12. The Power of the AI Narrative

[00:00:07] David: Hi, and welcome to the IDEMS Podcast. I’m David Stern, and I’m here today with Kate Fleming, a fellow director.

Kate, this is our final episode in our long series now relating to the Empire of AI. Looking forward to it. What’s it on?

[00:00:25] Kate: The thing that I think we have to talk about that is under lane, is behind everything we’re talking about here, is the power of PR and messaging. In one of the early episodes I talked about just the renaming of automata studies as artificial intelligence, how much work that did to set in motion a lot of different things.

And I think what we see, and Karen brings this up, you go back to 2020 through 2022 and what you have is a very organised – I think OpenAI is behind a lot of this, that’s her focus but I think other companies I’m sure were involved in this – this very organised lobbying, messaging, PR machine that is going out and capturing the narrative of what AI is, what it should be, the inevitability of it.

She tells the story of basically, by 2022, you have in the US senators, congress people, they’re just calling OpenAI when they get stuck on like, oh, I don’t understand this technology, tell me, Sam Altman, how does this work? Well, I’m guessing he had some very particular ideas on how this works and what they need to understand and here’s the policy we need.

And you see how they’re playing, the PR machine that is playing on geopolitical concerns, I would argue underpinning a lot of assumptions, about AI is a race that is US versus China. I think you see a lot of political forces there, and I think a lot of the conversation has been around AI, you see this bubbling up where, when Deep Seek came out of China, it wasn’t just like, oh, here’s another interesting AI – and I’m not even gonna get into the politics and whatever – but it immediately was like, how did they get that and what’s happening here?

The assumption is, oh, OpenAI, these are the good guys and this must be the bad guys, they’re all these forces at work that are shaping the messaging. So what I really wanted to talk about is: how do we counter that? I guess what are the forces there that we see at work? Yeah, and how do we start to create a force that’s different, that’s creating different messaging, where people come to us and ask, I always forget their name, Hugging Face, the open source models, there are all these different groups that are like, why don’t you ask them?

And so that’s what I wanna talk about is: what is that power, how can we change it? All of that.

[00:02:55] David: How do we liberate AI from the AI empires? 

[00:02:59] Kate: Yes, in the popular conception and in the political policymaking realms.

[00:03:05] David: And of course therefore in the research agendas, because we’ve discussed that as well in the past, that it’s taken a particular route, which is serving particular goals. So it’s a really good question. I, of course, don’t have the answer to that, I don’t know. If I knew, we’d do it.

But I guess the first question is to even recognise that really what Karen has articulated in her book is that there is a need to liberate AI from the AI empires. First, you need to recognise that these are AI empires, which is what she’s done, and she’s articulated so well. And then once you recognise that these are the empires of AI and you recognise that AI is just a tool, a tool is nothing in terms of actually what it can do, it’s just a tool.

And so maybe let me rephrase that, a tool is nothing more than what it can do. It could be used for many different things in many different ways, and what it is, is ill defined. So the idea then of being able to articulate that we need to liberate AI from the AI empires is maybe just a narrative that needs to get out more widely, more people need to take up and adopt and actually then adapt and make their own, because the key is that the AI that we’re looking to support is diverse. It isn’t a monolith, it has many different forms in many different contexts, it serves many different purposes, it’s owned by many different people and many different things.

So recognising that that popularisation, not in terms of simply the use of it, but what it is, that’s what we need to get so that people can say, well, this is AI as well, and so is this.

[00:05:01] Kate: Yeah. And I think that you are identifying what the core of it is that I feel right now, like a lot of activism against AI is only reactive, it doesn’t have a vision for alternatives. It is just, we don’t want this data centre. I agree, I don’t want it either, but I’m also engaging with, well why are we building these, what are the problems people think, what do we think is being solved here, why do we think we need these, why is everyone supporting them, and how might we build things that look different where now we don’t want it because we don’t need what you’re building?

It’s not that we don’t need the AI, we don’t need more of the AI that does this. We need more of this other kind of AI that doesn’t need this. Or this AI isn’t even the right tool for these problems that we’re trying to solve as a society. So I think it’s that where it’s equipping people to not just be standing in the way of progress, which I think is the Luddite, what Luddites have been lost to history for, where they recognise that industrialization was taking away their jobs.

They were fighting against that, but they didn’t yet really.. Actually, I’m not even sure this is true, I should ask, we should have a Luddite historian on. I think they might have actually had some ideas of how the wealth of industrialization might get spread out better and serve people and not be so exploitative. So they may have, even in the historical telling, been reduced to less, you know, visionary reactivity than they were.

But anyway, so yeah, I think it’s how do we avoid just seeming like you’re burying your head in the sand and in the way of progress?

[00:06:41] David: Let’s take that very specific example you’ve got about data centres. The simple route is to say we should oppose data centres because they’re not environmentally sound, they’re taking away water, energy, they’re disrupting society and ecologies in ways which are not positive, and so on.

I’m not saying that that fight is bad, but I’m saying what you are articulating is that fight is a losing battle if, as a society, we accept that we want to use the AI which is coming out of the AI empires, and that AI requires more and more data centres to be able to run. So if a data centre doesn’t get built here, it’ll get built somewhere else. So the bigger problem and the bigger question is, do we need those data centres in the first place?

Now you cannot, at this point, simply have the funding shift for big tech, but there are more and more cases. There was an Nvidia researcher who wrote this very nice article about small language models, which articulates the fact that actually the big large language model work which is happening now, which is so pervasive, which is leading to all this data centre work, is probably gonna be short-lived. Because as we actually make progress on the mathematics and the implementation side, we will be able to do the same with less.

That is going to be part of the advances that happen, and when we do the same with less, it is now very likely that a lot of that infrastructure that is getting built will never actually get used if doing more with less, if those technologies come to the forefront.

I would just give my favourite example of this, that was, I believe, the transatlantic cables that got, when the first companies were really getting stuck into this, they became hugely wealthy by increasing the bandwidth of the transatlantic communication networks, and then suddenly the whole thing crashed because, well, there was more than enough for the capacity. Simple supply and demand.

The data centres have the same, simple supply and demand. It is not clear that as the supply goes through the roof, the demand will keep growing if the work goes into saying, how do we do more with less, or even just the same with less.

[00:09:27] Kate: It’s so interesting, that cable example. I think we might’ve read the same book, which is that “Good Strategy Bad Strategy” book, I think we’ve done an episode on that, but it is that, right now, the strategy in big tech is to create this inevitability. It’s more and more, the strategy is you’re trying to corner the market. But I think what you’re arguing, and I would agree with, is that it’s actually a really bad strategy because they’re not gonna be able to deliver on it.

So, at some point, this is all gonna come crashing down. Those data centres, if they do get built, are they really gonna get used? Well, I guess they are, and this is part of the power of PR and I would say marketing and messaging. But companies are like, you have to buy this, you have to have this, and it is useful in a certain way, but pricing is going from, well, this is $15 a month for one user to, it’s now a thousand dollars a month for one user .

So you see this strategy that is unified around creating inevitability, creating scarcity, but actually it’s a very fragile strategy. Everything we’re talking about is there’s no inevitability here. 

[00:10:36] David: Let me just go a little bit further. Very concretely we are using these AI systems to do certain pieces of work, but all the work, the methods that we are doing for this, we are doing it in such a way that, if and when the small language models actually are able to do the specific bits better, we’ll just swap them out.

And this is actually, the more people go towards agentic AI, more and more should always be benchmarking against the latest models versus the smaller, more cost-effective models. And my hypothesis is that, that swap out, it’ll start happening and it’ll start happening fast. And the more the companies start increasing the prices, the more the incentives will be there to build these smaller models, which are just as effective for certain tasks, and the more things will get swapped out and that’s when they’re in real danger.

[00:11:29] Kate: Yeah, and it’s funny, even as you’re talking, I feel like maybe I haven’t said it enough on this podcast, I love a lot of things about AI, I use a large language model AI frequently. It’s just that I recognise, exactly as you’re describing, I use it for these very narrow use cases often when I’m using it, it’s actually not nearly as expert as it needs to be. So it’s helpful to a point, and then diminishing returns.

And this is where you made the point earlier that the work continues for skilled labour, because you recognise pretty quickly, well, thanks intern, okay, I guess I’ll take it from here, you know, it’s like there’s all this other work. So it is that if you are more skilled, you really are wanting that much more specialist, I just want domain expertise, I want you to be reliably great at this work. And you can see sort of how all the forces, they’re not pointing to what currently has market capture in the popular and political conception.

[00:12:33] David: And this is something where this is recognised by pretty much everyone at the forefront of AI. I loved a quote – I think I mentioned this in a previous episode already – but it was only a few weeks ago that San Altman made this quote where he said: and you may be tempted to do more with less, but no, you should resist that temptation, let’s shoot for the stars.

This narrative around what can be done with less is a growing narrative. There’s real evidence around this, in fact, this is exactly what good mathematicians want to be working on, because that’s the interesting maths to understand how you can get the same performance with less and so on. This is exactly what is natural as the next innovation space.

[00:13:22] Kate: Yeah. And it’s so interesting ’cause even as you’re talking, I think what I hear in so much of this is they are treating AI as if it’s just more software as a service subscription innovation. But really, what they’ve created is infrastructure. And that infrastructure you just, you can’t capture it and gate keep it and have it be useful.

It’s like when I was in Nairobi, there are these Chinese highways that have been built, which are lovely, but they’re so expensive that the only people who take them are a few, the people who can afford. And you can correct me if I’m wrong in this, but I think it’s an example of infrastructure that people get priced out of that’s useful, but it’s infrastructure so people just really are using it to get from point A to point B.

And at some point it’s too expensive, it’s the strain, the load of it is just not serving its purpose and it’s exactly what you’re describing. You’re gonna build this small, lean, gets you from point A to point B cheaper and more efficiently.

[00:14:25] David: Yeah, I mean, we’ll see, we don’t know, nobody knows. But if I was a betting man, I would not be betting on the big data centres right now, I would be betting on small language models and what is it that actually will do this cost effectively, efficiently?

[00:14:43] Kate: And this is what I would jump in on. You just said, nobody knows, and that is the truth for all of this. And all we have set up here is an alternative vision of the future, which is just as valid. This is what I’m talking about with PR and policy grab, is that the big AI companies are trying to control what the future looks like by making it inevitable because every force has been captured and corralled to centre what they’re doing. And if you’re not helping it, you’re just not standing for progress.

Whereas our viewpoint is just as valid. Nobody knows, why can’t we hold this vision and why can’t we work toward it and why can’t we put resources to it and why can’t it be the case? The only reason the other vision is happening is because people did exactly that for their point of view. We need these way stronger forces of capture for alternative stories. 

[00:15:47] David: Of course what’s really interesting and important about this is that, when you talk about capture, of course what has happened is the capture of the money around it. So much money has been sucked into this, which is why that narrative being coming out is so easy to sell now, because it actually has so much financial weight behind it. You know, this is a real David and Goliath, if you think about this from a sort of PR perspective.

[00:16:22] Kate: Yeah. 

[00:16:23] David: The weight behind, the financial weight behind the PR machine around what one vision of AI and the future could look like is very different from, and we are not alone in having an alternative view on this, there are a lot of others out there, but there’s no other Goliaths, it’s very much a David and Goliath sort of situation here. 

[00:16:51] Kate: Yeah, I’m glad you’re named David, seems very, very appropriate. Yeah, absolutely, I mean, there’s not much more to say about that, but I think I just want to really call attention to that machine that is creating a story of the future and inevitability that is very disconnected from what actual experts say. We’ve cited that Nvidia researcher, but there are other researchers, it’s just that that example is top of your mind. There are plenty of people who are actual experts in this field who are saying what we’re saying.

We are not saying something that is unusual, it’s just that a lot of those people are not pundits, they’re not out, they’re not PR people, they’re just not telling their story in the same way and they’re certainly not organising, they’re often just doing research or whatever. There are probably a number of reasons why they’re not out, they’re not funded to be out there, but yeah, there’s a lot of force.

[00:17:54] David: But all you are really articulating – and it’s a really good place to end the series – is exactly what The Empire of AI book has articulated. These are empires, it is about the propaganda, the propaganda of empires. That’s an important piece of it, keeping the population under control, allowing the empire, the vision of that empire to grow as an empire. The capture, the proportion of society, the implications for what it means in terms of how it’s discussed, how you are allowed to discuss it. These are all things which we are not the ones creating this narrative, but, man have we been sparked by. I suppose first for me it was entering into that activist space, the Earthkeepers versus AI Empires, but really that is fueled by, and so much credit to Karen for articulating this so well in the book. 

[00:18:56] Kate: Yeah, I was going to end with, everyone should read Empire of AI by Karen Hao. It is just a great book and it does just tell this story, the arc of these forces, all of these different players in a way that is so readable that, yeah, it seems like a good place to end.

[00:19:16] David: And we should clarify, we were not paid by anyone to promote the book, but we have been influenced by it and it is helping us to actually find our voice. These are things we’ve been thinking about, we’ve known about for years, but we didn’t have a voice, we were just like the other people who are not getting their voices heard on this. And the book has given us language to be able to have a voice.

I do think we should finish by trying to liberate AI from the AI empires. I like that statement, it really does sum up what I think is the big effort where it needs a liberation effort. This is not an individual or a small group to do, this has to be a collective.

[00:20:04] Kate: Agreed. Thanks, David.

[00:20:06] David: Thanks.