The Cost of Compute: AI, Data Centres, Energy, and Australia's Future

[00:00:00] Speaker 1: All right, we are back with David Pocock and today we are joined by Dr. Sue Kay, who is the Director of the UNSWAI Institute and an expert in AI, robotics and automation. Data centres are the physical infrastructure that underpins a lot of our online lives, the websites we love, the services that we like. Can you give us an understanding of the difference between these data centres that we've had for a long time that underpin that infrastructure and these ones that are increasingly becoming part of the conversation that are going to help train and [00:00:31] Speaker 2: develop AI into the future? Well, we've had data centres for a long time ever since we've started to put things into the cloud because the data centres are actually the necessary infrastructure to be able to support putting files and things up into the cloud, but they in many respects have been, you know, not very noticeable because they haven't had large requirements for energy and water and land space. Whereas now what we're seeing with artificial intelligence is we're seeing the data centres actually having to host a large number of graphical processing units to be able to help train AI models or to do AI inference for businesses and inherently that uses up a lot more energy and water than before. Because of the scale of the undertaking of developing these AI models, they're also taking up a lot more space. And so they're having a much more visible impact on the Australian environment. [00:01:37] Speaker 1: So it really just comes down to that. The main difference is really just the scale. [00:01:41] Speaker 2: Well, and the fact that artificial intelligence has just taken off over the last five years, particularly since the advent of ChatGPT, when, you know, everyone has been able to access that AI technology on their phones. Whereas before it was a fairly nebulous concept and people really would not be thinking about where their data lies and what's being done with it. But now artificial intelligence is making that take a much more primary position in our thoughts. [00:02:10] Speaker 3: So if you, you know, back up your photos on your phone to the cloud, that's posted in a data centre somewhere. That's been happening for a long time. But a lot of the new data centres that we hear about are compute to actually train AI models. [00:02:25] Speaker 2: That's right. And we don't actually have the reporting methods to actually understand what type of compute they're doing. We can infer that from how much energy and water that they're using. But in general, there's no need for the tenants of data centres to tell us how they're using that compute. [00:02:42] Speaker 3: And we hear data centres talked in megawatt or gigawatt. Why do we use a unit of energy rather than compute power? [00:02:50] Speaker 2: Because energy is being converted into compute power, you know, within those data centres. So typically when a data centre was only supplying cloud services, it would maybe take 10 to 20 megawatts worth of power. But now increasingly, we're seeing artificial intelligence requiring between 100 and 500 megawatts. So the term that we often use for that, the people who are supplying the compute for artificial intelligence, we call hyperscalers. So they include some of the largest tech companies in the world, like AWS, Microsoft. And to provide the services that they do, they require, you know, this huge amount of power. And so in Australia now, there are plans to build what they call gigawatt facilities. And that really is just an extension of this, of the hyperscalers moving into those, that territory. But, you know, when we're talking about gigafactories, then that is essentially using up the same output as, you know, a standard Australian power station. So it's a significant investment of power. But what that buys you is what we call tokens. So a token... We were hoping we would come to this. What is a token? A token is, so if you've had experience with prompting a system like ChatGPT, generative AI, then every word that you put in is treated like a token, or even a fragment of a word might be treated as a token. And then once you have put your prompt into the system, that then sets off a whole chain of, it creates a lot of vectors, it goes through and it does, you know, a trillion parallel, you know, processes to check how your prompt actually matches up against the information that is contained in these large language models. And in essence, to do that mathematical computing requires the conversion of electricity into essentially the maths required to give you an answer through these generative AI systems. And so the way that companies make money from that is by charging you per token. But in essence, that's really equivalent to how much energy is being used. So it's converting energy into artificial intelligence. [00:05:19] Speaker 3: And so how many prompts make up a token? Or how do you actually sort of quantify what a token is? We hear them talk about, oh, we need tokens. What, how much is that? [00:05:28] Speaker 2: How much is a token? Well, I guess it depends on the provider. So I probably can't give you two, you know, like an exact unit price. But what I can tell you is from the university's perspective, where we have a lot of researchers developing artificial intelligence models. We have a number of ways that people can get tokens. They can be using GPUs at their desks and operating off those devices. We have our own supercomputer. We also access our national supercomputing facilities, the NCI here in Canberra at ANU or PAWSI in Western Australia. And occasionally we will use the hyperscalers. But in general, that costs us three to five times as much as it would to use our own infrastructure. So by hosting a lot of this computing infrastructure in these data centers is where companies like AWS, Microsoft make a lot of money. [00:06:28] Speaker 3: I'm keen to ask about sovereignty and hyperscalers. You make the argument that we need sovereign capability and we do have some Australian owned sovereign data centers that have been in the game for a long time and are doing some work in terms of compute data centers. But we've also got this huge push from big multinationals that view Australia as the place to invest at the moment. How should we be thinking about the differences between sovereign capability investment from these big hyperscalers and what that might mean for the choices we make going into the future? [00:07:08] Speaker 2: So I think what's inside the data center is what we need to be focusing on. There's a lot of focus on the data center itself, how much energy and water it uses and on placing requirements on the data center operators. But it's actually the tenants of the data centers that have the control over most of those things because depending on how much compute they are drawing on at any one time that actually is what is putting the demand on the electricity grid and requiring cooling from water. It really has to do with how much compute those companies are doing on the behalf of either Australian companies or or completely foreign companies. So in Australia we have data center operators. We also have our own class of what we call neocloud providers who are able to provide computing in a sort of a niche. They're not direct competitors with the hyperscalers but they do offer a lot of infrastructure that you could use for artificial intelligence and have the advantage that they're Australian taxpaying companies. So I think there needs to be a balance. The hyperscalers are already here. They already service a lot of our corporate companies here in Australia and an important part of the ecosystem. We also potentially can leverage their investment in Australia to our benefit but only if we are looking beyond the construction of the data centers themselves and the ongoing revenue that is going to be generated by the computation that's occurring on our shores. Because I think what some people don't appreciate is that those companies don't have to sell any of that computing capacity to Australians. They could be selling it to anyone in the world. One of the things that makes Australia an attractive place to build a data center is because one we're considered geopolitically stable. We're also geologically stable. We have submarine cable connections to all of the major markets to Europe, the US and Asia. And in general, although I certainly don't feel it when I pay my electricity bill, we're considered to have competitively priced electricity. And with our focus on renewables, if you're a company that wants to meet your net zero targets, it makes sense to do your AI in a data center in Australia because then it's not going to impact on your path towards getting net zero if that data center is powered through renewable energy. Does that make sense? [00:09:27] Speaker 1: Yeah, yeah. That's a really interesting point that you just raised in regards to the cables that we have to different parts of the world. We've had some of the big companies sort of, by no means are they threatening, but they're highlighting the fact that they're also considering going to Canada if Australia is not going to be willing to play ball. Does Canada have the same sort of connections to those parts of the world that we do? I haven't looked at it, but quite possibly. Okay, so it is possible that Canada also has those kind of connections. And then just on the NeoCloud providers that you mentioned before. What's a NeoCloud? Well, these are these Australian companies that are providers of compute that could be used to help train AI as well. Well, should we, is it something that we should be considering to compel some of these bigger American companies to work with these Australian NeoCloud providers or is that not an option? [00:10:24] Speaker 2: Well, I think co-location can be really effective. So, you know, where you have some of the hyperscalers operating, also making sure that there are NeoCloud providers and, you know, just seeing what synergies might exist there. But in general, there is always going to be a demand for hyperscale services. And they are very effective for providing a lot of our large corporates with the AI inference that they need at scale. And so there is a place for them to play. But from a, you know, a university perspective, for example, NeoCloud providers often have quite specialised, you know, high performance GPUs that are really well suited to training AI models, which allow us to do things like develop models for drug discovery and other scientific breakthroughs that are important. [00:11:09] Speaker 3: Mm-hmm. You mentioned the cost of using the hyperscalers for compute. Do you think we should be looking at something like a compute reservation policy so we don't end up in the same situation we are with gas where we have all of this infrastructure and are exporting huge amounts but aren't actually reserving some for Australians? [00:11:27] Speaker 2: Yes, definitely. And I don't think it's too much to ask for, really, given the scale of the investments that is being considered. So yeah, it would be a real shame if we are hosting all of this compute and we're not able to get any access to it. And it has to be accessed beyond free tokens because I think we really need to have some form of either offtake agreement or some way that that compute can be accessible to Australians and Australian businesses, you know, potentially at pre-arranged prices. I'm not sure exactly what the mechanism needs to be, but what will probably be offered are free tokens. And so it needs to be something that means you don't have that vendor lock-in and that is interoperable so that, you know, you could be doing that compute at one location but be able to move that when the [00:12:17] Speaker 3: need arises, if that makes sense. Right. And that's why the free tokens thing doesn't work because then you then have to use from that particular data center of hyperscaler. Yeah. [00:12:26] Speaker 1: So you mentioned an offtake agreement. What's that? [00:12:29] Speaker 2: Where you have an agreement in place, you know, right at the start of how much compute demand you anticipate and how much you're prepared to pay. [00:12:38] Speaker 3: Okay. And to the broader question about data centers, how do you think Australians will benefit? I guess this is a real live debate on a few levels. One, is this just going to speed up the development of AI, which means it's coming for, you know, white collar jobs and then eventually everyone's jobs when you combine it with robotics. So should we be doing it? And then on the benefit, long-term benefit from data centers, how do you see that playing out? What do you think about it [00:13:07] Speaker 2: all? I think Australia has a really good opportunity to be an exporter of sustainable computing. So as I mentioned, one of the big advantages that we have, particularly as we lean into more renewable energy, is that for companies that are serious about meeting their net zero targets, the amount of energy that they are consuming by using artificial intelligence is going to make it difficult for them to meet those [00:13:33] Speaker 1: targets. It's blowing it out of the water, isn't it? We've seen Microsoft already, I think, drastically walk back their targets. [00:13:40] Speaker 2: Yeah. But if we as Australia is supplying sustainable compute to the rest of the world, then that is something that we can actually have a very positive impact on. However, if we don't own any of the compute that's inside those data centers and are not generating, you know, Australian tax to, you know, help fund the operation of our government and society, then we're not really going to see those benefits, which is why I think if we can encourage more of that ecosystem of Australian neocloud providers so that they are the ones who are making money, so Australian tax paying entities are the ones who are making money out of the sustainable computing that we export, that is where Australia can get benefit. But apart from that, as you said, I mean, we could be looking at a future where we have all of this infrastructure, we've gotten, you know, the sugar hit of, you know, some foreign investment immediately over the next couple of years, but then nothing. And in fact, actually probably paying at higher than market rates to be able to access the compute that is actually onshore. [00:14:44] Speaker 3: And what do you think about the argument that, well, this is additional energy that the grid will need to supply. We haven't really planned for that. That's going to have an impact on households and businesses, energy costs on transmission. And then there's also the water concern amongst communities. How, how do we, how do we think that through in a really sober way? Because, you know, we, we hear it's all upside from these big hyperscalers who are desperate to to invest here and, you know, and the government to a large degree, but there's a lot of concern amongst people in the community. [00:15:18] Speaker 2: I think people are right to be concerned. And those are the right questions to ask, because if you're looking at gigafactories, then our grid, as far as I'm aware, is really not set up to be able to handle that. And even if you were to look at a situation where the data center, you know, is some way self-contained in terms of its energy requirements. I mean, is that actually the outcome that we want? Surely they should also be contributing to help stabilize the grid and putting any excess energy that they have onto our grid. And at the moment, I'm, what I'm not seeing is a lot of commitment as to how these investments are going to support our, our grid, you know, the actual lines, the power lines, and the infrastructure that's required to distribute energy across the whole country. Now, I'm not an energy researcher, but, you know, I'm sure a lot of them have, have views on, on how this could work. But I haven't heard of anyone who thinks that Australia's current grid system is set up to be able to cope with this huge increased level of demand, nor how you're going to be able to switch on these renewable energy sources overnight. So to my mind, the timeline isn't making sense. Like, how are we going to be able to build the data centers and have this compute, but magically not have an impact on our water or energy? [00:16:38] Speaker 1: Well, that's been raised with you previously, the timing of lining up this new energy that we need in the system with the data centers actually coming online. And I think I wanted to ask, I think, about that syncing up of the energy, about ensuring access for Australian businesses and researchers. So we've seen the statement of expectations from the government, which, correct me if I'm wrong, but I think I've heard you speak before a little critically of the fact that they're just expectations and there's nothing enforceable about it. Should we be locking down exactly what we want these things to look like and putting it down as actual requirements that they must meet, not just [00:17:21] Speaker 2: being expected to meet? Yeah, definitely. So, and those requirements have to cascade down because the person that is applying, the people that are doing the planning approvals, those are the data center operators. But as I mentioned, they're not actually the people that control what type of compute is happening within their data center. In many cases, they don't even know what is the architecture of the compute that's in the data centers. You know, these trucks come in with their racks. They go up in their own lifts to their own part of the data center. And it's completely, that information is not accessible because it's a competitive information. So I think really a lot of those requirements have to flow down from agreements from the data center operator through the tenants. And the European Union has thought through this, but I think unless we have enforceable requirements that are not just on the data center operator, but actually flow through to the tenants of those data centers in perpetuate, I can't pronounce that word, you know, permanently. So it's not just for the anchor tenants, it's also for future tenants, then we're going to face problems because it is the suppliers of that compute who control, you know, essentially how much water and energy is being used. Now, the only upside I would say is that the newer generation of these, you know, graphical processing units tend to be very energy and water efficient. But at the scale at which they're being deployed, that kind of gets absorbed because, you know, in general, they are producing a lot of heat, which requires water to cool. But then, you know, I think we have a lot of clever people in the energy space here, you know, we could be looking at how you can convert that heat into energy that then gets supplied back to the grid. Like there are a number of different ways that we could look to resolve some of these issues. But, you know, I think we've I feel like we're being put in a position of having to make decisions in advance of us actually having the opportunity to think through some of the long term consequences. [00:19:26] Speaker 3: This is my big concern. It seems like there's this mad dash and the government sort of saying bring your investment and we'll maybe sort out all these other things down the track. Whereas I think most Australians would say, hang on, we want to have you sorted out the energy, the water, the location and social licence. And also answering the bigger question of, you know, is this just going to mean mass job displacement in the future? And we really don't have any safeguards or a plan for that. [00:19:54] Speaker 1: Yeah, I mean, as we mentioned at the start, you are unique in regards to being so well versed in AI, but also robotics. What's your view on the future of work in Australia? Are you more on the side that it's just going to augment human work and we're going to all going to be doing a lot more but still working? Or that companies who are focused on their bottom line are just going to take it as an opportunity to lay people off and reduce their overheads? [00:20:27] Speaker 2: I think we're seeing both things happen at the same time. But, you know, the interesting thing coming from, you know, I guess more of the robotic side, which I guess people now referring to more as physical AI is the fundamental difference that we're seeing in the standards that we apply to non-physical AI versus physical AI. And I'll just give you an example of what I mean. If you were to deploy a robot on a manufacturing factory floor, and that robot were to injure someone, then it's pretty clear under Australian standards and Australian law who is responsible. And we could have a situation where, you know, the directors of the company face criminal prosecution over that injury, because for that to have happened, there's been a number of things that must have gone wrong. And, you know, so standards haven't been followed, you know, functional safety checks haven't been made, and it's been an unsafe work environment. But we are seeing non-physical AI being deployed at scale with none of those safeguards. And that troubles me, because I feel if we've been able to solve that for physical AI, why are we not able to solve it for non-physical AI? And really, the main reason is because you can just download non-physical AI so simply. So people can come into your workplace with AI on their phones. And, you know, that's not a decision that the company has made. But I do think we should actually take a leaf out of the book of how physical AI is governed and apply that to how non-physical AI is applied. Because I think increasingly, we're seeing that some of the harms actually enter the physical world, where we see that people are committing suicide based on advice that they've got, that people are asking generative AI for advice on health matters. And, you know, I think recently, it was shown that some of these generative AI tools are giving health advice out that's wrong three out of five times or, you know, some enormous percentage. And in fact, you can't rely on generative AI to be right more than 90% of the time under the best circumstances. So we're placing a lot of eggs in the AI basket. And in terms of workforce transition and change and impact on workforce, that you would never countenance, you know, if it had a physical form. And so I think that robotics is really interesting. And perhaps, maybe we could use that in from a more optimistic view. Like if we can, we're able to do it with robotics, maybe we just need to apply some of the same principles to non-physical AI and have greater expectations of employers in how AI is being deployed. [00:23:01] Speaker 3: We've been looking at some of the really troubling stories of young people having relationships with AI chatbots and just the path that leads them down. And then, yeah, just zero responsibility from these huge AI companies. I think that's the sort of thing where we should be saying you are actually liable for what your model says to someone. [00:23:23] Speaker 1: Well, this comes down to something that you speak about in regards to sovereignty of AI and our sort of path to achieving that, that you've said that we're trying to regulate AI systems built overseas, over which we have no control. And is that part of the tension that we're trying to figure out here, if, if physical AI or a robot makes that kind of a mistake, well, you know, it's physically happened here and you can be held accountable here. But how can we try and, as a country, Australia, have influence over these companies that are just doing AI overseas and it's, there's nothing physical here to represent that? [00:24:04] Speaker 2: Yeah. Well, look, I think we can take a leaf out of the EU book. You know, I'm not suggesting, I'm not a legal scholar, so I'm not going to get into the whole argument about whether we should have an EU AI act. I'll just, you know, share that observation that in where physical AI is concerned, we treat it completely differently. And yet I don't think some of the outcomes are that different. So we need to be reframing that. But on the idea that all of these systems are produced overseas, that's where I would push back because I find it very frustrating that Australia is not having more ambition in this space because working at a university, I see a lot of the great talent that we have in this space. But then sadly, I also see it leaking overseas. So my sister's the managing director of Silicon Valley robotics. So, you know, for the last decade, she can, you know, spin a tale of how many Australian robotics startups have moved. I mean, the number of AI startups and AI researchers that we have, you know, generated, you know, generated, but then have gone on to move overseas and create overseas companies, I think, you know, is unfortunate. And I think that with not a huge amount of effort, we could actually be generating that value here, developing our own AI systems. And obviously, they're not going to be head to head competitors with the big frontier models. But I think there's still a lot of things that we could do. And for a lot of the things that we rely on, in particular for our corporates who are building models of, you know, US models that can now be closed or of Chinese models that, you know, the Australian government probably doesn't sanction for use, we really need to be coming up with some of our own alternatives. And maybe they're not going to be as functional as some of the more mature offerings from, you know, other countries. But I think, you know, that we have to address tech dependency. And like I said, what I find frustrating is that we don't have more ambition in this area, given that we are blessed with, at the moment, quite a lot of good talent in the space. So we could be doing things, but we're not backing our ability in that space. [00:26:13] Speaker 3: In the last parliament, managed to get an inquiry up into sovereign tech capability. And it was tragic the number of times we were asking questions of, say, government departments about various schemes and procurement issues. And they would say, well, we don't have that capability here. And literally within five, 10 minutes, we'd have a few Australian companies email and say, listen, I do this exact thing. And actually, I have to sell to foreign companies and governments, because the Australian government won't give me a, you know, won't even look at what we've got. So do you feel like there's this sort of cultural shift that we need, and particularly in procurement here in Canberra, where we look to the big primes, the big hyperscalers, rather than saying, well, what is actually happening right here in Australia that we can back, test out, and actually build more sovereign capability? Yes. [00:27:05] Speaker 1: Well, that, and I think also what you've, as well as procurement that you've spoken about a lot, the R&D spend, would that be part of trying to address this in Australia? You know, we're at, what are the levels of R&D? [00:27:18] Speaker 3: Lowest on record, tragically low, way below, below the OECD average. And every time you speak to a university professor, they're telling me about all these students that are going, going overseas. And yeah, these are people that we've, we've got to retain here and here in Australia. [00:27:35] Speaker 2: Yeah, I can't explain why we're not investing in this space. But actually, maybe you can give me the answer to that. I mean, is Australia just broke? I mean, is that the answer to why we are considering all this foreign investment, potentially without, you know, much thought as to what's happened? Or why? But I mean, and if that's the case, that we don't have money to invest in R&D, that if we don't accept this foreign investment, then everyone's going to take a serious hit on our lifestyles. And then why are we having that conversation? [00:28:04] Speaker 3: I mean, we're so wealthy, we can give away half of our gas for free. So there's money there if we, if we want it. And, you know, the return on this sort of thing, it's a really sound financial investment for us as a country. So certainly something going to keep pushing. [00:28:21] Speaker 1: Yeah, I wasn't expecting this to go in the R&D direction. So I don't have a few of the stats on there. But I can say with fair confidence that any of the research around money that is put into R&D spend, you get huge returns on that, on that money. So I think one of the issues that we speak a lot about as an office is that, that long term thinking that we seem to be lacking in politics. And this would certainly be part of it, because while you do get that kind of return on R&D investment, it may not be until 10, 15 years down the line that you actually see that return. And so as a, as a government who might be looking to the next election, do I spend, spend the money on something that I might not see the benefit from for a decade? Or do I spend it on this thing over here that, you know, has a shorter turnaround? I think that's definitely part of the, the issue. [00:29:20] Speaker 2: But are we the only country in the world that has this problem? Because you just compare us to any other nation, we seem to be the only developed nation that thinks that we can go through this AI transformation with zero public investment. [00:29:34] Speaker 3: Yeah, I mean, you know, the last, last term of the Albanese government, they did a lot of work on an AI safety act. They did a lot of work standing up a sort of expert AI safety committee. And this term that was all scrapped. And we've basically gone with a statement of expectations, sort of light touch approach and relying on foreign investment in the form of data centres. That's how we cash in on the AI boom. And I do think it's tragically short sighted. We're not saying what's happening here in Australia, how do we ensure that we're backing local researchers, businesses, startups, and then you're looking at these foreign investors and saying you can come here, but here that here are the terms and conditions. And this is how you're going to benefit, but we'll also benefit from it. [00:30:23] Speaker 2: And also, how are we going to upskill the Australian population to be best suited to adapt to a lot of these changes. So in Finland, you know, they invested in getting all of the universities together to develop a program to educate everybody with artificial intelligence. And I think one of the other traps where you need to be careful not to fall into is substituting that sort of investment for vendor specific AI programs. Because that is often positioned as an investment by these companies into education in Australia, when really, it's just developing future customers. And that was something that came up in [00:31:04] Speaker 1: the inquiry during the last parliament, the need to have those sort of the upskilling of the general [00:31:09] Speaker 3: population as a key part of this. Yeah, I was looking at the other day at the percentage of GDP spent on retraining. And I think we're 0.3% here in Australia, some of those Scandinavian countries are 2%. And so it just shows the amount of effort that is going into preparing the workforce and getting people into new jobs, different jobs. It's going to be a huge part of the challenge here. [00:31:34] Speaker 1: Yeah. And another person we spoke to again recently, who hopefully we can have on here for a chat soon at some point was making, I think, a very strong case around this being a countrywide change management project and helping people understand how to actually progress through this quite revolutionary technology. I feel like we've gotten a little bit dark just in the last five minutes. If Australia successfully harnesses its unique strengths, like we've spoken about renewables, we haven't had the chance to speak yet about data sets, but it is something that Australia is quite strong on. What does our ideal global competitive AI economy look like 10 years from now? [00:32:21] Speaker 2: I think Australia has always been really good at developing technologies for resource constrained environments, which is why Australia was the first country in the world to automate its ports. Australia is the first country in the world to automate many of its mine sites. And that's because as a country with a small population inhabiting this gigantic landmass thousands of kilometres away from anywhere else, we've often had to develop technology that doesn't have a lot of energy requirements that can operate without necessarily being able to communicate with satellites. So in communication-denied environments where often you're having to have programs that can operate on a device rather than having, being able to communicate up with the cloud. And so these are things, and in an environment of water scarcity. So I think if we lean into all of those strengths, there's a lot that Australia could benefit from. Because, I mean, surely if there's anything the world needs now, it's AI that is energy and water efficient. So that's something, an area that we can lean into. [00:33:27] Speaker 1: And can you just speak about Starfish quickly? [00:33:30] Speaker 2: Oh, yes. So I guess sometimes people wonder, you know, what is the benefit? Because I think, you know, there's no reason for us to lean into this whole AI transformation if people aren't seeing benefits. You know, in particular, I think the benefits for individual health and education is something we need to explore. But a lot of people don't think about the benefits for the environment that we could be aspiring to. And one of the technologies that was being developed when I was at the Queensland University of Technology and is still ongoing is a little robot called RangerBot, which is able to identify crown of thorns starfish, which unfortunately is eating most of our Great Barrier Reef, and can look at controlling the numbers of the crown of thorns starfish. But importantly, what the RangerBot can also do is harvest coral spawn and collect it and move it into water temperatures that are more conducive to coral growth. Because unfortunately, as our water temperatures rise, coral is spawning in areas where now it's difficult for the coral to grow. We actually need to be moving the reef further south. But you can imagine we don't have enough people to do either of those things. So at the moment, to control crown of thorns starfish, what we do is we send human divers down into areas where tourists go. And that's the only part of the reef that we try and protect because the reef is the size of Italy. It's an enormous area. And unfortunately, diving is fundamentally unsafe. It's, you know, if at all possible, we shouldn't be sending people down to do this work. So imagine if we had a whole swarm of these robots that could be helping to control crown of thorns starfish numbers that can be helping us to grow coral and actually help to remediate the damage that's currently being done to the reef. I think that everyone would probably be pretty happy with that outcome. So there are a number of positive things. I mean, I wouldn't be, I wouldn't have the role as director of UNSWA Institute if I didn't think that there are a lot of positive benefits of artificial intelligence. But we have to make sure that we're alleviating those risks and most importantly, making sure that everyone benefits and that those benefits are as evenly distributed as can be. [00:35:46] Speaker 3: I mean, there's clearly some incredible developments, particularly just with invasive species. You know, I know of a company that's built a edge, edge, edge, edge AI model that can identify a cat and then a feral cat and spray it with a little bit of poison and then grooms and dies, doesn't pick up any other species. Others who are using it to live catch things like cats and foxes, like some really impressive things. More of that, but on the broader issue that I think people feel with AI is, is this going to take all of our jobs? How should we be weighing that up as a society where this mad rush to invest in data centres to speed up AI model development, which it seems to me the amount of money being invested in AI, the only way you recoup that is to displace jobs at scale. Do you think, how should that factor in? [00:36:44] Speaker 2: Well, I think again, it's where government can play an important role. So I mentioned, I'm not sure that I mentioned, but you know, in Singapore, they had a scheme which was to encourage companies to adopt robotics. But in return, they had to guarantee that they wouldn't reduce headcount. And so it actually led to an upskilling of the people who worked in those companies that took on the scheme. So it was kind of a clever policy mechanism to ensure that the companies were keeping up to speed with the latest technology, but that it wasn't having a negative impact on workforce. And I think in general, one thing I'm not seeing enough of in Australia is a lot of thinking around that workforce planning and transformation and what the workforce is going to look like in a few years' time. And there are notable exceptions. I mean, there are some legal firms that are really having a look at, well, what does it mean for graduate entry jobs? If we take less graduates, then what does that mean in terms of the demographics of the whole firm? And what do we need to do in place? But then unfortunately, you have other companies that just go, oh, great, we don't have to employ graduates, and we can save, you know, some money this year. So it really is a matter of short-term versus long-term thinking. And I think the other challenge is that this change is happening much more rapidly than many changes in the past. So as to your point, it really is a huge change management issue. Should people be concerned about work? I think they should be. I think that it's not going to be a job apocalypse and that there's probably going to be a lot of other economic factors that are going to impact on the job market. It's important that people stay curious, but I think it's important that people feel that they can have a voice on this issue because I think it's actually probably that individual action collectively that will end up making a difference. So what I guess what I'm saying is this is likely to have an impact on jobs. I would hope, over the long run, that there will be a balance and that there will be a whole bunch of new occupations. But really, that's a human choice and a choice by employers over whether to take that short-term economic gain that might, you know, have a detrimental impact to social cohesion. So what are the incentives that we can put in place that encourage businesses to think, well, what is the extra value that I get from having people here? So yes, that particular task or element may cease to exist. But you know, I think humans are endlessly creative. I think, and I think not very good at not working either. So, you know, this utopia where we don't have to work and we're all on universal basic incomes, um, you know, I think people just inherently look to find things to do that are useful. I'm sure that there are useful things that we can be doing. It's just a matter of how we value them, how we're going to pay for them and then how we potentially redirect human labour towards those things. [00:39:41] Speaker 3: Yeah, I just don't think we've got any examples where in the absence of regulation and the government potentially using sort of a combination of sticks and carrots to say, we want you to do this, we don't want you to do this, that it works out well for people. You just have to walk into your local supermarket where you're essentially checking yourself out and sometimes don't even have an option to talk to a human and go through a checkout to, you know, flying in Australia now where you essentially do the same thing. And I just think the incentive structure for businesses will mean that as soon as one business competitor takes that step, everyone else then has to follow to keep up. And that's where I think we need the government to step in. And I agree with your thing on UBI. In one sense, it could be a great thing that everyone has some sort of financial security, but as humans, we need meaning. And I don't see UBI giving us that meaning. I think that's a whole other conversation as a society where I think we are all feeling this sort of deficit of meaning in our lives to some extent. [00:40:49] Speaker 1: Yeah, well, I feel like we could definitely keep going for quite a long time, but it is a sitting day and Dave, I've seen the messages flying in. You're going to have to get moving, but thank you so much, Dr. K. [00:41:02] Speaker 2: Yeah, thanks Lincoln. Thanks David. [00:41:03] Speaker 1: Thank you very much. Well, thank you so much for joining us. If you haven't already, please subscribe so that you don't miss an episode. And if you're loving the work that David is doing, make sure you're following him on all your social media, or go to davidpocock.com.au and you can sign up for email updates. We'll see you next time.