Ex Aula - Logo4-cropped

‘Artificial Intelligence Just Discovered New Planets’!
‘A New Sensor Gives Driverless Cars a Human-Like View of the World’!
‘Google supercomputer creates its own ‘AI child’ that can outperform any machine made by humans’!

 

Recent headlines reporting technological developments could easily be confused with science fiction, thanks to very real advances in artificial intelligence (AI).

In fact, the use of AI in technology has become so widespread that it can often be taken for granted. We can now instruct our phones to call friends or bring up information using our voices, translate text to almost any other language instantaneously, and we will soon be passengers in our own cars as driverless cars become mainstream. KITT, the smart car from the 80s classic, Knight Rider, really could be parked on your driveway’.

 

But what is AI?

Essentially, AI is when machines display intelligence usually associated with humans – such as the ability to perceive the outside world, learn and problem solve – to carry out tasks. Through AI, computers are capable of processing human speech, sorting through vast amounts of data, or manipulating objects, in order to achieve a goal.

AI is not a new phenomenon. In 1950, Alan Turing wrote an article called ‘Computing Machinery and Intelligence’, in which he first posited the idea of ‘learning machines’ which could simulate the brain of a child. However, advances in AI, and in particular machine learning, have proliferated in recent times due to increased computing power and the huge amount of data made available by the internet.

Machine learning is a sub-field of AI where large data sets are fed into algorithms that can make predictions and decisions based on that data. In one such instance, they can do this through artificial neural networks, which are systems modelled on the relationship between neurons in the human brain. This allows the computer to create outputs that classify or detect relationships between data. Such algorithms are adaptable to allow the computer to learn from its mistakes and ultimately become more efficient at completing tasks.

 

So where does intellectual property law come in?

Legal protection is available for new, inventive and industrially applicable inventions.

Patents, which have been enshrined in English law since the Statute of Monopolies in 1623, provide an inventor with a monopoly over their intellectual creation. They provide the inventor with the exclusive right to make, sell, use or import his or her invention for a period of 20 years. The classic justification for such a monopoly is that it will allow for the inventor to recoup research and development costs and incentivize further innovation.

Since the genesis of computers and computer-related inventions in the twentieth century, there has been a debate over whether software inventions should in fact be patentable. Large companies and patent attorneys have generally argued that software inventions ought to be treated like any other inventions and patent protection should be available. Others, usually smaller companies and individual programmers or lobby groups, have argued that software is inherently different from physical mechanical or chemical inventions and ought to be treated differently.

The way this issue has been dealt with by the law has been inconsistent.

Currently:

In the United States, Alice v CLS Bank, a 2014 decision by the Supreme Court held that software inventions are patentable provided that if they are directed to an abstract idea they must provide “something more” which transforms it to become patent eligible. In practice, if an algorithm is applied to achieve something in a technical field, it will be patentable in the US.

In Europe, the European Patent Convention sets out that programs for computers are not patentable “as such”. The two words “as such” left it open for computer programs to be patentable in some instances. As a result, a body of law has developed whereby software inventions that have a “technical effect” are in fact patentable.

While the purpose of patent protection is to support and stimulate innovation, it is possible that in some cases patents have the opposite effect. It is therefore important to reconsider the wider effects of patent protection for software, in particular where AI and machine learning processes are the subject of such patents, given the surge of inventions utilizing these techniques.

 

Patents for AI inventions

Patents for AI inventions have increased exponentially since 2014 for both inventions that apply AI such as driverless cars, and for AI algorithms themselves which set out techniques such as supervised learning. The latter in particular present a potential problem for the future development of AI. One recent example is US Patent No. 9,760,834 “Discovery systems for identifying entities that have a target property”. Essentially this patent describes the creation of models to analyse proteins by using machine learning. This has been called out by the Electronic Frontier Foundation as ‘Stupid Patent of the Month’ for reading “like the table of contents of an into to AI textbook”.

Through my research I am considering empirical data detailing the AI-related patents being granted, who they are being granted to, the possible effects on the future of innovation in the field, and potential alternatives.

Given the still relatively early stages of development of AI innovation, it is important that key techniques already used across the board are not monopolized by one entity. Otherwise this could lead to restriction of innovation to one (potentially unworthy) source, or hinder innovation in a particular area altogether if they fall into the hands of patent trolls (entities which simply buy and threaten to enforce patents but do not produce or develop inventions).

A key underlying question is whether the traditional incentives for extending patent protection hold up in the modern era and in the current innovation landscape. It is possible that the nature of software and the rate of development even without patent protection (consider, for instance, the ubiquity of Open Source Software) suggest that the recouping of research and development costs is not the primary incentive driving development of this technology.

Perhaps pro-active consideration of these issues by law and policymakers will act as a timely deus ex machina to resolve any potential difficulties.

 

 

– Sarah Rosanowski