18
Examination of witnesses
Paul Fleming, Dan Conway and Dr Andres Guadamuz.
Q13 The Chair: The Communications and Digital Committee is conducting an inquiry into the future of the creative industries. This is our first session after a break for the recess following Her Majesty’s death and the conference recess. There has been a bit of a gap since our first session, so I remind everybody that we are looking at the risks and opportunities of making sure that the UK’s creative sector grows, thrives and does not lose out to any new global competition.
Our session prior to the break was all about the big picture, understanding the economic and civic value of our creative industries and identifying where some of those risks and opportunities are coming from.
Today, we are moving more specifically to the impact of new technologies. We have two sessions. I know that one of them has attracted more attention than the other, but both of them are very important. I am delighted that we have the witnesses here.
The first session is about the opportunities and challenges around AI for performers, writers and visual artists, including intellectual property, the use of image and audio likenesses, and the automation of work.
The second session will look at a particular project, Ai-Da the robot, as a case study for some of the issues that I hope we will be able to explore first with our witnesses.
Please introduce yourselves and tell us the name of the organisation that you represent.
Dr Andres Guadamuz: I am a reader in intellectual property law at the University of Sussex. I am also chief editor of the Journal of World Intellectual Property. I have been studying this for about 10 years.
Paul Fleming: I am the general secretary of Equity, the trade union for performers, directors, designers, dancers, choreographers, children’s entertainers, burlesque artists, comedians and so on.
Dan Conway: I am chief executive officer at the Publishers Association. We are the trade body for publishing in the UK across consumer publishing, education and academic journals.
The Chair: Thank you. I should have said that we are transmitting live on the internet. A recording will be taken, both a written transcript and video, and they will be published on our website in due course.
Q14 The Lord Bishop of Worcester: May I add my own thanks to you for being with us this afternoon? I am delighted to be able to question live human beings and not be confronted with a robot to begin with.
We hear an awful lot about the potential of AI. I was delighted to read that a NESTA report suggests that “The UK is currently one of the world’s leading research centres in AI, has a strong position in AI research and domains potentially relevant for the creative industries such as our work on image text and sound data”. There is also, of course, a lot of hype around AI.
The first question is a rather general one, and I would be grateful if each of you said something about what you see as the possibilities, the potential and the pitfalls for AI in changing the way creative and cultural content is produced in the next five to 10 years.
Dan Conway: It is a real delight to be here, so thank you for having me.
The Publishers Association has been thinking about the opportunities of AI for a few years now. Some two years ago, we produced a report, People Plus Machines, which looked at the opportunities for AI across the publishing value chain, all the way through from content generation and acquisition to product development and sales and marketing.
We did that because of a general perception of the debate in this area, which is, we feel, that often there is AI innovation going on over here and the creative industries over here in that they actually stand in opposition to one another. The same would go for AI and IP—that one stymies the other.
We wanted to produce the report to show that in a sector such as publishing there is already a huge amount of AI innovation ongoing. Our overarching view is that AI, in the creative industries, enhances human creativity and does not replace it.
We have seen a huge amount of AI activation across publishing, throughout the whole value chain. Usually, that is about saving the human creator from the jobs that they do not really want to do. Take an academic researcher, for example, who is trying to make sense of the world’s academic content on a particular issue. It is incredibly helpful for that academic researcher to be able to use a piece of AI to collect all the information to allow that content and creativity generation to take place.
Further down the supply chain in publishing we have publishers using AI to tackle plagiarism—copyright infringement. In education, there are some really interesting things going on with adaptive education technologies. Learners can engage with education platforms, which can teach that specific learner in real time, dependent on where they are in their learning cycle.
Much like other sectors, we use AI in our sales and marketing divisions. It is probably not as headline-catching as the witness who will follow us, but it is things such as stock management, demand forecasting and real-life business things.
Our overarching position is that AI is already employed all the way through the publishing sector and is overarchingly a force for good in the sense that it helps creative businesses to become more efficient and reinforces the human creator rather than replacing it.
Paul Fleming: It is perhaps important, first, to caveat any talk about AI in an historical perspective. The first ever Equity general secretary to give evidence to a parliamentary Select Committee was in 1931. In 1931, item 5 in our annual report talked about the threats and opportunities of recording theatre productions and the rise of Hollywood. There is a little bit of history repeating itself with some of the hyperbole around AI.
As a trade union, we see it dividing very straightforwardly into two areas. There is the growing area of work for members—computer games and the ability of your voice to be used, the synthesisation of your body, your movements and so on—and some of the labour-saving things from jobs that our members do not particularly enjoy doing. You probably do not realise, but an awful lot of commercial radio jingles that you hear now are entirely synthesised and created from a melody of different voices. That work is not particularly creatively fulfilling for members; it is not work that they particularly relish doing. It is, nevertheless, work that they do and it has taken some of the brunt out of that. So an awful lot of extra work is created through AI in both a broad and a quite narrow sense.
The real risks of AI, from our members’ point of view, are in the labour market. I will try to make it really clear that our members are no more affected than somebody who encounters AI because they work on the checkouts at a supermarket. Somebody who works on a checkout at a supermarket has not seen their hours decrease, their pay stay stable or increase because the profits that are potentially garnered are larger. They have seen their hours decrease, their relative pay decrease and their work become more precarious because of the introduction of AI into their working lives.
Radio jingles, for instance, tended to pay broadly quite well; it is quite repetitive work and you have to pay quite a lot to incentivise an artist to do it. It is those relatively higher levels of wages that facilitate our members working in poorer-paid areas of the performing arts, most notably theatre and live performance. People who are walk-on or supporting artists have now essentially been largely erased from big-budget films and are replicated for battle scenes or crowd scenes and completely and utterly cooked up in the background.
Is our members’ pay from other engagers increasing? Is the pay and ongoing usage payments for that work fair, proportionate or in a way that will sustain a creative workforce? No, it is not. That is doubled up with the problem that where we have dealt with AI in very established areas of work—95% of British TV and film is made on a union collective agreement—we have managed “Star Wars”, all that sort of stuff, quite well within that framework. However, if you think about new and emerging areas, we have one agreement for video games. Those areas of new work are more difficult to unionise.
We will talk more later, I suspect, about the overall intellectual property framework, the rights our members have, the moral rights they can depend on—that framework in this country is weak. It is not helpful for our collective bargaining position and it is not helpful in protecting our members from unreasonable exploitation.
There are opportunities of new work, but there are serious risks in the impact of those opportunities and our ability to respond to them.
The Lord Bishop of Worcester: Thank you. That is really helpful.
Dr Andres Guadamuz: I would like first to address the question of hype. There is definitely quite a lot of hype about artificial intelligence. Some of it is justified, some of it is unjustified. Looking at the area, as I have been doing for many years, something that has struck me is that there have been very clear advances, particularly in the last couple of years, in the technology itself.
Some things were not possible seven years ago; I look at the things that I was using seven years ago, but now the capacity of artificial intelligence is at an entirely different level. Even in the last six months, we have seen that things are changing, particularly in the creative industries. Take art. I do not know whether anyone here has played with some of the tools—DALL-E, Stable Diffusion, Midjourney. It is like magic. You type in a few words and you create something that appears like magic. So some of the hype is justified.
Having said that, these revolutions also bring possibilities and challenges. I cannot state this enough; it is a revolution akin to the personal computer and the mobile phone, the smartphone. Once you start seeing it, once you start using it, it is amazing. However, there are challenges—challenges that have been mentioned already. There is quite the potential for artists and creators to be affected by this. There are people who are very angry, very vocal that their livelihoods will be affected by the technology. We have to explore that as well.
The Lord Bishop of Worcester: Thank you. You both outlined the potential and the pitfalls. Mr Conway, you were almost exclusively positive. Do you see negative sides in your area?
Dan Conway: There are some real difficulties about the way the IP framework is working, but I know that we will come on to that.
In terms of labour market substitution, in consumer book, traditional publishing, it is just not something that we are seeing. I can see that that is perhaps a concern in other areas of the creative industries.
Particularly in fiction publishing, the creative process and the audience, the reader’s reception of that creative process is quite complicated. I heard Ai-Da on the “Today” programme this morning, as I am sure all the members of the committee did. The conversation was around, “Is creativity just a process?” The publishing view is, “No, it isn’t”. The reason for that is also about how that process is received.
Part of the beauty of the creative industries is the emotional connection between a human author in my sector and a human reader. I have seen Ai-Da, who is incredibly impressive, and I am looking forward to staying and listening to her. She is able to produce Oscar Wilde’s poetry through being fed Oscar Wilde’s poetry. That is incredibly impressive, but it is not Oscar Wilde’s poetry. If Oscar Wilde’s The Happy Prince was written by a machine, it would not be the tear-jerker that it is.
While I can see that there may be some labour market amendments in relation to AI, probably in my sector in relation to potentially more specialist works with a smaller readership, perhaps away from fiction, we are just not seeing the same level that, with respect, Paul may be fearing in his sector.
Baroness Bull: I have a really short question on tear-jerkers. You say that The Happy Prince could not have been a tear-jerker if it was not done by a human hand. We were hearing that some jingles and jingle-like music, and perhaps some schmaltzy themes, are being written by AI and may well be tear-jerkers if they learn about the little catch moment. Can you think of a tear-jerker piece of music that has been written by a machine? Does this argument go across all art forms or is there something specific about the written word?
Paul Fleming: I have never been brought to tears by a windscreen replacement advert, so I am not sure—prices perhaps, but the adverts, no. The idea that AI would, for instance, read a novel in the place of one of the audio readers of our members and produce the same emotional reaction in an audience at the moment certainly seems untrue. We are, perhaps, naive to think that it would not become untrue.
Ultimately, what is AI? AI is reliant on our own intelligence, and anything we can do a machine can do as long as it advances far enough. It is certainly true that it is not an immediate challenge to reproduce that very intense artistic process. Can the effects of a battle scene or of a video game that you have controlled, which are all real human movements but are responding in very different ways, produce similarly heightened emotion? They probably can.
Baroness Featherstone: Theoretically, could a robot generate a book based on whatever you fed into it, and could that not ultimately replace publishers having to pay authors?
Dr Andres Guadamuz: The answer is yes. It is already happening; it is possible. It is not going to be a tear-jerker, it is not going to be the height of literature, but it is already possible. The technology is already capable of producing passable works. It is producing art that looks very nice. I have made a few myself that people have liked. It is possible to produce music; again, it is not going to be the most amazing.
Baroness Featherstone: That is all cheaper than human beings.
Dr Andres Guadamuz: Exactly.
Baroness Featherstone: What is going to happen?
Dr Andres Guadamuz: Yes, that is potentially the competition aspect; it will produce lots and lots of work that is low-quality, not the most amazing.
Baroness Featherstone: That is the point.
Dr Andres Guadamuz: Yes.
The Chair: Before we move on to the next question, I would like to ask Mr Fleming: in the context of performance, I understand the concerns you have expressed. We will come back to the impact on the labour market. Is there anything positive that you want to say about AI in choreography—anything that makes the production of a show different or improves it in some way that was not there before? I am trying to see whether there is another angle to the positive side of this that we have not heard.
Paul Fleming: I am a trade union bureaucrat and am not particularly well qualified to comment on artistic matters, as my members will tell you.
The Chair: No, but you have members who are.
Paul Fleming: Yes. What we would say is that it opens up a new plain of art. It opens up a new mode of expression. It opens up new opportunities for our members to work, whether it be through a completely new format such as a video game scenario or films that are completely constructed. We will be very familiar with the star vehicle in a production. You have a theatre show or a TV show or a film that is made around an individual. The possibility of reanimating people or putting them in different scenarios does open up a range of different artistic possibilities.
The art question is perhaps a different one for the union. We are not utter Luddites on this; there are definitely new spheres of art, new spheres of expression, new spheres of work that are opened up by AI. That is singularly and universally positive. The real question is securing the best terms and conditions in that exploitation.
The Chair: We will come back to that in due course.
Baroness Bull: Speaking as somebody who has worked quite a lot with choreographers, you can see that there is a difference.
The Chair: I was only using it as an example.
Baroness Bull: Yes, you know exactly that there is a difference between this baseline, standard generation of okay material and true works of art. Great choreographers, like Wayne McGregor, have been using technology in ways that supplement, enhance and create an experience that would not be possible otherwise. It goes back to the point that, yes, you can churn stuff out from a machine or you can use the machine to create something that is possible only with the machine and the human partnership.
The Chair: We are moving on to intellectual property.
Q15 Baroness Bull: This question is about the challenges that AI presents to the creative industries in terms of intellectual property.
I understand that, quite unusually in the UK, copyright law protects works that are generated not only by individuals, including individuals partnering with AI, but by a computer where there is no human involved. I understand that that is quite unusual. I am interested in that argument because, as I understand it, the legal definition of “originality”, which has to be met before copyright law kicks in, is based on human characteristics, and no robot can have those.
There are differing views on whether the law is still fit for purpose, whether the moment is now to update or whether it is too early. I am keen to hear from each of you. I will start with our academic, if I may. What are your views on the main challenges for the creative industries in terms of intellectual property?
Dr Andres Guadamuz: Yes, thank you very much. I would like to start by making the distinction that usually when we talk about artificial intelligence and copyright we are talking about both authorship and liability. As you mentioned already, there is the question of authorship but, very importantly, we are looking at potentially changing the law with liability.
Let me start with authorship. A few other countries have copied this, but the UK is unique in that it was the first country that passed Section 9(3) of the Copyright, Designs and Patents Act. It says that the author of a computer-generated work will be given copyright. Whoever made the arrangements necessary for the work to come into being will be the author.
This is unique, because not many countries have this. Some people argue that this clashes with the intellectual creation requirement of originality in copyright. We have this potential dichotomy that we have protection for computer-generated works and a requirement of intellectual creation.
At the moment, we do not have case law; this has not been decided on any case. Most importantly, this provision was generated before artificial intelligence. People were thinking initially more about robotic painters, as you will see later; that was the origin of the provision.
There was a consultation last year—it concluded this year—by the UK IP office, which asked three questions: “Should the law be changed?”, “Should this computer-generated provision be repealed?”, or, “Should we have a new sui generis rate?” Many other academics and I chose to go for option zero, which was keep the law as it is, and that is what the UK IP office decided.
Apologies, but I am going to go on a little bit. Right now, that is the solution, and we will wait and see what happens. The provision is there, it is unique and only a very few countries have this provision.
Moving to the liability part, could a robot, could artificial intelligence, infringe copyright? That is potentially a very important question. The UK is also a leader in this area. As far as I know, the UK was the first country to pass an exception for copyright for text and data mining, back in 2014. There were a bunch of exceptions passed at the time: parody, pastiche and other exceptions. One of them was for text and data mining for research purposes.
That is the law as it is, and in the consultation that was taking place—the same consultation I mentioned—the same question was asked: should we keep that as it is written or expand it? I am sorry that I am going on quite a lot, but it is a complicated area.
In 2019, the European Union passed the digital single market directive, which contains two exceptions to text and data mining: the training of artificial intelligence in Article 3, which is similar to what we have in the UK. In Article 4, it has an exception for text and data mining but for commercial purposes. The UK exception is for research only, and its exception is now for commercial use but as long as the author, the owner, does not reserve the right; it is an opt-out system. The UK IPO consultation came up with the solution that the UK should implement something that goes beyond that: an exception to copyright for commercial purposes, for all commercial purposes without an opt-out. That was the consultation that was presented in June 2022.
I am sorry that was a very long answer.
Baroness Bull: Will you explain the implication of having the commercial opt-out for creators—for publishers, for instance?
Dr Andres Guadamuz: It will allow someone—let us say an artist—who does not want their art to be included in a dataset that will be used to train, to be in a machine learning program. In Europe, with that opt-out, an artist can say, “I don’t want my art to be in this dataset”.
The implication of that opt-out is that it allows—I think it is a good compromise—owners and copyright owners, copyright authors, to have the possibility of saying, “I don’t want to participate in this new future of machines and machine learning”. What will happen? I do not know; I do not think anybody knows. It has not been implemented much in Europe. It was in 2019, and with the pandemic countries are just starting to implement this, so we really do not know.
The UK IPO’s consultation states that it wants the UK to become a leader— a leader specifically to have companies come and establish artificial intelligence, data-mining operations here.
Baroness Bull: Thank you. That supplementary was very helpful. Paul, from the perspective of your members—
The Chair: Before Baroness Bull continues, are you, Lord Hall, looking for a point of clarification?
Lord Hall of Birkenhead: Yes. When you say “that work” in “that dataset”, it would be very helpful if you explained exactly what “in that dataset” means, both commercially versus for research purposes.
Dr Andres Guadamuz: To give an example, in the very large dataset of art now being used for training, as in tools such as DALL-E, there is a very popular dataset called LAION. That contains 4.5 billion images: photographs, art and so on. The important thing is that it contains images and description. The machine knows, when you are asking for “Paddington Bear painted by Van Gogh”, what Paddington Bear looks like and what Van Gogh paintings look like, because the dataset contains both.
From the perspective of someone putting together all that data, this could infringe copyright. There is a copy in the dataset, so the dataset is public; you can download it and it is available. There are people who have created browsers where you can see whether an image of yours is there. I looked myself and three photographs of me are in the dataset. That is what it means. It is a large collection of works, a large collection of pictures, images, that has metadata with it. From a corporate perspective, it could be infringing copyright—that is what we mean, including making a copy into a dataset. An exception, particularly for research purposes, is that that dataset is not infringing copyright—putting all the images in the dataset—if you are doing it for research.
Lord Hall of Birkenhead: The end usage is key.
Dr Andres Guadamuz: Exactly. End usage is absolutely key.
Lord Hall of Birkenhead: Thank you.
Baroness Bull: Back to you, Paul. I am keen to understand from your members’ perspective the intellectual property challenges that AI presents.
Paul Fleming: We would say that there are three principal deficiencies in the legislative framework as it is. I will go with them in ascending order.
The first is image rights and the lack of clarity around image rights in this country. American states have been dealing with this for some time. California, as you might expect, had legislation brought in in the mid-1980s to try to deal with it, and now there is a range of case law to support the rights of deceased artists, to control their image, their work and so on. That is critical to starting to deal with this synthesisation phenomenon, the deepfake phenomenon, the use of very identifiable images, movements, the speech patterns of people to promote products they are not in favour of, to promote political causes they are not in favour of, to create videos that they do not want to be created, and so on.
That is issue one, where there is a deficiency in the legislative framework from our members’ perspective. The second is the data-mining issue, which has been touched on far more expertly than I can. It is pretty clear what the worry from our members’ perspective is. If you look at where AI is directly impacting terms and conditions right now, it is the creation of synthesised adverts, commercials, movements. Why do you need to engage several artists to put together all the movements that go into a video game if you can wantonly data mine?
The opting out of it is highly complex, particularly for an individual. It is one thing if you are a corporation with intellectual property that wishes to remove it or a big photojournalism firm, but if you are an individual performer trying to pick that out, that is quite difficult.
That gap in the legislation rests on what is probably the biggest problem: the protection of audio but not the protection of audio-visual performances with the same sorts of copyright as other areas of work. That comes from a long-standing tension—is it the director’s copyright? Is it the artist’s copyright? Is it the choreographer’s copyright, and so on? You can understand where that tension comes from.
Fundamentally, there is no copyright protection for a performer’s performance. Laurence Olivier’s “Hamlet” is not protected, not really. That is a very serious problem as you start to move into a world where AI gets better, and TV and film get better.
We are not in favour of an overregulated system with statutory remuneration for our members. We have very strong, collectively bargained agreements across the audio-visual sector that protect our members. Frankly, they are more flexible and responsive than a legislative approach; they deliver better returns for members. Our members receive significantly better royalties and secondary payments than their colleagues in continental Europe, because we bargain them better. That has allowed us to be flexible as new technology has arisen.
This is not a question where we are asking for audio-visual performances to be protected to create a statutory framework of royalties, which is often the ambition elsewhere in the world. What we would like is an acknowledgement of it in law. That is what the Government of this country have signed up to in signing the Beijing treaty. They have said that they will do it, but they have not ratified that treaty.
There is this enormous gap for audio-visual performance specifically, which are, of course, some of the biggest exporters of this country and whose work is among some of the biggest exports of this country, whose work is essentially unprotected in the face of growing AI. That is the biggest overarching IP issue: allowing us to have the framework to collectively bargain with it to deliver the answers that our members need to protect them in the future.
Baroness Bull: That is very helpful. It seems as though there is a potential solution to the last one. On the first two, is it moving so quickly that freezing legislation in time right now would be a bit meaningless because we do not know how this will develop?
Paul Fleming: It is absolutely a risk that you overengineer these things. By the time you have finished consulting on them, they are broadly irrelevant. We are very lucky in that we can look at what is a highly unionised sector. What we are looking for, essentially, is the framework within which we can bargain to ensure that individual contracts and agreements have a legal underpinning. We know that some of these image rights will go on and on. What we would like to do is negotiate ourselves, through our collective agreements, appropriate payments.
We see that already in film. When they are re-animating big actors in “Star Wars”, they are on our collective agreements—on British collective agreements. We have delivered the proper outcomes and those people can protect their intellectual property. That is fine, because it is within that framework. One starts to get worried in the world of computer games and elsewhere.
There is the question simply of the deepfake issue. The deepfake issue is very different from an artistic and creative issue, but it impacts our members very directly because of the nature of the work they do. It is very important to separate those issues from synthesised performance and deepfakes; they are very different issues. The framework that is required is a much stronger moral intervention, frankly, from the state to protect people’s rights to not be degraded.
Baroness Bull: If I have time, I would like to go to Dan with the same question from your perspective, please.
Dan Conway: Thank you. Taking one slight step back, it is often said of the UK legislative framework that we have the global gold standard for intellectual property and copyright. That is how we feel and that is how the rest of the creative industries feel about the legal backdrop.
Amazing numbers were quoted in the previous session that you ran as part of this inquiry about the creative industries being worth over £100 billion to the economy. The reason for that is because we have an appropriate legal framework that allows a market for creativity, allows a market for ideas and allows creative businesses to grow and thrive.
There are exceptions to copyright in UK law—rightly so—to balance the interests of the user with the interests of the rights holder involved. Our view on the AI text and data-mining issue is that where the Government have arrived at is a bit of a sledgehammer to crack a nut. It is a really broad new exception without any limitations, which potentially falls foul of our international IP obligations in relation to the Berlin convention. We would like the Government to look at it again.
In terms of what it means practically for publishing, at the moment, as Dr Guadamuz was saying, there is a research exception in UK law. If you are conducting research, you can, for free, access all the datasets you want and produce your research. If you are going to commercialise that research, you have to pay a licensing fee to do so. In a market, that feels fair, as long as that licensing agreement process is working.
In response to the consultation that Dr Guadamuz was talking us all through earlier, we alighted not on saying the status quo; we said option one, which was, “Let’s have a look at the licensing regime. Make sure that, if an AI product is looking to mine a load of data for commercial purposes, those licensing agreements work”. Obviously, a huge number of different companies are looking to do that—everything from large US tech firms on the west coast of the US, through to small and micro businesses all over the world. At the moment, the legal provision would allow any of those businesses of any size, located anywhere in the world, to access all my members’ data for free for the purposes of text and data mining. There is no differentiation between a large US tech firm in the US and an AI micro start-up in the north of England.
We have been talking to the IPO about the licensing regime and working on a more bespoke legal response that avoids a brand-new, wide exception and focuses on this being a market that works. We have put a number on it for our sector. Currently, TDM licensing, we think, is worth £335 million a year to publishers and is a nascent area of investment. Without a legal provision that undermines it commercially, that would grow over time.
Baroness Rebuck: That was really interesting, because arguably the publishing industry is a growth industry, post pandemic. You have made a very eloquent case for looking at the data-mining issue in a more holistic way. This campaign comes hot on the heels of, say, international exhaustion. In one of our earlier inquiries we looked at legislating the legislators. You referenced the fact that government is always playing catch-up to the speed of transformation of industry. From a general approach point of view, do you have any thoughts on how you could work more collaboratively so that you could avoid launching great campaigns every five minutes and look at some of the excellent use of some our legacy creative industries in using new technologies in a very positive and fair way, so protecting the past but also adjusting for the future in a slightly more subtle way?
Dan Conway: Thank you. To us, we are very important. We are a £6.7 billion industry and hugely culturally important. I am also aware that, in the grand scheme of things, government is a big beast and it is trying to legislate for a huge number of sectors. That is partially why this collective moniker of the creative industries is so useful: because it allows industries such as mine and the others to club together and carry some real heft.
It has been a shame for us that we have faced these intellectual property issues over the past few years. The threat of international exhaustion has been a big one, which has not quite settled yet, and there is this text and data-mining issue as well.
I would love there to be a forum for co-ordination and co-ordinated thinking on specifically intellectual property and the creative industries. One of the ways in which we would like to see this moving forward is a forum that perhaps government could convene and would bring together academics, industry, our industry and technology firms around the creative industries to look at it as a bit of a test case for how you bring AI forward. A bit of co-ordination, discussion and planning, perhaps convened by policymakers, could only be a good thing.
Q16 Lord Foster of Bath: I thank all three of you. It has been absolutely fascinating so far. I was particularly struck by Dan’s comment that AI enhanced rather than replaced human creativity. He went on to say that it was a force for good. I think we would all agree that if we are to maximise its potential, we need to have the right legislative framework to minimise the possible downsides and challenges.
In relation to those challenges, we have already discussed IP; others have been touched on around workers’ rights, such as contracts for artists engaging in AI work or labour market substitution. We are already well aware that there are a number of others, such as data bias and so on.
I wonder whether each of you could talk about what you see as the broader challenges and how best we go about addressing them. Perhaps I can start with Paul, not least because you said very clearly you were not in favour of overregulation or overengineering, so we have to find ways of doing it that do not do that either.
Paul Fleming: I go back to the core of where we are. The creative industries, as Dan said, are worth more to the British economy than banking, but it is an incredibly low-paid sector. We have 47,000 members, about 10% of whom are earning in the very top brackets that people associate with celebrities or performers. We reckon that around 80% to 90% of our members are earning less than £20,000 a year from the industry itself. That is incredibly low pay.
Fundamentally, there is a growing problem where the areas of work that sustain them through periods of low pay are about to be removed entirely from the market because of the speed of AI intervention. That is not necessarily a bad thing, because it is opening up other opportunities for employment, but what needs to be created from our members’ point of view and from the point of view of audio-visual performers, who are in a very different position from that of other artists, is a legislative starting point to ensure that we can go directly to emerging producers and creators of work, such as video game producers and so on, and say, “There’s a statutory obligation on you to provide some form of remuneration going forward to acknowledge the moral rights that our members have in their performance. Therefore, come together and negotiate it with us”, because the threat of legislation and that straitjacket, as happens in most of continental Europe, instead of a collectively bargained solution to this, which is what works in the vast majority of the areas in which our members work, will not happen without that as a bit of a kick to the producers and creators of content who are not used to operating in a unionised environment.
In terms of the broad challenge, you risk removing large swathes of people from the workforce. After the pandemic, 90% of our black women members are talking about exiting the industry. This is a group in growing cultural demand for their input into products, but they are talking about leaving the industry.
The figures for our deaf and disabled members are almost as high: 56% have considered leaving the industry. Those are people for whom AI should be facilitating more access to work; it should be making work easier for them to access.
The risk is that, as pay stays low, new opportunities for pay are not supported with the legislative framework, and people who struggle to find work and be represented within the industry when times are good end up being forced out when times are hard.
Of course, the trade union official will say that the biggest issue is people’s terms and conditions moving forward and their ability to bargain them. I think that is overwhelming.
Lord Foster of Bath: You make that very clear and it is helpful, but do you have wider concerns that perhaps we have not touched on yet and you think we should be thinking about?
Paul Fleming: I do not, because I do not think it is too complicated an issue, even though sometimes it seems somewhat overwhelming. There are intellectual property issues; there are issues about people’s rights when they are at work in the first instance; there are the creative issues and the intersection with intellectual property that need to be considered. It is quite easy to distil this into some very straightforward questions that legislators need to answer instead of getting slightly distracted by how overwhelming the whole thing is. All we need is a simple, straightforward framework, and I think the questions being asked are the right ones.
Lord Foster of Bath: It would be very helpful if you jotted down those questions and sent them to us to have a look at.
Dr Andres Guadamuz: Besides all the things that we have been discussing about intellectual property, leaving my field a little—because I have been watching what has been happening—to me it is interesting how fast things move. I do not know whether you remember when artificial intelligence translation started. People would laugh at it. If you tried to translate something, it would translate it very badly—until it did not. It completely changed. Now it produces text that is not only readable but which I can use in many circumstances. This is replacing translators. It is a profession that is almost dying out because artificial intelligence has an effect.
Things we do not even know are potentially changing right now. There are other industries. I was reading today that students are using artificial intelligence to write their homework. One student got artificial intelligence to write an essay justifying the use of artificial intelligence to do homework. It was quite passable.
These are the things that we have to grapple with. I think that in general society has not realised that changes have already happened, are happening and will happen. I do not know what the creative industries will look like or what will happen in academia. Students can write essays by just putting in a few instructions. It is already happening with art; it is potentially happening in music, video and all sorts of things.
I do not know what will happen, but as a society we need to think about these things. We will have almost unlimited content that will compete with human content. It will not be very good, but it will be free, potentially. I do not know what the answer is.
Lord Foster of Bath: In summary, the challenge is that we do not know what the challenges will be.
Dr Andres Guadamuz: Yes.
Lord Foster of Bath: We also have to think about how we prepare ourselves for that.
Dr Andres Guadamuz: Yes. As we say, there are known unknowns.
Dan Conway: Without wanting to repeat what I said earlier, I do not think that in the publishing sector we are expecting the kind of labour market difficulties that Paul is perhaps talking about in his own sector. I was interested in what Dr Guadamuz said about unlimited content competing with human content, but it being passable or not very good.
To go back to what I said at the beginning, is that not the point here? We are dealing with an information boom by human creators but also by AI creators. To go back to the report we produced, some of these technologies will be really helpful in making sense of a lot of that and supporting human creativity.
To return to what I said before, our perspective is that AI can be a force for good for human creativity within that framework.
Lord Foster of Bath: The one thing none of you has touched upon is real concerns around data bias and the way in which AI then uses the information to create new stuff, so we end up getting more and more of exactly the same stuff. In that sense, you could argue that we will miss out on that creativity issue, and we are interested in the creative industries. How do we address that challenge? It must be just as true about books as it is of plays, television, theatre or whatever.
Dan Conway: It is a really important point. We do not want to have just an AI boom; we want AI outputs that are accurate, that work and are usable. To go back to the issue of TDM, the real problem for us as a sector is that, if you disintermediate the people who are responsible for making that data accurate and ruin the investment case for building datasets that are correct and balanced, effectively you are undermining the investment case in the underlying data. You might be able to expand access to all types of commercial AI software that want to use it, but you are undermining the fundamental investment case and the data that underpins that. Therefore, the outputs that you get, as you rightly point out, will not be the right ones. That is why we need the right legal framework to underpin this.
Paul Fleming: To some extent, capitalism itself tends to overproduction. The consequence of an AI boom in terms of its impact on creativity is probably the same as it would be anyway. As legislators look at this, they have to consider it in the broader context of whether we are having the right state intervention in the creative industries. Are we adequately funding our theatres? Are we adequately funding the BBC? Is the structure right for Channel 4? Are we having these structural interventions in these industries? What is AI doing? It is responding to markets; it is producing content for markets and consumers that already exist.
The real question for the creative industries is: do we have a sufficient framework that adequately subsidises human content in order to intervene in that market?
You are quite right in that one of the bigger worries from our point of view is the impact it has on structural bias. I talked about the opportunities that AI should be bringing for our deaf and disabled members, the impact particularly on black creators and so on of what AI reinforces, and the already structural problem with accessing the market, the sort of content that is produced, what we believe markets are saying and what content we believe people want to consume.
That is probably a bigger short-term issue, because the impact is to limit opportunities for the workforce in the short term. I think the issue of overproduction can be dealt with by adequate state intervention in the arts and creative industries to make sure that they are stimulating the right sort of content.
As for the workforce issue, you are absolutely right to ask, because at the minute our membership is so precarious that AI intervention makes it harder for underrepresented groups to access the workplace. That will be a bigger problem.
Dr Andres Guadamuz: This is not my area of study. There are people who specialise in bias and particularly the ethical aspects of AI, so I cannot really comment on that. It is a very interesting study. People are working on this. There are specialists working on the ethics of AI. I can only say that the biases are there. Just yesterday someone using DALL-E asked it to create an image by Diego de Guevara of rich Mexicans and an image by Diego de Guevara of poor Mexicans. The poor Mexicans had skins that were considerably darker than those of rich Mexicans. That bias is embedded in all this technology. How to tackle that I leave to scholars and colleagues who are working in that area, but the bias is there.
Q17 Baroness Harding of Winscombe: Thank you all so much. This is fascinating.
Mr Fleming, it is interesting that you started with the 1930s. I want to take you all the way back to 1436 and the invention of the printing press. What we have been talking about today are the challenges of new technology. History shows us that in the case of the printing press, probably going even further back than that, you cannot stop new technology coming, but what you can do is choose to embrace it or not. You can also choose whether or not to mitigate the harms.
We have spent most of the session so far talking about how to mitigate the harms. I would like to take you to what we should be doing to embrace the opportunity. If you look at the history of the printing press, the single biggest issue in the 15th century is that the countries that chose to embrace it drove literacy and growth for the next 300 years and the ones that chose not to embrace it did not, regardless of how much mitigation there was of the harms brought by the printing press. What are the opportunities that AI could bring the UK, and how would you assess how well we are doing in embracing them?
Paul Fleming: I would not want to talk about the printing press but probably the cries of the Luddites, who have had a rather bad rap as far as I am concerned. They were concerned about the deceitful use of new technology. That was their position. The Luddites saw machinery coming in to undermine standardised terms and conditions and pay.
I think the real question about whether AI is a success for this country is: does the creative workforce become more diverse, and does it get to share more equitably in the fruits it creates? If what we have is a shrinking creative workforce, or certainly a less diverse creative workforce, and we have a creative workforce that remains as precarious as, or becomes more precarious than, it is now, the AI intervention has failed. When we talk about the printing press, we talk about increasing literacy and social benefits and the consequent economic benefits of that.
If there is no direct benefit to the workforce that is creating this stuff, we need to take a serious look at it. That is not the experience at the minute. The experience at the minute is that there is an awful lot of fear. It is not just reactionary. This is not a question about how much is paid per hour per job; it is a question about whether we continue to be a world-leading centre for dubbing or video game creation, which we are. That is driven by AI, but also by the cultural environment in this country. Can that be sustained and grown? Does that result in the overall share of the workforce in the fruits of the labour it creates? That is very much the test. The opportunities are absolutely there.
Baroness Harding of Winscombe: You have made the point extremely eloquently about the impact on your members, but what else should we be doing, if anything, to bring about the opportunities that AI presents? Will that just happen automatically without the need for any government intervention, or any R&D spend or incentives? Are you happy that we will be up there with the best in the world on AI if we do nothing else? How do you assess government’s role and the role of UK commercial sectors in bringing the opportunities?
Paul Fleming: I would worry that if we do not have a fit-for-purpose moral framework for audio-visual performances in this country, which is being developed relatively successfully in the United States—it certainly has a bedrock in Europe—you will see production move elsewhere. That is not being driven by bosses or producers; it will be driven by core talent, which will say, “I do not want my image put at risk within this weak framework”. That is very straightforward: ratifying the Beijing treaty, creating that moral right impetus for audio-visual performances and then facilitating the industry to bargain itself. Our union and others bargaining with producers is absolutely the way forward. It does not require legislative intervention in the same way.
That is a very real worry. If you have this huge data-mining operation that potentially undermines areas of work for our members, it will also undermine the ability of others to create. Those are the areas to be really worried about. It is not so much overengineering that will be the problem in the current trajectory; it will be a lack of protection, but there is a balance to be struck.
Dan Conway: We are really positive about AI. You are entirely right. AI has arrived and there is no point in trying to turn back the tide. Publishing is an example of one of the creative industries that has embraced AI technology across all its value chain, particularly in the world of academic publishing where we are talking about really big data companies employing AI to use recommendation engines to support research all around the world. There is some really well-developed thinking in that area. We are an example of a rights holder industry that hopefully challenges the assumption that there is AI innovation and technological innovation happening over here and creative industries over here and never the two shall meet. That is just not the case. Our members are investing in and collaborating with tech companies, and some really interesting things are happening.
On your question about what government should do and what we think should happen, there are three things. First, we would like legal certainty on the text and data-mining issue. We have covered that extensively, but we respectfully feel that the Government have arrived at the wrong conclusion. There should be a rethink about how we articulate the legal framework around text and data mining.
There is also the skills issue, which was covered more extensively in your previous session, but when I talk to my publisher members about where it is hard to recruit skilled data scientists, AI skills are a real area of weakness. I know that the Office for Artificial Intelligence is looking at working with higher and further education colleges to make sure that the UK education system is designed in a way that builds up the skills needed in this area, but obviously there is a time lag to that. We would be very supportive of an enhanced skills package to bring forward skills in AI. We certainly see that as an area where it is difficult to recruit.
The third one would be increased partnership in collaboration between academia and the creative industries, tech and government. The committee is running a really interesting inquiry in relation to the creative industries and AI. There could be an ongoing forum that looked at how this works in a more holistic way rather than the piecemeal conversations that we feel we are having at the moment.
Dr Andres Guadamuz: Even without any changes in the law, as things stand the UK is already a leader in artificial intelligence. Companies are already being created in the UK. One of the big players in artificial intelligence is Stability AI, based in the UK. Having changed nothing in the law, it is a start-up based here that already has a $1 billion valuation. So the UK is already a leader.
In order to attract potentially more companies and become a greater leader, the text and data-mining exception should be changed at least to meet the standard of the European Union; otherwise, some companies may just go to France, Amsterdam or wherever, because they have a better exception for text and data mining. At least that should be met, and that is probably what the Government can be doing.
I think the UK IP office has been doing a fantastic job with the consultations and leadership it is showing in doing research in this area, and long may it continue. The Government have to continue doing research. I also agree that we need more research on the effects of this on the creative industries.
The Chair: Thank you very much to all three of our witnesses. This has been a really helpful session. I am very grateful for your openness and willingness in answering our questions so clearly. Mr Fleming, if you are able to follow up the questions in response to a request from Lord Foster, that would be brilliant. Mr Conway, I reassure you that skills is an area that we will look at in greater detail in the second part of this inquiry.
