135 – Resource Constraints and Technological Innovation

[00:00:06] David: Hi and welcome to the IDEMS podcast. I’m David Stern, the founding director of IDEMS and I am delighted to have Johnny McQuade, one of our Developers, somebody who’s been behind the scenes on the podcast since it started, doing his very first episode with us. And so hi Johnny, how are you doing?

[00:00:29] Johnny: Hi David. Yeah, it’s good to be here.

[00:00:31] David: Yeah, at last I’ve got you discussing something and I don’t even know what we’re going to discuss.

You have an idea, you’re keen to discuss, so I’m going to hand over to you and find out what we’re discussing.

[00:00:43] Johnny: Yeah, I think I must be one of the people that’s listening to the podcast the most, because I help Santiago with editing. So yeah, very much behind the scenes of the podcast, but excited to be here and actually discuss something and hopefully something people will be interested in. This was, something I was thinking about following the discussion you had with Lily on the podcast last week, about the Paris AI summit.

And I’m very concerned about the resource usage that these big AI pipelines can use. That’s a great podcast called Tech Won’t Save Us that have been doing a good series about data centres and their impact on general kind of energy usage, but also specific local resources like water is a big one that they can be a big drain on the resources locally where they’re built.

And a few weeks ago, there was this big stir caused in the industry because the Chinese company DeepSeek AI released a new model, which seemed to perform just as well as some of the big US models, but both in terms of training and responses was orders of magnitude cheaper in terms of cost and also resource usage.

And these things are always a little bit of a grey area. It’s a bit hard to know exactly how much like they’re being subsidized or what resources are being used, but certainly seems to be, significantly different kind of scale of resources. 

And one of the interesting things about this was that they seemed to do it in spite of not having access to the latest hardware. The GPUs that are most commonly used by the big U. S. companies were banned from export to China. And so it appeared that they’d achieved this amazing feat without using cutting edge hardware.

I think in the context of the scale of money and resources that are being pumped into the big AI companies, there was this figure of something like 400 billion or something that the US tech industry was saying was required for levelling up their AI capabilities.

This is very significant that it could be done on a much different scale. And I think what I wanted to talk about was the extent to which DeepSeek achieved this in spite of these hardware limitations versus the extent that it was because of those limitations. So really how technology working within limitations, be that, I suppose hardware is the main one, but any kind of restrictions can actually be more productive or make more progress.

[00:03:05] David: Absolutely and I think this, of course, is a recurrent thing on the episodes that a lot of, My experience is working in very low resource environments and I found consistently that actually you approach problems very differently when you have resource limitations and actually I believe you can build much more efficient systems and much better systems from those restrictions rather than the restrictions holding you back from what you can build.

So let me speak very briefly to DeepSeek specifically for two reasons. One is I know this is an episode Lily and I are going to have in more detail later. And the second is that, I feel that this is illustrative of something which I think is more general, which you’ve hinted at.

It’s this sort of resource constraints can actually lead to better efficiency to building things which are more sustainable and so on. And I’m actually going to link that back to the sustainable development goals eventually. So basically, I am not surprised at all at what DeepSeek have achieved.

I’m impressed, I’m very impressed, and I’m really happy that it’s happened, but I’m not surprised. The inefficiencies within the current AI systems and AI models, because of the amount of money being thrown at it and resources not being the limitation, mean that nobody is really thinking hard, I would argue, about doing so efficiently, because everybody needs to do it quickly to stay ahead of everyone else.

And so the fact that a competitor who has emerged, which is thinking about efficiency, because of resource constraints in different ways, is achieving similar results more efficiently. This is expected, the mathematical innovation, which is happening here. People are wowed by AI as this magic bullet, but it’s not. It’s mathematical approaches which have been applied with lots of data and with huge amounts of computing power. Any problem which is using computing power, it takes time to make it efficient and effective. It does not surprise me at all that we’re going to eventually get to a world where we don’t have these, ever increasing energy and data sinks in different ways where things do become more efficient, more effective at some point.

And it’s wonderful to me that DeepSeek is an instance of that happening. And that’s the more general process which I’m really keen to dig into. As you say, there were elements of resource constraints, which, are associated to that advance. And there’s a question about whether the advance was achieved despite those constraints or because of them.

And that question is fantastic. And my hypothesis is more because of them. Because within those resource constraints, you have to think mathematically differently about how you go about the problem. And therefore you are actually innovating in ways which I think are needed. So my hope is there’s going to be much more of this.

There’s going to be a lot of this coming from other corners of the globe and that it does mean that it’s not just about throwing money at the problem. It is also about thinking about the problem differently. And don’t get me wrong. Money was thrown at DeepSeek. This did not happen without substantial effort, substantial finance behind it, and particularly substantial human resource.

That’s the key thing, but it does not surprise me that human resource in a constrained environment was able to outcompete human resource in an unconstrained environment.

[00:07:23] Johnny: Yeah, that makes a lot of sense. And I think it is as you say, it’s not necessarily surprising. I think in all kinds of fields and particularly creative ones, constraints and limitations are acknowledged as being very important for progress and, yeah, creativity. One of my favourite artists is Brian Eno, the musician, he talks a lot about creative constraints.

There was a quote I found that, it was attributed to Orson Welles, who says the enemy of art is the absence of limitations. That indefinitely in art, you need these limitations, otherwise there’s too much you can do and you end up paralyzed by that. And I think it’s potentially, or it may at first glance seem surprising, but the same applies in technological fields.

You might think at first thought that technology and software is about making use of resources. And if those resources increase as they have been doing massively over the last few decades, then we’d have so much more power that we can make use of. And there was something I came across in just researching this about there’s a economic law called Jevons law, Jevons paradox that says that as I understand it, as technology advancements mean that efficiency increases on a given resource, then what actually happens is the demand increases so much that the resource ends up being, the total resource consumption increases, even though the efficiency has gone up.

End up more than adjusting to, to fill that gap. And I feel like there’s a case to be made well as a user of software, I experienced that, the amount of RAM that my, that is used by the apps on my computer is enormous and, it will be gigabytes and gigabytes, which wasn’t available a few years ago.

And that will be just to run some, a few Chrome tabs in the background or something. And I think there’s, yeah, software isn’t being written with these kinds of constraints in mind. And you wonder where we would be if people were writing software in such a way that it was still, really minimizing memory usage and things.

You’d think things could be much more efficient than they are, but because this is available and we can use all the cutting edge hardware, then it’s a lot easier to write software in that way. And I guess some of that is about the efficiency of writing the software, which is different from running the software where there’s tools like Electron, where you can write something in one language and then have it kind of, available on all these different platforms.

But I think this made me think about the resource usage of these big AI pipelines and how if that, if Jevons law applies there as well, that, resource usage will actually increase, then there’s a kind of pessimistic view that, um, the US tech firms are still going to be wanting this 300 billion dollars.

It’s not as if they’re going to say, ‘oh, in fact, we can do it more efficiently’. They’ll still be using the resources, the data centres are still going to be built. And I think, yeah it’s not as if a point like this means, um, the environmental impact is suddenly lessened.

[00:10:21] David: Yeah. Let me come into this because I think you are absolutely right in many ways, but I want to come in on on exactly the point about your, the number of tabs you have open in your browser. The fact that we have powerful machines means that we don’t need to close our tabs quite as much and quite as carefully, and the developers of the website doesn’t need to be efficient.

We, I don’t know if you remember the IOGT project, they did some work on this where actually one of the government ministries, I want to say it was Egypt, but I can’t remember which implementation it was, did some tests and they found that basically compared to their government website, which was just built standardly, IOGT was using less than one twentieth the amount of data.

[00:11:10] Johnny: is it worth quickly explaining to listeners what IOGT is?

[00:11:13] David: So yes, thank you. It’s the Internet of Good Things. This was work we did with UNICEF, where they wanted to bring people on low end devices onto the internet for the first time, and so they wanted this very lightweight platform so that people who didn’t have much bandwidth or didn’t have good phones would be able to always interact with the system.

And so we built this and the government was really interested in this partly because it meant that the load time went down. When you’re in a resource constraint environment where bandwidth isn’t great, then actually the load time for a web page is a big deal. And so having a web page which uses less than a 20th of the bandwidth is huge.

And this is something where, I wish we could be working more on this in the future, because my hypothesis is the following, that if you work only in high resource environments, you won’t get the incentives to actually build out these systems which are more efficient, more effective, exactly as you say. But poverty is increasing, or it’s not decreasing as it should, that’s the sort of evidence from the latest reports. Depending on how you measure it. And hence, broadly, they say that there’s been a lost decade in terms of reducing poverty, eradicating poverty.

And if this is something where we actually want to take that seriously, building technology for that low resource environment is something that I care deeply about and that I think is needed. And my hypothesis is that actually that technology, once it exists, if it’s all open source, if it’s all built well, and if you’re actually then using it and it’s really tested out and fleshed out and built up, with this efficiency in mind for low resource environments, my hypothesis is that will out compete, just as DeepSeek might out compete the big American tech, the big American AI, even though it’s coming from a resource constrained environment.

As we say, that can be a competitive advantage. And there is absolutely a train of thought and a set of theories, economic theories and others, about how the world works based on a single high resource environment. But that’s ignoring the power of actually a proper global view, which values other environments and the advances that they can bring.

And exactly as you say the benefits you can get from designing for a constrained environment. This is where I wish more people were thinking about that rather than looking to AI, to say, if we’re going to use this in, this is going to be useful all over the world, to actually think, no, what does it look like if we get people redesigning it in low resource environments, more suited to low resource environments?

What if we actually had a whole set of people all over the world interested and engaged in that question, in those problems? Now, relatively speaking, that’s cheap, because almost by definition, you don’t need the same resources, you just need the human resource. But the point at the moment is a lot of the human resource is being pushed towards the needs of the AI. The, if you want the whole data science world has got sucked into the needs of big tech.

Whereas actually if researchers had different incentives, many of them could be interested in researching and understanding efficiency, building for these low resource environments, understanding constraints, building for constraints. My hope is, this is part of what the AI summit, just recently in Paris was hinting at. It was hinting at the fact that the EU and Europe is looking to take a different approach.

It’s not looking to compete like for like. It’s looking for transparency. It’s looking for these constrained systems in different ways. Now a lot of its constraints are different. They’re ethical constraints. They’re all sorts of other things. They’re resource constraints, for environmental impact.

There’s all sorts of other constraints they’re putting on. But that’s creating potentially an environment where there will be a whole lot of people, working on these constrained problems. I’m excited, I don’t know what this is going to lead to, but I think that this could lead to, this could lead to a different sort of innovation.

I don’t think the US, the UK even, there’s good reasons they didn’t sign the declaration or whatever it was called. Because they’re on this other route and they’re sucking up the funds and the money for that. And that’s the growth path they’re on, that’s the bet they’re making. It may win, it may not.

But I think it’s exciting that for the first time I’m seeing this differentiation coming in. People willing to think and my guess is the EU is going to put money behind this to actually build AI differently to think about it differently. So I think there’s a, I’m not, I’m an optimist. I’m always an optimist.

[00:17:05] Johnny: Yeah, I think the, it does feel like at some level technology created with constraints, it’s not just that it ticks these boxes and so it’s, and so it’s morally better. I think at some level it needs to… it needs to out compete it from a user’s perspective or, and I think you’re right. Even when those constraints are no longer there, it still can, win out on the terms of the technology itself. One example of that, you know, quite a minor thing, but I’ve been, I’ve spent a long time searching for a good note taking app, and, my requirements are quite simple really, I need it to sync between my phone and my computer so I can quickly access it wherever I am and it doesn’t really need to do much more than that.

I like Markdown, so I want it to support Markdown if I want to, for when I’m writing bullet points or things but I used to use a, an app and, over the last few years, it just got more and more bloated with features and things I didn’t use and it slowed down and I think I suppose it’s more about the kind of commercial needs of the business.

They need to keep growing. They’ve got shareholders to answer to. And so they, they need to keep this growth. But it wasn’t, that wasn’t what I needed at all. I didn’t, it didn’t suit my needs. And I think if they had the restrictions of poorer hardware or yeah, kind of resource restrictions, then they wouldn’t have needed to essentially ruin something that was working perfectly fine.

[00:18:35] David: Let me take that example, because I love it, because it’s something we have seen in many different cases, where, this bloating of software, because of the nature of how software is commercialised, being something which is not always leading to a better user experience. Now, I’m sure there were some users who like the features which are being added because otherwise they wouldn’t be good for business, but the fact that this is something where that’s not improving the user experience for all users is very interesting.

And this is exactly where I believe if you take open source systems, there is a potential to develop that differently. But currently the business models for open source they are competing in certain ways, but they’re not competing in others. And so there’s been the move to open source as led, I would argue, by Red Hat years ago, showed that open source can out compete on a level of playing field.

But it didn’t start by out competing, it started by people with a vision who believed in the mission and then in very specific contexts it out competed. Servers pretty much anywhere will run Linux now rather than Windows and it’s it’s an obvious choice, and part of that is that, you actually, you don’t have that many users on the servers needing to know how to use the server software.

So the fact that with Linux, it is lean it’s highly customizable and all the rest of it, it’s the right fit for that. I believe we are missing some of the, and I believe they are social business models, that will out compete. For other types of software.

And your note taking app is a really interesting example. I think that’s something where, the right business model on something open source can and should be able to outcompete. I don’t yet know what that business model is because I’ve seen and I’m looking at all the different groups trying different things and I don’t see the business models which are going to outcompete yet.

But I think that’s what we need to be able to get innovation. And I think it’s interesting that innovation for software, and I think this is going to come in with things like the AI models. That innovation for software does relate to open source, but it’s not just about open source, because it is around the business models around the open source.

So if you take open AI models and systems, again, it’s going to be, are there the right business models to enable them to out compete on a level playing field? More than, technologically, is open source going to out compete? That’s my hypothesis. And I don’t know there, I don’t know how that could happen, I don’t know, this is, I think there needs to be research into this.

There’s a lot of instances individual want to learn from, but there aren’t systematic ways to learn how to do this. So I do think that this is one of the reasons that I’m so interested and engaged in the social enterprise communities, because I don’t think the innovations that are needed now are purely technical.

I think they’re also social, and particularly about social businesses.

[00:22:25] Johnny: Do you think there’s a, so I work at IDEMS mostly on the open app builder project. There’s an open source tool for building apps. With a focus on low resource environments as we’ve been discussing. And one of the strengths of that project as I see it is that it’s not just open source, is that there’s also a degree of customizability that we want to expose to people that wouldn’t necessarily require full coding knowledge, but would mean it could be adapted to their needs.

And I wonder if that kind of relates to any particular business model or if it’s, or if that is just more of a software model. In the case of my note taking app, if it was, I don’t really want to get stuck in and start coding things myself or working on the open source code base at all.

But if there was an ability to just, or with a tick box, turn off some features that I didn’t want, that would be great for me. Or if I could, if I could customize it in a straightforward way, that would be great. And presumably, sometimes it’s a business model thing, but there’s, some of these features are important for revenue streams or things, so they don’t want users to disable them.

[00:23:23] David: I think there’s a good question there. I don’t have a good answer. I do believe that customization and that customizability in different ways is really important. I don’t know, that we’re going to get the business models right on that and so on. If we got them right already, then things would be different in different ways.

I do think one of the technology that we’re building is a long, there’s a long way to go to get it to the stage where I’d be really confident getting it out really widely and getting it to the stage where we could be having or trying out different business models on it. But I think, even then, we have hypotheses around what those business models would be like. It’s this opt out of payment rather than opt in is one of them. And that’s something which we learned from organ donation on driver’s license applications. This is very effective. And so that’s something which I believe I haven’t seen much.

I think Moodle, the Moodle app is one of the few places I’ve seen that explicitly done. So there are a few places that have already thought about that and already implemented something like this. There’s a lot of things around this where I think there’s innovation to come. But the simple answer to your question is I don’t know yet.

These are things that I hope to be working on in the future and involved in and engaged in discussions around. But, there’s a hard problem here, which I believe hasn’t yet been solved, and if it is solved, I think could have really big ripple effect.

[00:25:05] Johnny: So anything else you’d like to say, maybe relating back to the hardware limitations or limitation technology?

[00:25:12] David: I guess the last thing that I’d like to throw back to you on this is that you started by mentioning the environmental concern of this, and I want to I don’t want to trivialise that, but I do want to, I want to challenge the sort of common perspective on this. That is a problem and it is something which does need to be addressed, but it needs to be addressed in so many different ways. You can address it by changing the hardware so it doesn’t need to use the same resources, by changing the resources it uses so that it uses them in environmentally friendly ways.

You can reduce the resources that are needed to be used. All aspects of this problem are things that I’m interested in, but not worried about that as a problem because I believe all of those directions have so much room for improvement right now.

There’s a big question about batteries, which I won’t get into, but I fear there’s so much room for improvement in terms of efficiencies of the algorithms, actually getting work to make the algorithms more efficient. At the moment, they’re basically stupid. They’re extremely powerful at being stupid, and so they work really well, but they’re so inefficient.

If you understand the maths behind them, in terms of where maths is, this is elementary school mathematics they’re at, basically, it’s not quite true. I’m an algebraic geometer so I do have a slightly different perspective on this. But, it’s not really, there’s no efficiency built into these systems at this point in time.

It is just both, it’s the fact that the compute’s got so big and you can just throw so much at it. The, may as well throw everything at it. There’s so much room for improvement there, and it will reduce the cost, and because it reduces the cost, at some point when the cost limitation becomes the issue where people want to improve profitability, they will care about that.

So that is something that at some point will get worked on. Is it too late? Maybe. But that’s something which will be worked on at some point, I believe. The data centres themselves, their environmental footprint, it’s very visible. Given it’s visible, people will be working on data centres being environmentally neutral, or even better.

That’s something where that is going to happen. I heard about these underwater status centres in different ways, which therefore reduce certain other issues, but then they had other issues with them and that got scrapped. So people are working on that. So I’m not worried about things where it’s so clear that we as a society can and should advance

I would be much more concerned about things. where there wasn’t scope for progress. And I’ll come back to the battery issue. For a long time, I just couldn’t see the technological advance that would actually, move that forward. More recently I’ve seen there’s a few things which have come out where it might be possible to get away from lithium ion.

Fundamentally, lithium ion batteries have taken over the world in different ways, in a way which is clearly unsustainable, and there is no alternative, and our need for them is just going up and up. And so that need to shift away from that as a technology, that’s one which concerns me.

That’s one which I’m really worried about because it’s not clear. You can’t do a similar process more efficiently because, by definition, this is the best chemical element you could use for this. So that, that’s the sort of thing I was really concerned about. I’m not concerned about data centres. Data centres need people working on improving them, but these are solvable problems, is my personal view.

I want to come back to check with you, is that, does that resonate, or do you feel, no, I’m missing something? The fact that these could be worked on doesn’t mean they are being worked on or they will be worked on, and actually the scale at which the increase is happening is so much faster than the progress that’s being made on working on them, we should be worried about these.

And it’s possible that I’m wrong but…

[00:29:32] Johnny: Yeah, I think for me, we probably don’t want to get too much into politics, but I feel like for me, as long as the main driver is the profit motive, then we have seen consistently that humanity isn’t able to self correct to the extent that is required to support the natural world. And, the ecological collapse that’s been happening over the last century is testament to that, that even when efficiencies are made and, obviously it is within these companies interest to make things more efficient and cost less, which comes with using less resources.

That is built in as a motivation. I think we as consumers, we’re able to turn a blind eye to a lot of terrible things. In the supply chain of all kinds of industries, like the meat industry, say, it’s not the case that people voting with their wallets, quote unquote, will always, push companies in the right direction to ensure that things are protected. So I think maybe I’m just more naturally pessimistic than you. And I think, motivations there just aren’t big enough to restrict these companies, especially these huge corporations that, that have, seem to have shown little regard for the environmental world around us, but also for people’s mental health and, the brain chemistry of children that I’d be surprised if those were issues that could be solved by, very, by brilliant people working hard within the current system.

[00:31:07] David: Okay, Let me come to you. And I think I absolutely agree on the timescale that you’re referring to. And I think this is, I think, a fundamental difference of our optimistic, pessimistic perspective. In the short term, I think you’re absolutely right. However, I think there are signs in our society of instability. which are coming from resource limitation.

And so my hypothesis on that is, and I don’t know what timescale this is happening, but is that if you think about this mathematically, these signs of instability, happen when you’re close to hitting a resource limitation and the system’s going to have to break. The exponential growth that we’ve been experiencing for over a century, I think the instability we’re seeing in the world with the polycrises that people talk about.

I think they are signs that we’re hitting some form of resource limitation. As you say, it could be the planetary resources that we’re, we’re hitting going over tipping points and thresholds, or it could be actual more practical resources in other ways. And I think that my optimism comes from the fact that what has been is not the same as what will be.

And this is always true. So I absolutely hear and I see what has been. And there are elements of that which, I cannot see how the current system can correct some of these things that have been. And it’s not clear that what will be is moving in the right direction. I learned, I talked about the fact, I wanted to come back to the Sustainable Development Goals just to finish off.

I talked about the fact that one of the core Sustainable Development Goals was zero poverty, zero hunger was another one, but zero poverty. And broadly, people are talking about the lost decade, the fact that the last decade really hasn’t, no progress has been made on this, on a global scale.

And, what’s interesting to me is the power of the sustainable development goals was this recognition that everybody had to move: high resource environments had to evolve as much as low resource environments. And the diagrams I loved for this were showing that the low resource environments, they’re not going to go via where the high resource environments currently are.

They can get directly to a sustainable future. That was the hypothesis put forward, really, and this is what was so great about the Sustainable Development Goals, is it wasn’t saying that, oh, those poor countries need to develop to become like us. No, they were saying we all need to change. We all need to evolve.

And everybody’s going to be on their own path to evolve to something more sustainable. And that’s what we’re all aiming for. And that vision to me is when I got really excited about saying okay, what if we work really strongly with the low resource environments to show a model of what that sustainable future could be. And that’s where we get these options by context, these different perspectives of how we get to a sustainable future. So I believe very strongly that taking a high resource environment, I hear you, but taking a global view, I think if we look at that sustainable future, if people start getting there from a different perspective, from a different root.

Other people will take note. They’ll have to, because otherwise they’ll be out competing you. That sort of, if you have the same standard of living, the same sort of things, but you’re much more efficient as a society and you’re working much better of course you’re going to develop faster in the future.

And so these societies aren’t potentially going to come up in ways where if high resource environments don’t take note, they may be losing out and missing out. It is why I wanted to come back and finish with the Sustainable Development Goals, because this broad view, the world view we have from a high resource environment is the word view, I believe is fundamentally wrong.

We need to look at it from many different perspectives. And I’ve learned so much from looking at this from the perspective of really low resource environments. And I think that’s why I’m optimistic. If I was only looking at what’s happening and what you could achieve in high resource environments, I think I’d be pessimistic.

But I believe in the human capacity. More people are being born into low resource environments than are being born into high resource environments. This is where the human capacity of the future is going to be. This is where innovation could be coming from. If we can get the right education systems and so on. I’m optimistic because of, not despite, the low resource environments.

[00:36:45] Johnny: Great. Yeah. I think that’s really illuminating. I’m sure we could discuss, this a lot more, but we’d probably better wrap it up there.

[00:36:52] David: Yeah,

[00:36:53] Johnny: Yeah. Thanks, great talking.

[00:36:55] David: And great to do an episode with you. Let’s do more, look forward to discussing further on whatever topic you want.

[00:37:01] Johnny: Brilliant. Yeah. All right.

[00:37:03] David: Thanks,