27: Schedule 1, page 185, line 11, leave out from “provider” to end of line 13 and insert “, including where the publication of the content is effected or controlled by means of—
(a) software or an automated tool or algorithm applied by the provider or by a person acting on behalf of the provider, or(b) an automated tool or algorithm made available on the service by the provider or by a person acting on behalf of the provider.”Member’s explanatory statement
This amendment is about what counts as “provider content” for the purposes of the exemption in paragraph 4 of Schedule 1 of the Bill (which provides that limited functionality services are exempt). Words are added to expressly cover the case where an automated tool or algorithm is made available on the service by a provider, such as a generative AI bot.
My Lords, the Government are committed to protecting children against accessing pornography online. As technology evolves, it is important that the regulatory framework introduced by the Bill keeps pace with emerging risks to children and exposure to pornography in new forms, such as generative artificial intelligence.
Part 5 of the Bill has been designed to be future-proof, and we assess that it would already capture AI-generated pornography. Our Amendments 206 and 209 will put beyond doubt that content is “provider pornographic content” where it is published or displayed on a Part 5 service by means of an automated tool or algorithm, such as a generative AI bot, made available on the service by a provider. Amendments 285 and 293 make clear that the definition of an automated tool includes a bot. Amendment 276 clarifies the definition of a provider of a Part 5 service, to make clear that a person who controls an AI bot that generates pornography can be regarded as the provider of a service.
Overall, our amendments provide important certainty for users, providers and Ofcom on the services and content in scope of the Part 5 duties. This will ensure that the new, robust duties for Part 5 providers to use age verification or age estimation to prevent children accessing provider pornographic content will also extend to AI-generated pornography. I beg to move.
My Lords, the noble Baroness, Lady Kidron, has unfortunately been briefly detained. If you are surprised to see me standing up, it is because I am picking up for her. I start by welcoming these amendments. I am grateful for the reaction to the thought-provoking debate that we had in Committee. I would like to ask a couple of questions just to probe the impact around the edges.
Amendment 27 looks as if it implies that purely content-generating machine-learning or AI bots could be excluded from the scope of the Bill, rather than included, which is the opposite of what we were hoping to achieve. That may be us failing to understand the detail of this large body of different amendments, but I would welcome my noble friend the Minister’s response to make sure that in Amendment 27 we are not excluding harm that could be generated by some form of AI or machine-learning instrument.
Maybe I can give my noble friend the Minister an example of what we are worried about. This is a recent scenario that noble Lords may have seen in the news, of a 15 year-old who asked, “How do I have sex with a 30 year-old?”. The answer was given in forensic detail, with no reference to the fact that it would in fact be statutory rape. Would the regulated service, or the owner of the regulated service that generated that answer, be included or excluded as a result of Amendment 27? That may be my misunderstanding.
This group is on AI-generated pornography. My friend, the noble Baroness, Lady Kidron, and I are both very concerned that it is not just about pornography, and that we should make sure that AI is included in the Bill. Specifically, many of us with teenage children will now be learning how to navigate the Snap AI bot. Would harm generated by that bot be captured in these amendments, or is it only content that is entirely pornographic? I hope that my noble friend the Minister can clarify both those points, then we will be able to support all these amendments.
My Lords, I rise briefly to welcome the fact that there is a series of amendments here where “bot” is replaced by
“bot or other automated tool”.
I point out that there is often a lot of confusion about what a bot is or is not. It is something that was largely coined in the context of a particular service—Twitter—where we understand that there are Twitter bots: accounts that have been created to pump out lots of tweets. In other contexts, on other services, there is similar behaviour but the mechanism is different. It seems to me that the word “bot” may turn out to be one of those things that was common and popular at the end of the 2010s and in the early 2020s, but in five years we will not be using it at all. It will have served its time, it will have expired and we will be using other language to describe what it is that we want to capture: a human being has created some kind of automated tool that will be very context dependent, depending on the nature of the service, and they are pumping out material. It is very clear that we want to make sure that such behaviour is in scope and that the person cannot hide behind the fact that it was an automated tool, because we are interested in the mens rea of the person sitting behind the tool.
I recognise that the Government have been very wise in making sure that whenever we refer to a bot we are adding that “automated tool” language, which will make the Bill inherently much more future-proof.
My Lords, I just want to elucidate whether the Minister has any kind of brief on my Amendment 152A. I suspect that he does not; it is not even grouped—it is so recent that it is actually not on today’s groupings list. However, just so people know what will be coming down the track, I thought it would be a good idea at this stage to say that it is very much about exactly the question that the noble Baroness, Lady Harding, was asking. It is about the interaction between a provider environment and a user, with the provider environment being an automated bot—or “tool”, as my noble friend may prefer.
It seems to me that we have an issue here. I absolutely understand what the Minister has done, and I very much support Amendment 153, which makes it clear that user-generated content can include bots. But this is not so much about a human user using a bot or instigating a bot; it is much more about a human user encountering content that is generated in an automated way by a provider, and then the user interacting with that in a metaverse-type environment. Clearly, the Government are apprised of that with regard to Part 5, but there could be a problem as regards Part 3. This is an environment that the provider creates, but it is interacted with by a user as if that environment were another user.
I shall not elaborate or make the speech that I was going to make, because that would be unfair to the Minister, who needs to get his own speaking note on this matter. But I give him due warning that I am going to degroup and raise this later.
My Lords, I warmly welcome this group of amendments. I am very grateful to the Government for a number of amendments that they are bringing forward at this stage. I want to support this group of amendments, which are clearly all about navigating forward and future-proofing the Bill in the context of the very rapid development of artificial intelligence and other technologies. In responding to this group of amendments, will the Minister say whether he is now content that the Bill is sufficiently future-proofed, given the hugely rapid development of technology, and whether he believes that Ofcom now has sufficient powers to risk assess for the future and respond, supposing that there were further parallel developments in generative AI such as we have seen over the past year?
My Lords, this is a quick-fire debate on matters where most of us probably cannot even understand the words, let alone the purpose and particularity of the amendments. I want to raise points already raised by others: it seems that the Government’s intention is to ensure that the Bill is future-proofed. Why then are they restricting this group to Part 5 only? It follows that, since Part 5 is about pornography, it has to be about only pornography—but it is rather odd that we are not looking at the wider context under which harm may occur, involving things other than simply pornography. While the Bill may well be currently able to deal with the issues that are raised in Part 3 services, does it not need to be extended to that as well? I shall leave it at that. The other services that we have are probably unlikely to raise the sorts of issues of concern that are raised by this group. None the less, it is a point that we need reassurance on.
My Lords, this has been a short but important debate and I am grateful to noble Lords for their broad support for the amendments here and for their questions. These amendments will ensure that services on which providers control a generative tool, such as a generative AI bot, are in scope of Part 5 of the Bill. This will ensure that children are protected from any AI-generated pornographic content published or displayed by provider-controlled generative bots. These changes will not affect the status of any non-pornographic AI-generated content, or AI-generated content shared by users.
We are making a minor change to definitions in Part 3 to ensure that comments or reviews on content generated by a provider-controlled artificial intelligence source are not regulated as user-generated content. This is consistent with how the Bill treats comments and reviews on other provider content. These amendments do not have any broader impact on the treatment of bots by Part 3 of the Bill’s regime beyond the issue of comments and reviews. The basis on which a bot will be treated as a user, for example, remains unchanged.
I am grateful to the noble Lord, Lord Clement-Jones, for degrouping his Amendment 152A so that I can come back more fully on it in a later group and I am grateful for the way he spoke about it in advance. I am grateful too for my noble friend Lady Harding’s question. These amendments will ensure that providers which control a generative tool on a service, such as a generative AI bot, are in scope of Part 5 of the Bill. A text-only generative AI bot would not be in scope of Part 5. It is important that we focus on areas which pose the greatest risk of harm to children. There is an exemption in Part 5 for text-based provider pornographic content because of the limited risks posed by published pornographic content. This is consistent with the approach of Part 3 of the Digital Economy Act 2017 and its provisions to protect children from commercial online pornography, which did not include text-based content in scope.
28: Schedule 1, page 185, line 23, at end insert—
“Public information services
5A A user-to-user service is exempt if its primary purpose is the creation of public information resources and it has the following characteristics—(a) user-to-user functions are limited to those necessary for the creation and maintenance of a public information resource,(b) OFCOM has determined that there is minimal risk of users sharing harmful content on the service, and(c) it is non-commercial.”Member’s explanatory statement
This amendment would allow OFCOM to exempt services like Wikipedia from regulation where it deems them to be low risk.
My Lords, as we enter the final stages of consideration of this Bill, it is a good time to focus a little more on what is likely to happen once it becomes law, and my Amendment 28 is very much in that context. We now have a very good idea of what the full set of obligations that in-scope services will have to comply with will look like, even if the detailed guidance is still to come.
With this amendment I want to return to the really important question that I do not believe we answered satisfactorily when we debated it in Committee. That is that there is a material risk that, without further amendment or clarification, Wikipedia and other similar services may feel that they can no longer operate in the United Kingdom.
Wikipedia has already featured prominently in our debates, but there are other major services that might find themselves in a similar position. As I was discussing the definitions in the Bill with my children yesterday—this may seem an unusual dinner conversation with teenagers, but I find mine to be a very useful sounding board—they flagged that OpenStreetMap, to which we all contribute, also seems to be in the scope of how we have defined user-to-user services. I shall start by asking some specific questions so that the Minister has time to find the answers in his briefing or have them magically delivered to him before summing up: I shall ask the questions and then go on to make the argument.
First, is it the Government’s view that Wikipedia and OpenStreetMap fall within the definition of user-to-user services as defined in Clause 2 and the content definition in Clause 211? We need to put all these pieces together to understand the scope. I have chosen these services because each is used by millions of people in the UK and their functionality is very well known, so I trust that the Government had them in mind when they were drafting the legislation, as well as the more obvious services such as Instagram, Facebook et cetera.
My Lords, Amendments 29 and 30 stand in my name. I fully appreciated, as I prepared my thoughts ahead of this short speech, that a large part of what I was going to say might be rendered redundant by the noble Lord, Lord Allan of Hallam. I have not had a discussion with him about this group at all, but it is clear that his amendment is rather different from mine. Although it addresses the same problem, we are coming at it slightly differently. I actually support his amendment, and if the Government were to adopt it I think the situation would be greatly improved. I do prefer my own, and I think he put his finger on why to some extent: mine is a little broader. His relates specifically to public information, whereas mine relates more to what can be described as the public good. So mine can be broader than information services, and I have not limited it to non-commercial operations, although I fully appreciate that quite a lot of the services we are discussing are, in practice, non-commercial. As I say, if his amendment were to pass, I would be relatively satisfied, but I have a moderate preference for my own.
20 of 116 shown
The right reverend Prelate the Bishop of Oxford is right to ask whether we think this is enough. These changes certainly help. The way that the Bill is written in a technology-neutral way will help us to future proof it but, as we have heard throughout the passage of the Bill, we all know that this area of work will need constant examination and scrutiny. That is why the Bill is subject the post-Royal Assent review and scrutiny that it is and why we are grateful for the anticipation noble Lords and Members of Parliament in the other place have already given to ensuring that it delivers on what we want to see. I believe these amendments, which put out of doubt important provisions relating to generative AI, are a helpful addition and I beg to move.
Secondly, can the Minister confirm whether any of the existing exemptions in the Bill would apply to Wikipedia and OpenStreetMap such that they would not have to comply with the obligations of a category 1 or 2B user-to-user service?
Thirdly, does the Minister believe that the Bill as drafted allows Ofcom to use its discretion in any other way to exempt Wikipedia and OpenStreetMap, for example through the categorisation regulations in Schedule 11? As a spoiler alert, I expect the answers to be “Yes”, “No” and “Maybe”, but it is really important that we have the definitive government response on the record. My amendment would seek to turn that to “Yes”, “Yes” and therefore the third would be unnecessary because we would have created an exemption.
The reason we need to do this is not in any way to detract from the regulation or undermine its intent but to avoid facing the loss of important services at some future date because of situations we could have avoided. This is not hyperbole or a threat on the part of the services; it is a natural consequence if we impose legal requirements on a responsible organisation that wants to comply with the law but knows it cannot meet them. I know it is not an intended outcome of the Bill that we should drive these services out, but it is certainly one intended outcome that we want other services that cannot meet their duties of care to exit the UK market rather than continue to operate here in defiance of the law and the regulator.
We should remind ourselves that at some point, likely to be towards the end of 2024, letters will start to arrive on the virtual doormats of all the services we have defined as being in scope—these 25,000 services—and their senior management will have a choice. I fully expect that the Metas, the Googles and all such providers will say, “Fine, we will comply. Ofcom has told us what we need to do, and we will do it”. There will be another bunch of services that will say, “Ofcom, who are they? I don’t care”, and the letter will go in the bin. We have a whole series of measures in the Bill by which we will start to make life difficult for them: we will disrupt their businesses and seek to prosecute them and we will shut them out of the market.
However, there is a third category, which is the one I am worried about in this amendment, who will say, “We want to comply, we are responsible, but as senior managers of this organisation”, or as directors of a non-profit foundation, “we cannot accept the risk of non-compliance and we do not have the resources to comply. There is no way that we can build an appeals mechanism, user reporter functions and all these things we never thought we would need to have”. If you are Wikipedia or OpenStreetMap, you do not need to have that infrastructure, yet as I read the Bill, if they are in scope and there is no exemption, then they are going to be required to build all that additional infrastructure.
The Bill already recognises that there are certain classes of services where it would be inappropriate to apply this new regulatory regime, and it describes these in Schedule 1, which I am seeking to amend. My amendment just seeks to add a further class of exempted service and it does this quite carefully so that we would exclude only services that I believe most of us in this House would agree should not be in scope. There are three tests that would be applied.
The first is a limited functionality test—we already have something similar in Schedule 1—so that the user-to-user functions are only those that relate to the production of what I would call a public information resource. In other words, users engage with one another to debate a Wikipedia entry or a particular entry on a map on OpenStreetMap. So, there is limited user-to-user functionality all about this public interest resource. They are not user-to-user services in the classic sense of social media; they are a particular kind of collective endeavour. These are much closer to newspaper publishers, which we have explicitly excluded from the Bill. It is much more like a newspaper; it just happens to be created by users collectively, out of good will, rather than by paid professional journalists. They are very close to that definition, but if you read Schedule 1, I do not think the definition of “provider content” in paragraph 4(2) includes at the moment these collective-user endeavours, so they do not currently have the exemption.
I have also proposed that Ofcom would carry out a harm test to avoid the situation where someone argues that their services are a public information resource, while in practice using it to distribute harmful material. That would be a rare case, but noble Lords can conceive of it happening. Ofcom would have the ability to say that it recognises that Wikipedia does not carry harmful content in any meaningful way, but it would also have the right not to grant the exemption to service B that says it is a new Wikipedia but carries harmful content.
Thirdly, I have suggested that this is limited to non-commercial services. There is an argument for saying any public information resource should benefit, and that may be more in line with the amendment proposed by the noble Lord, Lord Moylan, where it is defined in terms of being encyclopaedic or the nature of the service. I recognise that I have put in “non-commercial” as belt and braces because there is a rationale for saying that, while we do not really want an encyclopaedic resource to be in the 2B service if it has got user-to-user functions, if it is commercial, we could reasonably expect it to find some way to comply. It is different when it is entirely non-commercial and volunteer-led, not least because the Wikimedia Foundation, for example, would struggle to justify spending the money that it has collected from donors on compliance costs with the UK regime, whereas a commercial company could increase its resources from commercial customers to do that.
I hope this is a helpful start to a debate in which we will also consider Amendment 29, which has similar goals. I will close by asking the Minister some additional questions. I have asked him some very specific ones to which I hope he can provide answers, but first I ask: does he acknowledges the genuine risk that services like Wikipedia and OpenStreetMap could find themselves in a position where they have obligations under the Bill that they simply cannot comply with? It is not that they are unwilling, but there is no way for them to do all this structurally.
Secondly, I hope the Minister would agree that it is not in the public interest for Ofcom to spend significant time and effort on the oversight of services like these; rather, it should spend its time and effort on services, such as social media services, that we believe to be creating harms and are the central focus of the Bill.
Thirdly, will the Minister accept that there is something very uncomfortable about a government regulator interfering with the running of a neutral public resource like Wikipedia, when there is so much benefit from it and little or no demonstrative harm? It is much closer to the model that exists for a newspaper. We have debated endlessly in this House—and I am sure we will come back to it—that there is, rightly, considerable reluctance to have regulators going too far and creating this relationship with neutral public information goods. Wikipedia falls into that category, as does OpenStreetMap and others, and there would be fundamental in principle challenges around that.
I hope the Government will agree that we should be taking steps to make sure we are not inadvertently creating a situation where, in one or two years’ time, Ofcom will come back to us saying that it wrote to Wikipedia, because the law told it to do so, and told Wikipedia all the things that it had to do; Wikipedia took it to its senior management and then came back saying that it is shutting shop in the UK. Because it is sensible, Ofcom would come back and say that it did not want that and ask to change the law to give it the power to grant an exemption. If such things deserve an exemption, let us make it clear they should have it now, rather than lead ourselves down this path where we end up effectively creating churn and uncertainty around what is an extraordinarily valuable public resource. I beg to move.