industry because Google has been accumulating almost all data for the last 19 years and leverages the data using AI.
Almost all the providers of activity trackers typically sell their devices without a high margin because the data they accumulate can be monetised to possibly serve out targeted advertisements to sedentary people for gym memberships or weight loss programs. Smart assistants like Google Home and Amazon Echo are also priced low because the data that is gathered is the profit centre.
What about traditional moat-maker – patents?
Patents are also a source of sustainable competitive advantage for a limited duration. Patents prevent others from practising the invention without authorisation. However, there are drawbacks in choosing to obtain granted patent rights. It is expensive compared to other forms of intellectual property (IP) protection (copyright and trade secret).
On average it takes about 20 months for a patent application to be initially examined, and 30 months until it is granted (assuming the claimed invention is new, inventive and is directed to patentable subject matter). There is also a risk that if a patent is sought but not granted and then refused or invalidated, the subject matter may have become public, rendering not only patent protection, but also trade secret protection unavailable.
However, having a pending patent application has some value because it may attract investors, is perceived as an indicator of an innovative business and may ward off competition.
Certain AI components are suitable for protection as trade secrets rather than patents. The general rule should be that the more valuable a particular AI trade secret is to a business, the more resources should be applied to its protection.
If an AI innovation is more suited towards patent protection rather than a trade secret, for example, because it is easily reverse engineerable from public observation, then an appropriate enquiry is whether such an innovation is considered patentable subject matter.
Business methods are unpatentable but is AI innovation eligible for patent protection?
In Commissioner of Patents v RPL Central Pty Ltd  FCAFC 177, the Full Federal Court of Australia suggests that a computer that “evaluate[s] the user’s input to provide [an] answer … [and] functioning in the nature of an adviser or an artificial intelligence” is an invention or ingenuity in the program or operation of a computer. Consequently, such innovations would constitute patentable subject matter in Australia. It is the actual technical implementation of a method by the computer that is important, even if that method is a business method.
Characterising the problem solved by the invention in terms of a business or commercial problem should be avoided, and the focus should be exclusively directed to a technical problem. Some examples of technical problems include assembling text in Chinese language characters using a non-Chinese keyboard (AU patent 61654); production of an improved curve image by computer (AU patent 629173); and eliminating delay in awarding customer benefits the first time the customer uses their smart card at a merchant’s point of sale terminal (AU patent 712925B2).
However, there are examples where certain software-based inventions were found not to be a technical implementation providing a technical contribution that improved the computer.
The generation of intellectual information in the form of a musical notation or scale that can be used in composing or playing music in the case of Svetko Lisica  APO 44 (1 August 2013) was not considered an improvement in the operation of or effect of the use of the computer, or an improved use of computers by the Australian Patent Office.
Automated safety assessment in the case of Swiss Reinsurance Company Limited  APO 12 (15 March 2017) was rejected by the Australian Patent Office because the substance of the invention did not lie in the application of artificial intelligence to safety assessments.
Determining a trigger condition based on engagement data at an engagement engine by continuously evaluating data in the case of Rokt Pte Ltd  APO 34 (11 July 2017) was rejected by the Australian Patent Office as constituting the substance of the invention. It was found that there was no contribution in the computer systems involved in the evaluative process, and that the contribution lies in the evaluative process itself (which was found to be a mere scheme for improving customer engagement with advertisements).
AI is expected to have a very significant impact on most industries. The businesses that innovate the technology to develop a sustainable competitive advantage based on data leveraged using AI should also consider placing additional barriers to entry by choosing an appropriate type of IP strategy most suitable in the circumstance.