In a recent chapter, Ryan Abbott and Elizabeth Rothman present the utilitarian argument for granting copyright in AI-generated works (hereafter AIGW). Aspects of their argument also find expression in the recently launched UK Intellectual Property Office (UKIPO) consultation on AI. In response, this post outlines my scepticism. The utilitarian arguments supporting copyright in AIGW are empirically speculative and theoretically dubious. Our society’s welfare will probably be better served by leaving AIGW in the public domain.
Utilitarianism and economics of copyright: incentives are only half the battle.
Classical utilitarians argued that the state should act in a way that maximises society’s utility, summed up in the phrase ‘the greatest good for the greatest number’. Modern welfare economics makes the same argument. There are a few differences between the two – particularly regarding the definition of ‘utility’ (happiness versus the satisfaction of subjective preferences) but these should not detain us too long here. For more detail on the relationship see here (p 18). Assuming we agree with that political philosophy, should the state grant private property rights in expressive works?
As an initial matter, granting copyright might seem curious from a utilitarian perspective. Expressive works are beautiful things because of one wonderful characteristic: they are non-rivalrous, i.e. use by one person does not diminish the ability of others to use them. Most things in life can only be used a finite number of times and therefore need to be carefully managed and put to the best possible use to prevent waste. But the great thing about a work like the new Wicked movie is that you and I can watch it a hundred times and it never diminishes. So, what is the point in locking it up behind legal chains? If you and I get utility from watching the Wicked movie, how can society’s utility be increased by restricting our ability to do so? Of course, this is precisely what copyright does: it grants the owner the ability to restrict access to those willing and able to pay a fee. This is, in a nutshell, what economists mean when they say copyright causes ‘deadweight loss’, i.e. the lost utility of those people who cannot pay for access. And that is to say nothing of the increased transaction costs, enforcement costs, and administrative costs associated with copyright as well as its ability to cause other distortions in the economy.
But copyright has beneficial effects too. In addition to being ‘non-rivalrous’, expressive works have two further characteristics: they are ‘non-excludable’ and have high fixed costs of production (relative to marginal costs). Such works are non-excludable in the sense that producers cannot, absent copyright, restrict access to paying consumers. This causes a problem for the producer. Creating the first copy of an original work is often a costly endeavour. The budget for Wicked was $145 million. If Universal Studios cannot recover that cost, it is unlikely to create the work in the first place. And so we have a market failure. There are empirical questions about how widespread that market failure really is (i.e. how many welfare-enhancing works would fail to be created under a free market?), and there are also a range of policy tools to address such a market failure of which copyright is only one. Nevertheless copyright is the conventional tool to solve the non-excludability problem. By restricting access to the work, the law gives the owner an ability to limit access to paying consumers, thus generating the profits necessary to ensure continued supply for works. And so copyright has the dual effect of limiting access (bad for utility) and encouraging creativity (good for utility).
That said, the utilitarian (or welfare economic) argument for copyright is not simply ‘copyright encourages creativity’ but is far more nuanced. Properly stated, the argument pro-copyright utilitarians make is that the state should grant copyright if – and only if – the benefits of copyright outweigh the costs and, furthermore, that the cost-benefit ratio is more favourable than other alternative policy tools. Of course, the argument is challenging to test empirically. But for the moment, let’s just assume there is a good utilitarian argument for copyright in non-AI generated works. For more detail see here (pp 1848-58) and here (pp 728-735). For an introductory level discussion on all of this see here (pp 1-6 and pp 32-35).
The putative utilitarian argument for copyright in AI-generated works.
Granting copyright in AIGW has all the negative costs associated with copyright in regular works. If an AIGW is subject to copyright, then use of a non-rivalrous resource is restricted causing deadweight loss. Consider, for example, the following AI-generated image of Pope Francis that went viral in 2023. It’s a fun picture! You and I are currently gaining some utility benefit by viewing it and enjoying it. However, were this image subject to copyright, we would likely forgo this benefit. I am a poor academic after all. It’s unlikely that I would stump up the money required for a license fee to use the work. And, even if I were able to do so, I doubt I would be willing to go through the rigamarole of locating the copyright owner and transacting over a license.
With that in mind, what are the supposed benefits of granting copyright to AIGW and do they outweigh the costs? Abbott and Rothman suggest several benefits, some of which are found in the UKIPO consultation. For space reasons, I will consider here the two most plausible and most widely discussed putative benefits, namely:
- granting copyright will encourage people to use creative AI to generate and disseminate socially valuable works.
- granting copyright will encourage people to develop generative AI technologies.
Encouraging use of AI
If copyright is granted therein, can we reasonably expect meaningfully or significantly more generation and dissemination of AIGW than we would under the free market? I am doubtful. Of course, I don’t doubt that copyright will have some positive incentive benefits for some people in some situations. I do doubt, however, that such a policy intervention would cause a significant net increase in creativity.
Why? Well, because AIGW production is just so easy. Consider: I have never painted anything in my life but I have already produced hundreds of images with Midjourney. This is what Dan Burk referred to as ‘cheap creativity’. Because the costs of generating such material are falling, then one of the chief barriers to creativity (i.e. high fixed costs of production which need to be recovered somehow) is increasingly disappearing. As production costs decrease, one can predict that producers’ individual cost-benefit calculations will change, making it possible for more people to create in the absence of copyright. And nor is distribution of that material particularly problematic. Distribution is practically costless in the Internet era.
Overall, the opposite problem seems far more likely: that even in the absence of copyright, so much AIGW will be created as to put human creators out of business.
Encouraging development of AI
So what then of the second argument? The UK IPO consultation suggests that the costs associated with AIGW copyright could be justified if such copyright ‘is likely to encourage development or investment in generative AI services’. Similarly, Abbott and Rothman point to the costs involved in ‘development of creative AI like Dall E 2’ and the subsequent improvement thereof.
However, we have a much better tool available for encouraging technological development: the patent system. As an intuitive illustration consider: do we grant copyright in literary material to encourage investment in typewriters or ballpoint pens? Or do we grant Microsoft copyright material produced in Word to encourage development of word processing technologies? No. Instead, if someone makes a novel form of typewriter or pen, they are eligible for a 20-year patent monopoly. Underpinning this basic intuition are three important points:
- First, regardless of what copyright does in relation to AIGW, the rate of technological development will always be determined by the strength of innovation policy interventions, the chief one being the patent system. If Google, for example, cannot prevent OpenAI using their technology, then Google faces the usual disincentive to invest in that technology. Even if Google is granted a patent allowing them to restrict competition, the incentives for technological development in said technology will end when the patent ends. Copyright in the outputs cannot change either outcome. It is the non-excludability of technological inventions (not creative works) which is the problem that needs to be addressed to ensure continued technological progress.
- Second, to the extent that copyright might increase technological development incentives, then this outcome is duplicative and wasteful. We don’t need two costly policy tools that aim to achieve the same outcome (increased technological development) in the same way (through a private property right).
- Third, it is likely that the two policy tools will interact unpredictably and negatively. Both technological invention and creativity are highly complex systems. It’s hard enough already trying to fine tune the patent system to optimise our rate of technological progress. I doubt that retrofitting copyright to that goal will make it easier for us to achieve that goal.
Overall: do the benefits outweigh the costs?
Ah, the million-dollar question. And in truth, I don’t know the answer. We rarely do when it comes to the consequences of IP protection. But if I were a betting man, I would bet ‘no’. There are concrete negative consequences associated with copyright protection. Meanwhile the proclaimed benefits to AIGW are highly speculative at best. And so I think we are probably better off leaving those works in the public domain.
This is not to say I disagree with Abbott and Rothman or the UKIPO on everything. The UKIPO is clearly on the right track in this consultation. And Abbott and Rothman are also entirely correct when they say copyright is not a good policy tool for fixing the very real problem of automation-caused unemployment for human creators. It just so happens that copyright is also not a particularly good policy tool for encouraging technological development, nor a necessary one for encouraging AIGW.
