You're listening to Enough About AI, a podcast where roughly once a quarter we return to discuss some of the key developments in artificial intelligence and how it's affecting our lives and society. I'm Dónal Mulligan. And I'm Ciarán O'Connor. And we've spoken plenty of times on the podcast about AI hype, where we're constantly being told about the opportunities to be gained from generative AI. You've heard enough about those. So in this episode, we're going to talk about the costs. On a lot of previous episodes, we've talked about the intricate details of models. We've also talked about the vast amounts of money pumped into generating these tools, generating these models. But today, on today's episode, we'd like to talk about the broader terms of costs, some of the social costs, some of the job costs, and even some of the kind of information costs that AI is bringing forth. So, Dónal, to kick things off, I think I want to start with some costs around electricity prices. It's often the most tangible, apparent way in which we notice the role of technology, the role of AI in our lives. And that particularly revolves around data centres. Yeah, just this week, a UN report has come out that has described Ireland as a "cautionary tale" in terms of the impact of AI-related data centres here. Currently, they're consuming, I think, over a fifth of our national electricity, and that's set to climb even further. And that's in comparison to data centers in other European countries where they're in the low single digits in terms of their percentage of electricity being used. So we're definitely greatly out of alignment, perhaps, with other countries on this. So there's an interesting case there. And that report was drawing attention to the financial costs of that electricity, but also the environmental costs of it, too, as it described us in that way. we're recording this in june 2026 and it's warming up a little bit for us now but we've come through a winter and a pretty cold and rainy spring where we all felt a lot of those costs very directly in our pockets there was government subsidies in prior years for electricity in winter and so we didn't perhaps feel it as strongly as we we just have in the past few months um but uh another report out this week one from friends of the earth was really effective in tying the higher cost for energy in ireland to the growing number of data centers here as well it's worth maybe talking a little bit here about why this data center explosion is happening everywhere really but disproportionately happening in ireland too so it's perhaps useful to say that data centers aren't something that's completely new of course cloud computing is already something that we you know take for granted as part of our lives we are very dependent on storage that's off our devices in many cases so lots of us have phones that back up our pictures or to our iCloud or various other cloud services that we use. It's part of our work. We use data centers, of course, for all sorts of online activity, commerce and remote computing generally, being able to connect to a remote server that does stuff for us. Of course, that's something that we've been doing for years and years. What's new here is that the AI data centers in particular are much, much more power hungry. And that's because they are bigger in terms of the actual physical size of them. Some of them extraordinarily large as some recent reporting will have shown. But also the kind of chips that they're using are ones that take a lot more electricity and a lot more water. And so what's happening there is that the often NVIDIA, we talked about that company before, the chips that they produce are chips that are kind of a cousin to the graphics card in your computer. They work slightly differently from the processor in your computer. And we've talked about that in prior episodes, but they do that so that they can do thousands and in fact, billions of calculations in parallel all at the same time. And part of the effect of doing that is that it generates a huge amount of heat and uses quite a lot of electricity. And that heat then, of course, is cooled away using water. So these data centres tend to use massive amounts of electricity and massive amounts of water. In Ireland, we have already been building these data centres and we're continuing to do that. Some of them were approved years ago and are just being built now. Some are on the way. So there's certainly, it continues to be the government's position that we should locate more and more of these here. In part, that's because of Ireland's general position as a tech hub. But also, I think that government policy is directly encouraging companies to build here. Initially, during the term of this government and the previous government, I think that has been related to our renewable energy generation. So it was considered that because we are doing so well, or we're doing so well in terms of our renewable output, that it would be green energy that could be used to run these data centres. But of course, despite Ireland's quite substantial progress, especially in wind energy in recent years and decades, we far outstrip that in terms of our energy consumption. And most of the energy now, of course, is from gas fired electricity. And that was already expensive because of the Russian invasion of Ukraine. It got further expenses to it from the kind of petrogas shocks from the Iran war. And now we're in this kind of "cautionary tale" that is described in that UN report where there's a perfect storm in terms of prices for us. And we're really feeling it. Yeah. Some of the numbers involved are very interesting. Ireland is the data centre capital of the world with 89 confirmed data centres in place with I think over 30 more in development, more dispersed around the rest of the country outside the capital. Within the capital, many of these data centres are located close together in business or industrial parks, putting lots of energy demands in very clustered areas. that some experts say that the kind of usage of these clusters of data centers is equivalent to small cities the size of Kilkenny for example, and I think what I found most interesting about that Friends of the Earth report was how they were some of their modeling suggested that this rapid expansion of data centers really since kind of 2017 onwards it seems like could be adding the equivalent of an average of 360 euro to household electricity bills all because of the way the the kind of state has pivoted towards allowing and enabling this massive surge in data centers across the country. And not something that the the state is backing away from either so despite admissions this week that from the government that we will definitely fail to meet our our legally binding climate targets for emission reduction by 2030. And, you know, we will have fines in the billions associated with that. And despite those household costs that we're just talking about, the minister responsible, Peter Burke from the Department of Energy, Trade and Employment, he was launching a report this week that his department commission from KPMG called "The value of data centres to Ireland". And in that he was effectively saying that, you know, the renewable energy that Ireland produces and will continue to produce. This is something that we need "energy users" for. He was identifying this idea that we need to find ways to use this. So rather than linking this to ways to reduce costs for people, I think he's interested there in developing that sector even further. And certainly that report talks a lot about the opportunities of AI. So with that, I would assume that it's very clearly government policy to continue to host these centers here and also to you know associate those with the development of these very large AI data centers these power hungry ones that we're talking about. Yeah and if the earlier generation, I suppose, of data centers had had cooling challenges perhaps the the AI-centered ones have something of a cooling crisis as you said the heat and the energy produced by these these clusters of kind of like gpus of of microprocessors type devices require enormous cooling systems and often require then enormous volumes of water as well i know that um some of the major technology companies microsoft and google and others have faced scrutiny particularly in in drought prone areas not something that we have particular challenge of here in ireland but in other geographies certainly yes and i think it it's also interesting that we are seeing in other countries such as France, for example, we are seeing a kind of glut of data centers that are kind of being tied to that country's ability to produce nuclear power, but also being financed by some other major players within the kind of AI technology and banking sphere in the case of SoftBank, which is quite interesting. There's some significant sums of money involved there too. Yeah, that one's interesting because we covered this previously. SoftBank had kind of taken a step back a little bit in terms of its investments recently. So where it is investing, it's going to be places that it's more sure about return and value. And I think the reason why SoftBank is interested in France in particular is because, as you say, France has a largely nuclear-powered grid for its setup, which is quite different to ourselves in Ireland, or indeed most of Europe, which is running on lots and lots of gas and then some amount of renewables. The nuclear power grid gives you greater stability, I suppose. So it's a more interesting place perhaps for them to try and position that. And maybe that's part of the discussion that recently was had in Ireland about whether Ireland should adopt nuclear power, which I found bizarre, to be frank, because it was strange to hear the Taoiseach kind of throwing that out there when we have absolutely no initial planning done for this. We have no idea how we would handle the waste or the cost. So I think it's one of those very pie in the sky things that seems like a good idea to make it look like the government was responding to the crisis in energy production at the moment, but I don't think it has any real legs. The one in France is interesting because it's sort of pointing at a need for stability in the grid from the kind of investors in this. And that's obviously something that renewables don't provide as easily. So renewables are great while the sun shines and the wind blows, but of course you need massive battery storage or you need something else as a backup when that doesn't happen. And nuclear offers one version of that. The other thing that, of course, fills in for that a lot of the time is local gas turbines. So in many cases, indeed, in several of the cases in Ireland as well, some of the large data centres in Mulhuddart and in Blanchardstown, Lucan, around Dublin, those data centres have on-site gas-fired turbines for their own electricity generation. And this is quite problematic for local communities around them, too, because those are noisy and they pollute. There's, you know, clearly byproducts of the combustion of that gas then released into those local areas. And this is a kind of secondary, but really important concern, this idea that as you build out this infrastructure, if you were allowing alongside it local additions to the power generation, that's, you know, it's effectively a private power plant on site. There's, you know, there are real questions there in terms of the environmental and health kind of consequences of that. And we've seen this not just in Ireland, but the Elon Musk backed Colossus facilities. So the ones that are being built in the US as well are often being built in communities that only later find out that there's on-site gas. In fact, I think in Musk's ones, he brought quite large mobile gas fired power generation to those sites after the fact and kind of shocked some of the people there. So lots of problems associated with this. Yeah and what we're seeing are quite often the communities where large data center clusters are being located are often themselves disadvantaged facing a long history of socio-economic challenges um task the think tank for action on social change recently in in May/June 2026 put out a report that found nearly 90 percent of the data centers within Dublin were located in disadvantaged areas um so we can see kind of mapping against the the deprivation index by Pobal there that a lot of these centers are clustered in areas that have faced you know a number of challenges throughout the last decades and then added to the mix are great greater i think clusters that are using up energy but also emitting emissions at a far higher rate as well whilst at the same time we see i think um large technology companies perhaps falsely portraying data centers and this transition as something green in all in all aspects but that's not quite the truth sometimes, no? Yeah, I think i that was a problem i think in Ireland too I- I think there was a period during which some of the discussion around ireland adopting these data centers was quite strongly tied to both our renewable generation and also the fact that we might be able to reuse the heat that's generated in some of them in local community heating projects, or that the water that's going into them would not be wasted if it was a source of heat that could be then piped to homes, et cetera. And that has not really come to pass. So, you know, while we have made great progress in Ireland in terms of our renewable generation, we've now outstripped that progress in terms of our consumption. And the data centres are a major part of that and seem to be continuing to be. And we haven't really had those secondary benefits in terms of cheap sources of local heating or other things that might offset people's need for electricity. So I think we're in a really problematic place there. It's something that people have long been talking about the environmental impact of. And now, of course, it helps often to get a point across when it also has a tangible financial impact. But it's quite clear that there really is a major effect on our bills that Friends of the Earth report, I think, did a very good job of making concrete for people in terms of monetary value. Yeah. And then just to kind of move on then, talking about the effects of AI in another area to talk about the effect of AI on jobs um we're seeing now fields on a recurring basis news reports announcements of job losses predominantly in technology industry technology companies but broader than that too and the kind of subtext of this always seems to be its labor displacement as companies transition to a greater role for ai within their workforces. I suppose the headline report of the period since we recorded last was hundreds of job losses within Meta in Ireland and globally as well as a case of this organization shifting towards AI, well, greater role for AI in training with models by staff as well. Yeah, there's, I mean, we'll come back to Meta's training of its models because I think there's some really quite dystopian stuff going on within that. But I think it's worth saying that there's a difficult to discern issue that's going on alongside this, where I'm sure to some degree, there's a lot of jobs that have been automated away by some aspects of AI and that there is that kind of effect that we sort of see as the simple cause and effect thing that's happening here. These job losses seem to be directly a result of an AI agent replacing a person in a kind of one-to-one fashion or whatever. But in many cases, I think there's also a downsizing that's been going on for a while in a lot of tech areas since COVID in particular, that maybe AI gives an extra fig leaf to, that if you want to at the moment cut costs and get rid of staff, you can say perhaps it's AI even when it isn't directly. And there's also a real change in the way in which shareholder value is now viewed in this kind of boom that's been happening because of these small amount of incredibly, incredibly wealthy companies. As we move into this period when the biggest IPOs ever will take place, judging by the figures coming out for the IPO for, that's the initial public offering, the sort of floating on the stock market of OpenAI and of Anthropic. The last while, as their values have increased and increased, there's been pressure on other companies, including in the tech space, to really cut costs as much as they can in order to drive their profits, in order to make each of their quarterly returns look better and better and better. And one of the ways you can always do that is to get rid of a load of staff. And so if your concern is really to drive the bottom line in order to make the next quarterly report look good, you might get rid of a lot of staff and you might say this is because of savings and efficiencies in AI, even if those efficiencies aren't as clear in every case. And so it's difficult to fully assess that space because a lot of it is obfuscated to some degree, but I don't think it's as simple as AI really is effectively replacing a lot of these jobs. I it's also a useful screen when other more ruthless decision making might be going on And yet we still see over the last few years as you mentioned it's a broader trend of of technology companies of social media companies shedding large amounts of staff even in an area closer to my my day-to-day work of trust and safety and looking at how platforms are kind of i suppose policing their own spaces we've known that the largest uh social media platforms out there the facebook's and X (formerly Twitter)s of this world have let go large numbers of staff. And that was a factor even before we'd entered this AI age as well. But the scale of some of these job losses, even within Ireland's tech sector are, well, they're large. In the year to March alone, there's been some 20,000 job losses. And they kind of seem to bring home the reality of AI and kind of larger labor displacement within this industry for the sector in Dublin, but also across the country, so many other kind of tertiary industries depending on those jobs too.(You) mentioned Facebook, there was an interesting piece I read during the week of this kind of job displacement, labor displacement impact of AI. Is Meta recently announcing, or at least speaking publicly, about what it's called its "Model Capability Initiative" and this is i think a fancy name for its its initiative whereby employees of meta are now being tracked internally to train the the tech the company's own ai models um through the use of monitoring of keystrokes through the monitoring of mouse clicks to train, in effect, the models that are going to probably one day replace their job or their type of job. And it all feels a bit dystopian, even the part whereby if employees wish to opt out of this initiative, they can only do so for 30 minutes at a time. Yeah, I mean, we're really in an episode of Black Mirror when we start looking at the specifics of that particular case. It's very dark. I think there's quite a few industries where people have reported this in the last while. So like you, I know people who are in that sort of trust and safety space as well. There was a lot of people working in kind of contracting firms for the larger companies. So rather than doing it themselves in-house, so companies like Meta, et cetera, might have another contractor take on a lot of their moderation work or their work to review content or whatever it might be. And in lots of those cases, the people employed in those were employed on different kinds of contracts. And many of them recently reported that they were effectively being, their work was being logged so that a model could be built based on their decision-making. So they were effectively training up the model that will replace them even as less secure workers to begin with. And of course, that's been a pattern that has existed across the space for quite a while. There's been a huge amount of kind of labor extraction out of African countries, Kenya in particular, was heavily used for cheap labor for the training of models, the cleaning of data at the kind of training stages. This is the sort of refinement of models for particular purposes to replace workers. And so it's part of a wider, yeah, really quite depressing trend, I think, that people are seeing. And I think it's making people more suspicious of these tools coming into their workplaces and whether it's something that, you know, like in that very obvious case where they're being tracked mouse click by mouse click in Meta, I think lots of people will start to become very wary of when, you know, co-pilot or another assistive tool is constantly running at all times, recording what they're doing, whether that might be part of an effort to eventually make them replaceable. Yeah, I think so. And it's really where a lot of my sympathies lie to our new graduates, our people coming out of university after a number of years studying these topics and finding that there's not the same demand for new graduates who are trying to enter the workforce not just in in these you know large technology companies or social media firms but also broader industries broader companies because lots of these companies are now automating their their workforces and I find it I'm curious when I scroll LinkedIn nowadays and see friends from college friends from secondary school who are now working in different industries but may have some one foot in the technology field and they're talking about the cost of "tokens" uh they're they're they're talking about essentially how the role how the role of AI is increasing in their own um their own places of work as well but we're now beginning to see in another way the costs to users and the realities of token billing coming back to well to bite, really, aren't we? Yeah, this is an interesting cost because uh whereas our our electricity costs or our costs in terms of job losses are costs to us for the industry as a whole expanding. This is a very literal usage cost if you choose to take on a particular model and start using it in your workplace. And so what has happened to a very large degree is that in order to get us, incentivize us to use these models as much as possible, the major companies involved, so Microsoft and OpenAI or Anthropic, have really heavily subsidized the usage that people have been able to make up until now. So even on the quite cheap monthly subscriptions, you were getting a huge amount of value in terms of the actual token usage. So to go back to when we mentioned them before, tokens are those little chunks of data that are the underlying units that the large language models are dealing with. So they're usually short words or part of a word. And so you are providing a load of those in the prompt or in the document that you're providing to the AI model. It is processing those by thinking about what tokens make sense together and creating, synthesizing some output based on the directions it's received and the tokens it had as an input. And it itself is then creating a stream of tokens back to you. It's typing back to you or it's creating a document or whatever it might be. And so up until now, the costs of those tokens were not really directly used in terms of the billing that you had for your usage. You were being subsidized and lots more capacity was given to you in order to you see how good these models could be and to impress you and to maybe draw you into perhaps using the models more, becoming more reliant on them, or as a manager, maybe perhaps thinking about replacing staff with some of these models or with one member of staff using various models to be more productive. What's happening now is as part of what we've been talking about for a very long time on this podcast, the need of these companies to start recouping some of the unbelievably huge amount of money that they've had invested in them, they're now much more accurately beginning to bill. And so Microsoft, which had Copilot built into GitHub, one of the kind of key pieces of infrastructure for software development, it's a kind of an online archiving resource where people can submit code and warehouse pieces of code or share a code base for things that they're developing. They struck a deal to bring Copilot into that and to allow people to very easily start using that to code alongside what they were doing themselves. They did that in part because a lot the success of Anthropic recently was very tied to Cloud Code, which was seen by a lot of people as, for a while, one of the most advanced ways to kind of co-code or to enhance what you were doing and to "vibe code" to just people who hadn't any or had very little background in coding could describe what was needed and have it create something for them. So that success there, OpenAI success with a product called Codex, which is very similar. And then Microsoft's, all of these were driven on providing a huge amount of token usage for a fairly low cost. And now all of them are starting to really realistically charge for that token use. And we mentioned in our last episode quite some time ago about the surprise bills that can come when you start billing by tokens and you don't realize quite how much you're using. This is very much the case here. And so the two things that drive this are this kind of complacency that people might have gotten into in terms of how much they depend on using these and so they might keep up a level of very high usage. But also because the models themselves often now have extremely complex chains of thought going on within them, all of which requires burning up tokens, using tokens internally before the response comes back to you. So we sometimes still erroneously think about these models like the very early versions of the LLMs like GPT that were an input and quick output that we said something and we fairly readily got a response almost right away. We're now talking about models more complex, that pass your prompt through various kind of iterations where it looks at and breaks down and compares the prompt. And it might take several seconds up to minutes to respond because it's doing quite a lot of processing in the background, all of which is using up your budget of tokens as it does that. And of course, sometimes it's doing that for a long time and returning something that's not especially good. You might try and create a breathtaking Shakespearean sonnet or a fantastic new app, and you might get a lot of rubbish back, but you have still paid for the tokens either way. So one of the things that's really making people quite unhappy now is that there's a combination of much higher bills, but there isn't in many cases a commensurate guarantee that when you spend all these tokens, you're definitely going to get the thing that you asked for. And so this is really shining a new light on AI tools and their use in terms of being an expensive, but perhaps not always reliable way of doing things. And so I think some of the stuff that we've talked about in advance of this recording. Some of the bills that have come in have been fairly shocking for people. There's companies really blown away by what they've seen. They have, yes. Axios, the US kind of business tech news website recently reported of a company over there who supposedly spent $500 million in a single month after failing to put usage limits on Claude licenses for employees. And that's just one company that was willing to speak about it. I'm sure there are some other horror stories of employees who've been left to, well, just left to work away, perhaps without the requisite awareness. And it seems as though with the initial lower cost kind of incentivization by AI companies for tokens to be used, for companies to get comfortable with them, we're seeing the kind of laying of foundation, laying of this kind of infrastructure and really dependency on these tools and the kind of wholesale movement from perhaps more normal kind of traditional staff basis towards more automated tasks and kind of seeing the displacement of that and the cost, as we've mentioned, of that as well. I mean, I think we were always looking at some version of this where this cost has to be recouped somehow. I think it's possibly not that surprising in this case that they've had to move towards this. We have talked about, I mean, we've had our own "Bubble Watch" on this for quite a while, and that bubble has only grown for all this time, but we still have not seen this huge return on investment that's supposed to come. And so this is a way that these companies, now that they're moving into this period prior to their IPO, now that they need to really start showing that there definitely could be some return, it's not perhaps terribly surprising that they're going to start upping the costs for their users in this way. But it's not smaller companies like the one you mentioned, I think Uber recently talked about the fact that their token budget for the year was already used up in the first three or four months. So their intent for what they would use over the course of a calendar year was already burned through. So of course, that would imply that their overall bill, if they stayed at that rate, would be four times what they expected. So these kinds of things, I think, are going to cause some tension because in order for this IPO to be successful for these companies, they really do need to show some sort of return. And so if there is now a shake in confidence, it will be interesting to see in the next month or two how this pans out in terms of usage and whether companies will, like some of the ones that you've mentioned or that are being reported on, scale back their use or have to think differently about the freedom that they allow employees to use these things alongside their work. The kind of shift towards greater AI, I suppose, activity or dependency is sometimes carrot sometimes stick uh the the what i'm kind of getting at here is the costs of information access and costs of knowledge and that kind of knowledge dependency that we kind of shift towards with the greater use of ai it's another thing i'd like to kind of get on to today um there was a recent piece written i believe by the provost of Western Sydney University in Australia. And I believe the provost was arguing against the use of AI, or at least to not overindulge and become too dependent on it themselves. But it turned out that, and this is one of numerous examples, of course, but you know where this is going, yes. Yeah. I mean, it's a particularly embarrassing one. You don't want to be an academic when other academics are doing this kind of thing, but this was a fairly flagrant. I mean, I read the original opinion piece that they wrote, and it's so clearly AI. As I think I've mentioned on the podcast before, I have to read an awful lot of AI because of student submissions for essays and things like that. And so you become attuned very quickly to the cadence of some sort of, there are tropes to how AI, especially ChatGPT, tends to talk.

"it's not just revolutionary, but it's also seismic!" you know, these kind of those doubles and yeah, those kind of adjectives are used. And so, I mean, this piece is full of those really from paragraph one of this, it's very clear that it was written using AI. But this is not the only case where we're going to see this. I think there's, there are so many people who have become dependent on it, who've become less critical in looking at the output and making sure that it even sounds like them or that it's something that they would say. I think as people reach for it so easily, we're going to see more mistakes like this person And I think in that case, it's particularly problematic because it's a very senior academic sort of talking to students about how important it is that they don't use AI. And it's a very lazy thing for them themselves to have done that. But we've seen this in other areas of media. We've seen this with journalistic figures using this. It's that same thing that we talked about with regard to the token usage, that effectiveness that was handed to everybody at such a cheap cost has really changed how a lot of people see the ease of picking up these tools and replacing what they would have done much more thoughtfully before themselves with a quick click or a short prompt. It's lazy, is what I think it is. I'm also reminded of the example of, I forget his name, but the senior figure with Independent News and Media who was writing on his own blog about the topic of artificial intelligence and was then found to have used generative AI on that blog and I think other blog posts as well and what's so I suppose striking about either an academic or someone working in in journalism for a number of decades is that so much of that profession is built around trust reputation the veracity of information that they are communicating and the standard of how they are trying to communicate it and to wholesale replace that whole cognitive process with something that is it seems quite haphazard ai maybe maybe elements of even slop i mean if it was written by themselves manually in the old-fashioned pen and paperweight probably wouldn't have ever seen the light of day because as you say the barrier was lowered there was a kind of um a logical bypass perhaps and we're kind of seeing the effects of these tools and these technologies especially on people who operate as you know a kind of very high in either intellectual or executive level within our society um and what that might say for people who are who are just starting off in university or schools you can probably you can probably speak to that um more so but the way in which we are engaging with uh generative ai is also increasing increasingly changing how we're accessing information how we're even engaging with these technology companies that we've kind of grown up with uh notably google has made a significant change to its own search engine um experience i suppose that they're now changing from maybe what you might have previously called quite blunt short searches and responses into more of a flowing conversation in which you are back and forth engaging with an AI um chat but there's also of kind of risks inherent in this too, isn't there? Yeah, I think this is part of one of the costs that we wanted to talk about today that's less obvious is this kind of cost for information and veracity, like whether we can access truth anymore is going to become more and more problematic as AI gets in the way of that. And it's doing it in both of those things that you've mentioned. It's both problematic that people are reaching for it as a tool to just output slop when they might previously have outputted something thoughtful. And so their contributions to the world are AI-mediated and might be, you know, certainly are linguistically of less quality. So they're definitely poorly written. They're full of those tropes that I mentioned. And there's a lot more of that stuff because it's cheap to produce. But then our access to information, exactly as you say, I think is really challenged by this decision by Google to go "AI-first", as they describe it, that the sort of results that you see by default, unless you go out of your way to disable this, is now going to be an AI synthesis of the information that's available. And this has really profound consequences because just in terms of the model of the web for a very long time now, that publishers were still able to get people going through to their websites and therefore able to generate income from that. And this is what kept a lot of news websites and things going. It will stop that because it will instead access information that's on those places, synthesize it and provide within Google without you ever visiting that site, some sense, you know - "terms and conditions apply" - of what that original information was. And even the fact that it's doing that may also blur something that previously was useful, which was the idea that if something was further down the rankings or was harder to find on Google, it might be less significant. So there might be more obscure information. There might be, you know, things that there was a reason why they were on page 14 of the Google search results, but now presented in this new way. They'll often appear equally nicely put together in a little narrative piece by the AI. And so that's problematic in that it means that things seem to have the same truth value, regardless perhaps of where they've come from or how much corroboration or how much significance there was for this thing in the first place. And I do worry that that's going to make people, again, even more unable to really get a handle on what is true and important and backed up in several places versus what is something that has been dressed up to look like it is that because of this AI sitting in the middle. Well, we see this quite often in in my regular job where we're i suppose examining significant instances and trends in missing or disinformation on social media and in that this infinite scroll world in which we now exist there is equal weight given to perhaps the the months-long BBC news investigation that took multiple multiple hours of investigation and corroboration and everything else to getting ready to publish at the same time some person who has quite conspiratorial opinions about the same topic is given equal weighting in how you and how that content is delivered to you how it's presented to you and in also the kind of time it takes you to process that from a cognitive point of view and in a similar way we're perhaps seeing kind of mirroring of that from generative ai tools which are uh perhaps perhaps just kind of being almost quite sycophantic and kind of confirming the the kinds of searches the kinds of information that it thinks you may be interested in but not giving equal weighting to exactly as you said the veracity of the message or the messenger in particular we know that there is also attempts to, I think we covered in a previous episode, groom or kind of inject perhaps less positive I think it's interesting to see that people are beginning to turn away from Google for their search, for example. I mean, Google is a verb because of how great a stranglehold it has in the idea of searching for information. And so it's interesting to see that just this week, the European Parliament has ditched Google in favour of Qwant, which is a European-based search engine, but also, very importantly, is one that has much stronger privacy capabilities going on and guardrails going on within it, but also isn't doing this kind of AI synthesis and presentation of information. So I think it'll be interesting to track whether this is part of a wider movement, perhaps more likely in Europe than elsewhere against this sort of thing. But certainly myself, I have I've disabled this and I've gone out of my way to find there is a small thing that you can append on the end of your Google search URL to stop it doing this, which I have my browser automatically do now. But it's annoying to do that. And of course, that's the idea. You know, it's if you make something the default, it is what most people will use. And most people will now see Google in this in this form and it will serve them information in this, you know, potentially problematic way. So it's a concern I would have for information access, for knowledge, and especially, like we've said before, for younger people who aren't as cynical as we are growing up in the much messier days of the early internet. I really do worry about the way in which they come to view and understand the world as it's mediated through these kind of sycophantic tools. So do I. And perhaps the way by which you can bypass that Google feature, something you will include in the show notes. I will absolutely do that. Yes, I'll put a little how-to in there. but just as you mentioned the paradigm there reminded me of a conversation i had recently with someone who works in politics works in kind of um who works at kind of the front line of public services let's say and a separate conversation i had with someone who works in the kind of public communication side of a media organization and those parallels to the conversation where they're now seeing be it either um applications for funding on the one hand or or kind of complaints or outreach from the audiences on the other hand that are again as you mentioned they're tropes of AI material that are very clearly being either created or assisted in the creation by generative AI and I just wonder as we kind of move forward so quickly into the future are we prepared for perhaps how public consultations perhaps any form of public service that requires public input like that maybe misuse might be might be the wrong word here but essentially will be now generative AI content and submissions will now be a part of that and on the other hand I'm under no illusions that perhaps public services or other organizations we all we all know that organizations use AI systems for hiring as well now we may be now moving into a space where there's AI talking to AI and that it's us humans who are at the kind of end result of a review or seeing what the what the net outcome of that might be and it just it just i don't know it just it just strikes me as curious as to how we're so quickly moving into this world without the kind of what feels like the necessary due diligence or due consideration for what we're giving up in the process as well i think yeah yeah i agree i've had the same worry for quite a long time because it was very acute for us in in the university space from early on that we could very rapidly end up in a place where students would use this for the submission of their work and then academics if they were lazy would use it for the correcting or the grading of that work and so you have exactly that same you know AI writes the essay and then AI grades the essay and then there's people on either side of this who have little involvement in the actual process over time and then of course what you describe is it's naturally going to happen with CVs and recruitment. It's going to happen in tender processes. I'm involved in a funding review at the moment and quite a huge amount more applications have been made this year. And the applications that have been made have the linguistic characteristics of things that LLMs were involved in. And so it's very clear that there is going to be use in this. And it's partly driven, I think, by a fear of missing out that people, I certainly have done focus groups with students and ask them why it is that they might do something as silly as to using AI. And often it's because, well, they're afraid everyone else is doing the same thing. And if they don't, they're actually going to do worse, which is of course not true. They're much more likely to develop a skill by, you know, taking the time to write the essay than they are by handing it off. But the perception is that this is something that you're stupid not to use. You should be taking every advantage you can. And I understand that. That's a really, you know, tangible and compelling argument, unfortunately. And I think that's why it's infiltrating so successfully. And it's linking back again to what we mentioned earlier. It was very cheap until recently, at least, to get quite a lot of value for very little money or indeed even on free models from some of these kind of services so that they could do a lot of work for you. So I think we're really in a space where I agree very strongly with you, where our humanity is something we need to really bring back into this and the effect on us as people and people in relation to one another is something we need to talk about. And we're in good company bringing this up because a man with a very big hat and much more opulent clothing and apartment has brought this up because the Pope himself, no less, has written about this and is talking about exactly these issues. Yeah, as you mentioned humanity there, he, being Pope Leo, of course, chose to publish an encyclical titled Magnifica Humanitas which translate to Magnificent Humanity and as you were speaking there about the kind of transition towards and kind of growing dependency and we've framed this episode around the cost of AI we could also perhaps if we were being slightly alarmist to phrase it around the increasing slavery of AI which we and the services institutions around us become greatly perhaps enslaved but it was something that was picked up on in this encyclical by Pope Leo as well. And why this encyclical is so significant is because the last time a Pope, also a Pope Leo published an encyclical was in the late, I believe, 1800s, around the time of the Industrial Revolution. And there seems to be clear parallels in how the current Pope Leo views this period of enormous transformation comparable to the upheaval of that period. But Pope Leo took a very critical tone on artificial intelligence within this encyclical, saying that it needs to be disarmed, that it is seen as a present danger and is seen as a real risk for society as well. Yeah. And I think what's really interesting in it is that a lot of the perspective he's taken is quite a kind of democratic one. And it's about the concentration of power, which is something that we've certainly talked about before, which is something that, of course, also is directly connecting with that previous encyclical that you talked about, because it was written, you know, in that gilded age of robber barons taking over society and accruing wealth in these massive ways that we now see this small handful of people who are leading these huge AI companies doing. So, you know, the Elon Musk of the present is the, I don't know, Vanderbilt or Leland Stanford of 150 years ago. And so the pattern, I think, is the same. And I think the Pope is very deliberately connecting it with that prior period because a lot of the themes need to be readdressed in his mind. And I find myself in the bizarre position of really strongly agreeing with the Pope. This is something that I've surprised myself on because I would not have thought that certainly 10 and 20 years ago in Ireland that we would come around to a point where I'm so on the Pope's side on this. But I think it is really well written. It's an excellent document. I was reading it over the past few days and I think it's accessible. It's very long, certainly. So, I mean, I understand these days that nobody wants to read terribly long things, but it is worth really having a look at even some of the summary of if you haven't done so, because the themes in it are important ones to talk about. There are things that we've brought up here many times in terms of not just the issues that you mentioned there about, you know, the kind of accruing of power and wealth and the problems for society, but wider things around how it is that we define ourselves as humans in regard to one another, how we give meaning to our lives. There's a really interesting question about the fact that work for many of us is something that is vocational and something that helps define us and provide us with happiness by giving us a way to relate to the world, to make change in the world. And if we end up in a place where there's massive job losses, well, what does that mean? Even if there is a universal basic income and you're paid to stay at home, you lose a source of meaning, you lose a way of relating to and meeting people. And I think those sorts of things are a kind of redefinition of what it means to be human, that it really is something that demands some care and attention and some deep thought. And we're skipping past that a lot of the time to talk instead about profit making and not being left behind. We are. We are. So I think that that is a after our after our collective review of the various costs of AI, I think that is probably a good place to leave for today. I don't think we can top Pope Leo on the podcast for today. Well, we'll certainly include a link to his encyclical. So listeners can take some time to read over that and to review some of the other things that we've mentioned in this particular episode. If that causes you to come up with some important ideas or indeed have some questions that you could ask us, please definitely get in contact. hello@enoughaboutai.com. Thank you. Thanks very much.