Google is now legally responsible for AI lies | Paris Marx

[00:00:00]It opens them up to a new kind of a front of of liability and of financial payments that they certainly wouldn't want to take. Um and especially when you consider that the economics of AI are already not great because of how much Google is used, it means that there's millions of incorrect answers generated every hour. That's a huge amount of false information that people are consuming and in many cases will not know it is inaccurate. I don't think it should be incumbent on regular people to have to be checking over all of the mistakes that Google's model is making.

[00:00:30]When Google is presenting this as authoritative and trustworthy, right?

[00:00:35]They have designed them in this way. Um and so then to turn around and say, "Oh, but people should know that there can be mistakes. People should know that they should be checking." Um I think it's really disingenuous.

[00:00:54]>> Hi, I'm Isaac Pound and on the Tech Pod with me today is the tech journalist, host of the Tech Won't Save Us podcast, and author of the soon-to-be-released book Hyperscale, which takes a look at data centers and AI, Paris Marx. Thanks for coming on.

[00:01:06]>> Absolutely. Great to join you.

[00:01:08]>> So, some big news that's kind of gone under the radar as far as I can tell has come out of a regional German court and it ruled that Google is liable for the inaccuracies of its AI overview.

[00:01:20]The plaintiff sued Google after the search engine's AI overview connected them with scams and some dodgy business practices.

[00:01:27]The ruling essentially saying that AI overviews can't be treated the same as a typical search engine results because they're not like an index of sites.

[00:01:35]They're created the the overviews are created by Google and the company is therefore responsible for them.

[00:01:41]What do you think of the ruling and the impact it has on the idea that these platforms are are in some way neutral?

[00:01:48]>> Yeah, I think it's uh fascinating, right? Another kind of crack in their armor because for the past what, 3 years, you know, since generative AI has become a thing, you know, since the release of ChatGPT, these companies have really been trying to escape liability, right? More than anything. They've been trying to escape liability on copyright and having to pay the people whose copyright that they violated in taking all this material to train their models.

[00:02:12]Um, you know, liability around privacy because of all the personal information that they took on people in order to train these models. Um, and certainly, you know, liability for the outputs of, you know, the the chatbots that they have created, right? And to see a court in Germany really, you know, kind of be the first one to make a ruling like this where they are effectively saying that, uh, you know, the the content that is being generated by these chatbots, by these models, is like, you know, new material and thus is not protected in the way that like if you were just showing some search results would have been in the past or like an excerpt from something that someone else has written.

[00:02:48]Because this is novel that that the companies have created a tool that creates, you know, new writing effectively, um, that they are liable for what that tool creates. Obviously, they do not want to want it to be seen that way, you know, as Google was arguing, uh, in the lawsuit, um, because it means that now they are opened up to being sued by other companies or individuals who, um, you know, find incorrect information on there, who, you know, uh, you know, maybe kind of take an action based on wrong information, who, uh, you know, have an AI overview produce text that, uh, you know, defames them or lies about them in some negative way. Um, and yeah, so I for me, like it's it's a positive development in trying to hold these companies accountable for for what they're doing. I guess the question is, you know, whether this kind of holds in Germany, whether it expands, uh, through the country, whether other countries start to follow suit on something like this. But I think I think it's a obviously a positive start, right?

[00:03:47]>> Part of Google's defense was that it's the user's responsibility to verify things that they know as they know AI makes mistakes, but plenty of people treat AI as an authority. And there's been lots of studies that have shown that this happens even when people are trying not to use cognitive surrender, I think it's what it was called.

[00:04:05]What do you make of the the hypocrisy behind business models selling shortcuts while legally relying on users taking extra steps of verification to not fall foul of these shortcuts?

[00:04:18]>> I I feel like it's not a big surprise, right? Um like it it's kind of what you would expect from these companies, right? You know, they they want the benefit of saying, "Look, this is going to make you more efficient, more productive, like it's going to make things easier for you." Um but they also don't want to have to admit when the tools make mistakes and you know, actually maybe even make things less efficient because if you are checking things over, then it does take more work. Um and you know, maybe for certain tasks that actually is more work than if you had just done the thing the the regular old-fashioned way in the first place, right? Um and so what we know about these AI models is that, you know, the what the industry calls hallucinations, you know, the incorrect things that they output is part of it, right? It's not something that can be worked out of the model. Um and this is something that researchers at OpenAI admitted last year in a research paper that they that they published. And so we know that this is not something that the companies are going to get rid of. Um and like when you look at these the the answers that say an AI overview turns out, and I've been trying to pay attention to it, you know, as I've used the Google search engine and taken a look at it. And like obviously there are bigger mistakes that can be found in, you know, these outputs in these overviews, but there are also a lot of small mistakes that unless you really know what you're asking Google about, then you're not really going to notice them, right? And I and I don't think it should be incumbent on regular people to have to be checking over all of the mistakes that Google's model is making when Google is presenting this as authoritative and trustworthy, right?

[00:05:53]You know, the way that these models output text is in such a way that is, you know, presented as authoritative, right? As though it understands what is happening, as though this is something that you can trust. They have designed them in this way.

[00:06:08]And so, then to turn around and say, "Oh, but people should know that there can be mistakes. People should know that they should be checking." I think is really disingenuous.

[00:06:17]>> A statement by Google in which they promised to obviously appeal the rulings, they kind of say that the overviews are meant to reflect what's already on the internet. But some analysis done by Umi Elmi, I'm not entirely sure, done for the New York Times find found that 56% of Google's correct AI answers couldn't be traced back to the sources that they referenced in those answers.

[00:06:40]So, assuming that those numbers are at least somewhat accurate, how can these tech companies justify sort of this level of inaccuracy while presenting it as facts?

[00:06:50]>> Because I don't think the facts really matter, you know, of of what they're presenting, right?

[00:06:55]You know, they just want you to trust the tool. Certainly, they know that some references need to be in there in order to, you know, quell you know, criticism of what they are doing to be able to say, "Look, see, we're citing the sources. You can trust this. It is accurate. If you don't trust it, look, you can go back and and you can check it over." Because they know that the vast majority of people will not actually be doing that, right? And that is reflected in what, you know, news publishers, what website, you know, people who run websites have been finding since AI overviews have been introduced to Google, where their search traffic, you know, the traffic that they're getting from the Google search engine, has been plummeting because they know that as Google is using its chatbot to churn out these answers in the AI overview, that more people are, you know, reading the overview, getting their answer there, and then not going on to a website in order to, you know, check the the facts of it or to get more information or whatever. It's just there in the overview, and as Google has developed this further, now they kind of want you to have a conversation with the overview, right? That they'll present you the initial information, then you can you can submit a second prompt to like continue to get more information.

[00:08:05]And so they're moving more and more in the direction of like you stay on the Google platform, you don't go somewhere else. Um, you know, and this is this is completely how it's designed.

[00:08:16]They know that, but obviously they don't want to admit it because, you know, again, it's more things that they that thing comes back on them. Uh, you know, they'll present an idealized version in order to try to get around scrutiny and accountability, but that doesn't change the fact that, you know, we know how it operates, and they know how it operates, too.

[00:08:32]>> Do you think big tech companies think that they are above the law, then?

[00:08:36]>> Um, yeah, I definitely think they do.

[00:08:39]Uh, you know, that's it's not something something that they would admit, um, but yeah, they I, you know, I think on one hand, I think they do think that they're above the law, and you can see that in the way that many of the kind of founders and billionaires that are at the head of this industry talk, you know, they really don't feel that they should be held to the same rules and standards as other people and companies and and what have you, right? They do not want that. Um, but then, paired with that is how they are very wealthy, right? Their companies are some of the biggest companies in the world, and they deploy their power and their wealth in order to try to make sure that rules are written to favor their businesses, right? And and you see that across many different aspects of the tech industry, but obviously most recently with AI, um, they have worked very hard in order to shape AI regulation, um, to benefit themselves, right? And we have seen this in the UK, where the government has been very focused on creating rules and frameworks that are very, um attractive to and open to the AI industry to try to attract investment from those companies, whether that is in the development of AI tools or also in the construction of data centers. Um you know, I mentioned copyright earlier, but the government was moving forward um you know, kind of a copyright bill that was very favorable to the tech industry um and eventually had to kind of pull that back based on the opposition that they were feeling in in large part from the creative community, but also from uh you know, people more broadly. Um and we've seen them overrule uh opposition to data centers um around the country as well, right? Um and so, we can see that not only I think is it right for them to feel that they are above the law, but in many cases, they can get governments to act in such a way that does make them in practice um often above the law. Um you know, and hopefully we would see that change more and more. Um but when governments are trying to attract investment and when there's so much investment going into anything that has to do with generative AI, then they are going to be very favorable to those companies and what they want in in order to try to make sure that that capital flows into whether it's the UK where you are or Canada where I am or whatever other country is competing for uh you know, this this limited amount of investment though there is a lot of it.

[00:10:55]Um yeah, then then they are going to take those actions to to do that. And you know, it's they're certainly trying to increase economic activity as a result of it or whatever. Um but I would argue kind of harm the public uh you know, when those actions are taken.

[00:11:09]>> Is there a a glaring enough kind of disclaimer that would meaningfully intervene in people's like overly trusting approach to AI summaries and AI in general I suppose. Or does it kind of go back to the the fact that the use case of AI is to offload effort rather than kind of adding to it or moving it on to verification which most people probably aren't as as used to.

[00:11:32]>> Yeah, it's even even as you say that I I think about like cigarette packets and like you know, the the kind of big warnings that are on them now and it's like when you put a search into Google and the AI overview comes up on the top like before you even get into the answer there need to be like a big kind of yellow bar being like this might not be accurate. Do not trust what we have put here.

[00:11:54]Consider the environmental effects of you know, what this overview is and the amount of fossil fuels that are needed to to generate it or you know, the places where that it's denying of water because it's built at these massive data centers. Like you know, maybe there is like a series of disclaimers that can be put around these tools.

[00:12:12]But to actually imagine that happening in practice and you know, the governments even moving it forward let alone the tech companies like kind of you know, letting it happen is very hard to imagine.

[00:12:24]>> So kind of depending on who you ask the number kind of changes but generally speaking Google's AI view is about 91% accurate. But that obviously leaves millions of inaccurate summaries every hour.

[00:12:37]And I mean hallucinations are a mathematical certainty of AI.

[00:12:41]So if companies are held liable for those hallucinations as per this this ruling in Germany, how how damaging could it be if this precedent becomes widespread?

[00:12:52]>> Well, it could be huge, right? Which is why they're trying to avoid it because it would mean that then they face financial, you know, penalties for all of this inaccurate information that they are creating. And again, as you were saying, the inaccurate information cannot be removed from the chatbot. Like it is part of how large language models work. You know, even though these companies talk about you know, the large language models getting to a point where they're understanding the world or they're they're thinking about themselves, right? You see these stories pretty regularly.

[00:13:22]That's not what is happening, right?

[00:13:24]These these computers do not understand what they are outputting. They have taken in a load of data. They are trying to find connections between that data, and they are trying to find connections between the prompt that you put into the system and all of that data that they have to try to churn out, you know, a response that is going to try to best line up with the prompt that you have given it. It does not know if that is accurate. It does not know, like, you know, the the larger context around it.

[00:13:51]Um, that that is not part of what is going on here, right? And I think that it's really easy to, you know, forget that or mistake what is happening because of the way that the companies have sold the technology, but also because of the kind of authoritative nature of how the outputs are produced, right? Um, but, you know, if these companies are going to have to start paying or or are going to be open to, uh, you know, being sued if what they output is incorrect, you know, if they output things about people that are incorrect or companies that are incorrect, um, you know, that can harm them in some way, that give people information that if they follow that information can be harmed themselves, um, then yeah, it opens them up to a new kind of a front of of liability and of financial payments that they certainly wouldn't want to take, um, and especially when you consider that the economics of AI are already not great. You know, companies like OpenAI are losing a ton of money in order to try to create and sell this technology to people, um, then, you know, it continues to open the question as to whether this really makes sense at the end of the day, whether it's financially viable for these companies, especially when, you know, all the hype and excitement dies down and you're left with the reality of the economics of it. Yeah, whether that makes sense. And and you see this with many of these companies. They're constantly trying to avoid liability like this, right? You know, the one that immediately comes to mind to me is like the self-driving vehicle companies. And the companies creating the technology that, you know, supposedly drives the vehicles do not want to be held liable if then those vehicles get into crashes or something like that, right? They want it left to this the the you know, kind of human who is watching over the system or something like that, right? Or they just want nobody to be at fault or they certainly don't want to be the ones who have to pay in many instances. We've seen some Chinese companies say they're fine with that, but largely as far as I'm aware the western ones have been trying to avoid that very aggressively.

[00:15:43]So, yeah, these companies are always trying to avoid liability because it means more cost for them, right? Which is not a surprise. Um but especially when you know, we have this tool that is presenting itself as accurate and it is very inaccurate. As you say, you know, because of how much Google is used it means that there's millions of incorrect answers generated every hour. That's a huge amount of false information that people are consuming and in many cases will not know it is inaccurate.

[00:16:09]>> If this ruling does become wider and it does hamstring the industry, it strikes me that it's better to be on the safe side and have an overabundance of caution than than it just kind of roam free as it were cuz even just talking about this like AI psychosis and and the violence that that has already been caused. And some people might make the point that oh China or let them surge ahead of America or whatever else, but I mean to from my understanding anyway, China has a much more coherent AI policy than than America does as well.

[00:16:40]>> Yeah, I think it's a ridiculous argument, right? And it's one that the industry often deploys so they can avoid regulation and liability and having to act ethically and responsibly, right?

[00:16:50]You know, we can't possibly think about the impacts of our technology because then China will win and and do you want China to win? You want the the authoritarian communist to win? You know, we can't have that, right? Which is which is foolishness. Um you know, I think unfortunately that we have really let these companies get away with you know, not considering the harms and the costs of their technology because you know, as I was saying earlier, because we want the investment, right?

[00:17:16]Because we want to make sure that these companies are growing as large as they can, are contributing to GDP, are, you know, potentially paying taxes or or whatever at some point. Um you know, and certainly governments have even bought into the notion that using this technology is definitely going to make things more efficient or more productive. And you know, whether in the UK or in Canada, we have seen them already pushing hard to roll it out into the public sector without having a proper understanding of what this technology is, how it actually works, what the impacts are going to be, um whether it is actually going to deliver those efficiencies that the companies claim. And I think that's a very dubious claim, personally. Um and so, you know, unfortunately, like I feel like we're in this moment. And certainly, you know, the UK is moving forward policy on this right now, where we're finally grapping grappling with the costs and the problems of social media, right? And this is something that we have flat kind of fester and persist for a long period of time because we didn't want to have to deal with these companies. And we knew that this was going to be a difficult kind of policy making area in order to move into. And even now, we're focusing more on, you know, the effects on youth rather than the effects on the whole of the population. When we know that there are a lot of people who are negatively affected by what happens on social media. And not just like how the content flows around, but explicit decisions made by the companies for how these platforms work, how they're designed, how they kind of amplify certain information and content. You know, these are all decisions decisions that are made by companies that they should be held accountable for. Um and so, we're getting to the point where we're trying to deal with that. But on the AI front, we can see, as you say, there's a growing number of reports of the harms and the costs that can come with this technology on people, on the environment, on society more widely. Um but we still don't want to think about that. And we don't want to have to deal with that because that would mean scaring away investment. And we couldn't possibly imagine doing that in this moment where there's so much money flowing around. And so, you know, we let these companies off the hook, and then in what, 5 or 10 years, we're going to be saying, "Oh, we should have acted sooner sooner on this." You know, we have so many people whose mental health have now been affected. Um and, you know, we need to do something about it now, uh even if it's too late, when we could have done it much earlier. And all of those people probably would have been or many of those people would have been, you know, kind of protected from what they felt as as a result of this technology being able to be deployed in the way it has.

[00:19:38]>> The thing that strikes me is that not only we kind of repeating the same mistakes, but in uh specifically Mark Zuckerberg's case, we're doing it with the same people as well.

[00:19:47]>> Totally.

[00:19:48]>> And Section 230 is is credited generally by some at least for sort of the creating or letting the internet flourish into what it has become today.

[00:19:59]With a like the limited liability for for content being posted on on platforms by third-party users. It's a un-nuanced but a simple simple way of putting it.

[00:20:10]Is it not testing AI is now testing the limits of of that, I suppose.

[00:20:16]And to your point and to what I just kind of said as well, while to Section 230 is unlikely to be repealed, what do you what do you think happens next in in that kind of space?

[00:20:27]>> Yeah, I think it's getting closer and closer to being amended, to be quite honest. Um because, you know, in my opinion, I think it has always been exaggerated the role that that specific law has played in creating the internet that we have today. Um especially when you look at the fact that there are many jurisdictions around the world that do not have comparable legislation and where the internet works just fine in those places as well, right? Um I you know, I think that this is something that uh you know, particular types of internet activists have really seized on um and, you know, have kind of acted as though we couldn't possibly change this because it would destroy the internet. And to be quite honest, I have heard that about a lot of policies that have been changed over the past 5 to 10 years um and the internet has continued to persist uh, you know, in the way that it has. Obviously, it has a lot of problems, but those problems are not the result of the repeal of those legislations that, you know, these tech activists said we couldn't possibly touch, right? Um, and of course, we hear that today with the regulation of social media and think every time that there's a new attempt to rein in what these companies do, all of a sudden, it is going to destroy the internet. And, you know, I I quite frankly have just had enough of it, um, because I think that we're long past time where we need to take seriously the problems that these tech companies have created in our societies, um, to really start to rein them in, to really start to limit their, um, you know, their ability to evade accountability and liability for the actions that they're taking. Um, and it is very clear to me that with how these, uh, companies have created the platforms that we use today, whether that is with chatbots that generate, uh, you know, novel content, obviously based on the large amount of data that they've stolen from from everybody else, or the social media platforms that have all of us kind of posting content to them, but then they have designed algorithms to select certain types of content in order to elevate them and to push them at people, that they are acting more as publishers in, you know, that kind of action, right? Uh, they are making explicit decisions about what people are seeing, um, and, you know, that is not something that is just left to users, that is something that they should be accountable for doing. Um, and so, yeah, whether it's on AI, whether it's on social media, I think there is real, uh, a real need for action in those areas to finally start to take, uh, you know, to finally start to hold the companies accountable for what they're doing.

[00:22:43]Obviously, then the question is what is the right approach there? Um, you know, personally, I think that the use social media bans are not the best way to go about it, and that we should be looking much more broadly at the the actions these companies are taking, the way that they are designing their platforms, the harms that come of those design choices and those algorithmic amplification choices and things like that. Um I think we should be focused much more on that stuff than, you know, keeping people who are between 13 and and 16 off of social media. I feel like that's kind of like a bit of a distraction, personally. Um and you know, I I think that a lot more of those kind of like design regulations that we're starting to see in some jurisdictions should really we really need to be looking at the chatbots and thinking seriously about the effects that those are having with, you know, the false information they're spreading and all those sorts of things as well.

[00:23:30]So, you know, I think that there's a lot of opportunity in this space. I think that we're seeing movement, but the movement is not always the the best movement or, you know, what would really be ideal in order to address the harms that we're seeing. But, you know, I think starting to hold the companies liable when they generate false information and and present that as as accurate and trustworthy to people, you know, decent start.

[00:23:51]>> To your point, the social media bans to me kind of seem like uh socializing the responsibility of what should be the the companies that have created these tools these platforms or however you'd want to define them.

[00:24:02]And I suppose it's already seeming like they again this those same mistakes that same approach is coming for or coming from the AI industry to make the the user responsible for the thing that they are being pushed into and not even necessarily understanding or having opportunities to understand it or the ability to understand it without a a master's degree in AI computer science.

[00:24:28]>> Yeah, to to a certain degree, right?

[00:24:29]Like there is still a responsibility on the platform. Like, you know, many of these platforms have already set an age limit of 13, which I think is something that is often lost in these discussions.

[00:24:39]You know, we talk about them setting an age limit and banning young people and whatnot. It's like, well, there's an age limit there. You're just like kicking it up a few years and saying, "Actually platforms, um you need to actually, you know, try to enforce this age limit in some way, right?" Because often the the 13 age limit is there, but it's very weakly enforced until governments or, you know, start to make a fuss about it or activists or something like that, right?

[00:25:03]Um but I do think that the kind of I feel like the opportunity cost, right, of talking so much about youth social media bans is that we don't talk about the broader harms of social media and and whether it's chatbots as well, you know, more broadly that affects the whole of the population, right? I think that, you know, we have all you know, been affected by things that we have seen on social media, whether that is extremist content, you know, whether that's really harmful content, whether that's stuff that, you know, makes us see our bodies in, you know, particularly negative ways. And I think that that can be more acute for younger people who are still developing, in certain social environments and and things like that. So, I do think that there is a reason to think differently about how younger people are engaging with these platforms and, you know, whether that means that there should be greater restrictions on the way that, um you know, the platforms you you know, the in the way that the platforms work when younger people are using them, right? Um and then maybe that opens up a bit more as you get older. Um but I you know what, I think we also see that, you know, there are a lot of older people who, you know, are using these technologies, who are on these platforms, using Facebook and whatnot, especially now with generative AI, who are running into a lot of false information and really don't have the tools to be able to understand, uh you know, what is false and what is not. Um and and are, you know, kind of consuming a lot of erroneous and and not just erroneous content, but stuff that is also trying to shape their opinions of the world, of politics and and things like that, right? In in really harmful ways. And so, you know, this is I think this is really not something that just affects people 13 to 16 years old, um but affects people more broadly. And when we only talk about the effects on people in that age group, then we really miss out on the broader effects of social media and we let these companies off the hook with what they're doing to everybody else and what they're doing to society more broadly. And that's why I think that we need this kind of broader regulation, and not just to look at, you know, what young people are doing online and trying to get them off. Um and and to me, it's like there's one other piece of this that I feel like is often left out of the conversation, which I know has been part of the conversation in the UK, but is like why are young people on social media so much, right? Um it's because that's where their peers are, that's where people go, but also because there's been a huge cut to like the funding of programs meant for youth, right? So they actually have something else to go and do uh when they're not in school or whatnot. Um so that maybe they're not looking at their screens so much, right?

[00:27:28]And I feel like the kind of like physical in-person aspect of it, and like actually providing things for young people to do to enrich themselves, to enrich their social lives, all those sorts of things, is something that we're missing. And I would say that that kind of goes for a lot of other age groups as well uh in actually making sure that there are things for people to do so that they're not just retreating into their screens and not having kind of like a vibrant social life.

[00:27:51]>> Just finally, and going back to the the liability ruling against Google, this was something that completely new to me it was when I was researching all of this, is that the output of code and algorithms in America are protected by free speech, or there's been precedent set that they are protected free speech.

[00:28:08]So European free speech laws are of course different from America's, and I suspect Canada's are probably different as well.

[00:28:14]>> Yeah, we're much more aligned with Europe.

[00:28:16]>> What do you think about the idea that these kind of outputs could be protected in that way? And I mean that that is something that is already being put to the test a little bit with Mythos and Fables as a defense for uh get getting it opened back up again, cuz somehow whether whether the fear-mongering is accurate or not, protecting people from something is somehow even if it's code is somehow also against free speech. I I'm English, I don't understand it, I'll admit.

[00:28:46]>> Oh, totally. I think that the US conception of free speech is so extreme compared to how we tend to see it in Canada and and in Europe, right? Um and to me I it kind of frustrates me when we then have people trying to push that conception of free speech onto our societies. Um and to expect us to, you know, kind of completely reorient how um we see things, how we see acceptable speech. Um and, you know, in the many ways that the US conception of it benefits these companies and and how they have evaded liability and accountability because of the way that free speech has been kind of constructed and and implemented uh in the United States, right? Um yeah, I I don't know. I I just think it's it's really really foolish and not something that we should be letting them get away with whether in the United States or beyond. You know, obviously we can't control what happens in the US, you know, we know they're a very unique um but uh no, I like I I think it it should have no kind of impact on the conversations that we have about how we approach these tools, you know, these technologies, how we regulate them. Um and, you know, to to not feel held to standards that are set in the United States that are often standards that are set in order to benefit uh their companies and their corporations uh even at the expense of of their own people, right? Um and of course we know that because it's been a thing that they, you know, have have really um held onto for so long that a lot of people in the United States also believe in that vision of free speech even when it doesn't really work for them, I I would say. But, you know, I'm sure that there's a whole debate within that whole question that I have no interest in having to be quite honest, but >> No worries.

[00:30:37]But on that note, Paris Marx, thanks for taking the time.

[00:30:40]>> Absolutely. Great to talk to you.

[00:30:42]>> If you enjoyed today's episode and you want to hear more of the Tech Report, please consider liking and subscribing.

[00:30:47]Also, you can get episodes of Tech Report wherever you get your podcasts.