March 2, 2021

The Rise of the Creative Machines

Recently I had the privilege of delivering a presentation/webinar to IPSANZ on the topic of 'The Rise of the Creative Machines'.  This article is a condensed version of that presentation.  

Developments in Artificial Intelligence

Up until recently, the idea of a “creative machine” was pure science fiction, but the last few years have seen developments in the Artificial Intelligence (AI) field that have made such machines a reality. AIs can now create original images with no human input, or write genuinely moving poetry. News outlets like Forbes use AI to generate financial news stories, which are  indistinguishable from those written by a human. Had these works been created by a human, they would attract copyright protection, but as they are the work of AI, it is unclear if copyright can, or should, protect them.  

Enter, Machine Learning

Although the “AI” label has been attached to a number of different technologies over the years, the AIs responsible for creative works are almost invariably built on some sort of Machine Learning (ML). Put simply, ML feeding an AI massive amounts of data and then training it, rather than programming it, to do a task. In the case of creative machines, this tends to be thousands of creative works (such as poetry, prose, images, screen plays), from which the machine will determine the works’ inherent characteristics, then construct its own versions. The most advanced ML machines use a technique called Deep Learning (DL). A DL system starts with the construction of an “artificial neural network” which is essentially a collection of algorithms modelled after a human brain. The network then uses vast amounts of data to train itself. The AIs that show the most “creativity” are a type of DL system called a GAN, or Generative Adversarial Network. It is these that produce the most human-like work.


Copyright protects creative works, assigning moral and economic rights to the work’s author. The rights assigned to the author are closely linked to the reasons for copyright’s existence. The first reason is economic, and serves to incentivise the creation of certain works by granting a monopoly over the economic benefits of those works to the copyright holder (be that the author, or some other party). This view has seen copyright described as “an incentive that, properly calibrated, can positively affect the creation and availability of knowledge”. Under this view, the purpose of copyright is to provide a mechanism to reward the creators of works that are beneficial to humanity. The second reason is the protection of moral rights, which consist of the right of an author to be recognised as such, their right to object to derogatory treatment of a work, and the right to object to false attribution of a work.

These reasons show that copyright exists to protect things that matter deeply to humans: their culture, their honour and reputation, and their ability to generate the resources they need to survive through their labour. Given that none of those needs or concepts apply to machines, justifying copyright protection for machine-generated works is problematic.

Copyright for AI Generated Work?

The first question, “can copyright subsist in AI generated works?” is not as straight forward as it sounds. Under New Zealand’s Copyright Act 1994 (CA) the answer appears to be “yes”. Section 5(2)(a) specifically states that in the cases of “computer generated works” the person who makes the arrangements for the work’s creation shall be considered the author. However, when the computer involved is an advanced AI, determining who this person is becomes very difficult. Is it the developers, or the company that employs them, or the customer using the machine? The question becomes even more difficult once Deep Learning (DL) technology is considered. An AI built on DL techniques (such as the creative GANs) has, for the most part, taught itself, so finding a person responsible for the creative output of an AI is exceedingly difficult. When you dig deeper into the way these machines work, its not unreasonable to conclude that the AI itself is the author. If this is the case, we are well outside anything the Copyright Act anticipates.

AI and the Law

Given that current legislation is not set up to deal with AI, we need to look at the common-law, which suggests that copyright will not subsist in AI generated works. The issue of non-human creators has recently been dealt with in a US case. In Naruto et al v. David Slater, more commonly known as the “monkey selfie case”, the Court determined that, because he is an animal, Naruto (the photogenic crested macaque who took the selfies) lacked standing, as an animal is “not an author within the meaning of the Copyright Act”. For our purposes, Naruto stands for the premise that a non-human entity cannot be considered an author under existing laws, therefore works created by non-humans do not attract copyright.

Having established that currently copyright probably does not apply to AI-generated works, the question now becomes “should it?”. Looking at the purpose of copyright as it currently exists then the answer is probably “no”.  Quite apart from the fact that neither of the reasons for copyright’s existence (discussed above) can possibly apply to a machine, if we were to say that an AI can be considered an author for the purposes of the Copyright Act, what we are in fact saying is that a machine has rights. The flipside of rights is always the associated obligations, however a machine will have no way to meet them.

Copyright is inextricably linked to the basic human needs and rights it protects.

No items found.

The overall impact of AI creativity will depend on how we deal with the rights inherent in AI created works. Various writers have discussed several options, and the two most suited are presented below:

A. The User, Developer or Owner of the AI as Copyright Owner

An efficient solution would be to say that any work created by an AI (as opposed to a traditional computer) is treated as a “work made for hire”. Then it would be simply a case of determining who had “hired” the machine, which would usually be covered in a contract, such as an End-User License Agreement, or in an AI-as-a-Service agreement.

This would incentivise the use of AI to generate works, but may result in a glut of machine produced low-quality artistic or journalistic output that may have a negative impact on society. Art and journalism are both vital to a well-informed polity, therefore if the overall quality of either degenerates, as would be the case if human creators were crowded out by over-incentivised machines, societies will be less informed and engaged, leading to lower quality  discourse, and a less informed, therefore poorer, democracy. This is, perhaps, a long bow to draw.

The second argument against copyright for machine created work is simpler:  as AI is just code, and that code is already protected by copyright, then surely granting copyright for the results of the machine executing its code is incentivising twice?

B. Direct entry to the Public Domain

This approach would see any an AI’s creative output immediately become part of the public domain, meaning everyone was free to use, and no copyright would subsist in, any machine created work. This is likely the most practical solution, there a number of potential problems with it. For example, it may operate as a disincentive to the AI industry. While this may be true on some  level, lack of copyright for AI generated works (as is currently the case)  has not hindered development.


Copyright is inextricably linked to the basic human needs and rights it protects. When a work’s creator is not human, and can make no claim to these rights, copyright protection can no longer be justified. With this in mind, we can conclude works generated by DL AIs should not attract copyright, partly because there is no justification for it, and partly because it would be a fallacy to attribute authorship to a human, and doing so would over-incentivise the creation of works with lesser cultural value than those created by humans. Placing machine-created works directly into the public domain  would avoid these problems while still allowing society to benefit from AI-generated works.

Article Link


Get in Touch