Transcription
Let's start with the basics, because in my experience, AI discussions are often quite vague. It's hard to get a grip on what we're actually talking about. So, I'd like to ask everyone first, what have you personally used AI for most recently? Let's start with you, Jussi.
- Well, I-, I used it for this-, I'm going to Hong Kong for a couple of weeks for a research visit, maybe to visit China, so I asked AI what kind of apps, applications I should pack for payments in China and what I need in Hong Kong. And that's what I asked.
- And the answers were good?
- They seemed quite sensible. They are existing apps.
- What about Vili?
- I got a pike from the market, so I asked what kind of sauce I could make for it. You just take a picture of the fridge; these are the things you have. So, as I've noticed, if you try to use a language model at work, and it takes about as much time to verify the answers as it does to ask the question itself, then in this kind of cooking matter, it doesn't matter if it's a bit more artistic, if the facts aren't that important.
- What was the sauce?
- Well, it was a kind of hollandaise variant, which also had a bit of umami, miso in it.
- Right. What about Mikko?
- Well, I inevitably use it whenever I do Google searches, because AI comes up as the answer at the beginning, so otherwise I'm quite lazy to go to AI. So, AI comes to me at this point. At home, my wife uses AI more, and we had a small incident this past weekend when a pair of cross-country ski boots from last winter had gone missing, and we had to figure out what type of ski binding it was, and I warned that it was old and recalled which one it was, and it's SNS. And during that time, my wife had managed to take a picture of the skis and got an answer from AI that it was NNN, and I said, tell them it's SNS and my husband says it's SNS. It gave a very arrogant answer, that your husband is wrong. [laughter] It's an NNN binding. Salomon has never made this. And luckily, my wife believed it when I went to the attic to get my own boots and showed which one fit, but the conversation continued the next day, and little by little the AI answer was retracted, so I don't quite believe it unequivocally yet.
- So, this time, for now, the professor's intelligence wins over AI, at least in ski and boot matters.
- This time.
- Good. Let's move on to the actual topic and what it takes for us humans to be able to do things like this with AI. Jussi, could you simply explain what your own request to AI was about, and how and what had to happen for you to get the answer?
- Yes, good question. The internet is our global network, and nowadays AI models practically run in data centers, which we'll probably talk more about later, but so, somewhere at the other end of that request-, at one end is my laptop, phone, at the other end is the data center, and in between is the network, and it might go, the cables go, well in Finland's case, if it needs to go abroad, typically under the Baltic Sea and- or overland elsewhere, and then at the other end is the data center, where my sent query is processed. And in order for it to be processed and fed to the AI model, the model had to be trained beforehand. So, a lot of training data is collected, the model is trained. If we look at the big players, they release new models maybe two or three times a year. So, training the model is something that happens rarely, and then every time someone sends a query there, it's run through that model and the answer is delivered back to the user's screen via the network.
- And in that data center, there are basically computers.
- A data center is a big, huge industrial hall full of computers.
- Exactly. Good, Mikko, let's go a bit deeper into what else is behind AI besides these data centers and computers, what's in those computers? You are a professor of inorganic material chemistry. What does a chemistry professor do in an AI panel at all?
- Good question. I'm also looking forward to it with interest. But no, seriously, Jussi is talking about computers, and computers have microchips, so there's the hardware, and without it, no models would exist. And what we do in our research group in Kumpula is we develop the chemical methods by which these microchips are manufactured. We aim to anticipate at least five, preferably 10 years ahead, what materials will be needed in the future and develop the necessary chemistry for it. Our method is atomic layer deposition, developed in Finland in '74 by Tuomo Suntola, which has now become very central in the manufacturing of these microchips.
- And what is a microchip? How big is it?
- A microchip? Well, they come in different sizes, but typically, they are probably 6 plus or minus 24 cm in size. They are made on silicon wafers. Silicon wafers are 300 mm in diameter, and the materials needed for microchips are built layer by layer on them, patterned, and the next layer is added on top.
- And what does their production require, and where are they made?
- Well, their production requires-, first, of course, you have to design what they will be like. So, there's chip design, from which arises, so to speak-, or not so to speak, but from which arises the blueprints according to which they are made in semiconductor factories. And semiconductor factories need a lot of equipment to manufacture them. When you go to these more advanced semiconductor factories, the price tag is 10 billion, preferably maybe 20 billion. And there are very expensive machines used to make these materials. They are added, patterned, everything is done in cleanrooms. Very expensive technology.
- And where are these factories located?
- Well, the largest manufacturer is Taiwan. Taiwan produces maybe 60% of all semiconductors in the world, 90% of these most advanced microprocessors. Korea produces a lot of memory chips, America produces some, Japan. Europe was once big, now it's quite small and is trying-, wants to regain its share, increase its share.
- Good. Well, let's briefly talk about what materials are in them. Rare earth metals have been discussed a lot in public. Trump wants them from both Greenland and Ukraine, and dependence on China. They are talked about a lot. But are they related to this in any way?
- This is perhaps a bit disappointing. Rare earth metals are not very important when talking about microchips themselves. We discussed this with a larger group, and there are very few rare earth metals, but where this confusion comes from is that there are a lot of critical materials and strategic materials, which include rare earth metals, but there are very few rare earth metals in the microchips themselves.
- And what are these critical minerals, if you could briefly mention.
- There are 37 on the EU list, I believe, so silicon itself is found there, but silicon is on the list because it's an important material, but there's no problem getting it and producing it in Europe at the moment. So, they've listed those that can cause problems, and some are already such that they must be obtained from outside Europe, and that can quickly become a problem, for example, especially through China, so for example, germanium, below silicon in the periodic table, which is needed there, germanium is much more critical.
- Good. Well, that might also provide a bridge to politics. Vili, you are a professor of technology policy. If you had to highlight just two political questions or themes related to AI, what would you choose?
- Well, one is, of course, this technological dependence, because these supply chains are so concentrated. At every level of the supply chain, starting from the raw material, which is only produced in a few places. Chip design, which is largely controlled by one American company, chip production, and then running and owning these data centers and developing AI models, which is largely in the hands of the same tech giants as our digital world otherwise, so this deepening AI use threatens to deepen our existing technological dependence. That's perhaps one theme. And the second is the change in the economic structure, which I find fascinating, because we had this idea that digitalization means moving from smokestacks and physical industry to this information economy, where only what you know matters. Intellectual capital, not so much physical capital-, and countries' competitiveness is based on intangible things, and now we've suddenly returned to a situation where competitiveness is measured by the ability to produce raw materials, produce gigawatts of electricity, and invest billions in these huge factories, practically. Programming is becoming a smokestack industry. I find this fascinating.
- It is indeed extremely fascinating, and as you described it well, we haven't thought about this physical side of the digital world at all in public discourse, at least. Good. Data centers have also been discussed a lot in Finland. They are being built in Finland and are on the one hand very welcome, as we are getting some investments here. But on the other hand, there's concern about how much energy they consume, and then again, perhaps about their ultimate economic significance and whether they create jobs in Finland. Let's perhaps focus on the energy issue for a moment. Jussi, what do we know at all about how much energy data centers consume, use energy?
- Well, there are many different sizes of these data centers, but generally the discussion focuses on the larger ones, like Google's in Hamina and others. And it's unusually difficult to get any generalizable numbers from them. These big players don't directly say how much electricity they use. So, here we have to rely perhaps more on general estimates of global electricity consumption. Currently, according to some recent numbers I found, current data centers would use about 1.5% of all the world's electricity. And according to forecasts from the International Energy Agency, by 2030, it would be about 3%, and most of this growth would be precisely the increase in electricity consumption for AI.
- 3% sounds like a small number, but is it?
- Well, when you look at these different industries and others, you might not find such big-, there are these-, I remember 15 years ago when we looked at the carbon dioxide emissions of different sectors and so on, almost everything was around 2%, and there wasn't necessarily a big single culprit, but with AI data centers, the most worrying thing might not be the one and a half percent, but the rapid growth, which is much faster than in other sectors.
- Well, then if we think about it from the perspective of emissions, on the one hand, public discussion also emphasizes that Finland currently has clean electricity, and that's one of our competitive advantages in attracting these data centers here, so how much does the method of energy production matter for this, or is it taken into account in calculations?
- Yes, that often gets a bit overlooked, that these big players might report their own emissions, but not the indirect emissions that come from electricity production. And as you said, in Finland, we have quite clean electricity compared globally, which means that from the perspective of these data centers, we have a good product. We shouldn't necessarily give it away for free, but perhaps set some conditions for establishing data centers here.
- Good. We'll talk more about that soon. Mikko, I'd like to ask you again, how do these microchips relate to energy consumption, can they develop to consume more or less energy in the future?
- That is certainly what is being strived for, and if we look back 15 years, or over the last 15 years, their efficiency in performing a certain calculation operation has improved by a factor of 10,000, and in the future, the same should be done, and it's not easy to do, but of course, the entire semiconductor industry in the world is working towards it.
- Right. Well, Vili, with what mindset have you followed this discussion about data center energy use, and Finland's role in general?
- Well, in my research, I generally examine technology companies from a rather critical perspective, but in this case, I've felt that the public discussion about data center electricity consumption in Finland has even gone into a surprisingly negative spiral. So, people here are now very interested in how many megawatts a production facility consumes and how it affects their spot electricity price. I'd like to ask, how many people know how much a paper mill consumes electricity? Because before data centers, we had pulp and paper industry here. And as is known, Google's Hamina facility replaced a paper mill. And paper mills consume a lot of electricity. And in fact, industrial electricity consumption has decreased significantly in the last 20 years, from about 50 terawatt-hours per year to 30 terawatt-hours per year. A massive drop in industrial energy use. And now, when we've moved from paper to bits and replaced paper mills with these computing facilities, and they also consume some electricity, but they don't even consume as much as paper mills used to consume at their peak, and now there's huge concern about it. I think it's gone off on a bit of a strange track. I have a hypothesis about why this has happened, right?
- Yes.
- Can we ask?
- Yes, let's ask. How many people here follow the spot electricity price diligently? Quite a significant portion, yes.
- Well, yes, but not as many as I would have expected.
- Well, we are in the center of Helsinki too-. [laughter]
- Spot electricity prices started being sold to consumers about-, when did that happen, 10 years ago. Before that, how many of you followed the spot electricity price? [laughter] So, when there was the paper and pulp industry and the forest industry consuming electricity in Finland, consumers didn't care at all how much a factory consumed, because they had fixed-price contracts. Now that consumers have become, so to speak, competitors for the electricity used in industry, everyone naturally wants to buy electricity at the cheapest price, so this new wave of industrialization is happening at an awkward time, because now consumers are suddenly an interest group demanding their interests be respected, which didn't happen before. I'm not saying it's wrong, but it's a new phenomenon, so to speak. And I think it would be a shame if this were to prevent an investment wave that, evaluated by other criteria, would be quite good for Finland. It's nice to get-. I also have a spot electricity contract, and I like to go to the sauna quite a lot, especially when electricity is cheap, but the fact that I heat my sauna with cheap electricity doesn't create any new jobs in Finland. So, if data centers create a few new jobs, the sauna is unfortunately the worst in terms of employment.
- Do Jussi or Mikko have any comments on this?
- Well, that's true in a way, but that's precisely where the whole data center discussion in the media comes in-, where everyone is defending their own positions from different trenches, and these things, like employment effects, and so on, yes, during the construction phase, although in Finland, construction is mainly done by Estonians, Lithuanians, and Poles, and then, so to speak, during the operation of the data center, what jobs are created, what type, some certainly, but Google has been in Hamina for over 10 years, and Hamina hasn't become any Silicon Valley of the Nordics.
- Well, does it matter where the data center is located, for reasons other than economic ones-, are there political reasons why it would matter?
- Well, I can comment on technical reasons, if we consider web applications, which I use for AI models or email online, browsing websites, and so on, then from this perspective, from a Finnish point of view, if it's in Europe, it doesn't matter where it is, because the response time is so small, a person won't notice it.
- What about in relation to these dependencies?
- I think it's interesting that with the internet and network connections, data processing and storage have become a tradable commodity that can be imported and exported. So, imported from abroad, produced locally, and exported. Like any such international commodity sector, industrial activity tends to concentrate where the factors of production are cheaper. So, at the moment, there are some data center investments coming to the Nordic region. More are still going to places like Frankfurt and Amsterdam, but some investments are coming here too, thanks to the cool air and relatively affordable and low-carbon electricity production, and above all, the electricity grid is in such good condition that it can connect a 100 MW facility with short notice, which is not possible in most countries. So, it seeks out where, any industrial activity, where it is most profitable to do. So, in that sense, economically it matters. And that's where the political aspect comes in, because computation is not just any commodity. It's not like oil, which can be stored, but computation cannot be stored. It's a service. It's consumed-, it has to be produced again every moment, what is needed, which means that it involves more risk of outages and so on, so it becomes a resilience and thereby also a political question.
- So, what can happen? What is the concrete threat?
- Well, if, for example, our public sector runs on computing capacity located abroad, then we are dependent on the fact that, of course, these foreign network connections will continue to be secured. That the place from which the computation is brought, that they will continue to want to bring it to us, and not suddenly decide, we're shutting down those servers for you. And then it involves-, well, we can talk later about the risk of concentration too, that if it's all brought from one place, it's even more vulnerable.
- So, it's a matter of security of supply to some extent, that they actually-?
- Yes, that's certainly the right word here, security of supply.
- In Ukraine, they have been concretely under this threat, so what has been the reaction there?
- Well, as far as I remember, four years ago, when the latest attack began, copies of key state administration systems were immediately made abroad.
- So, it was kind of the opposite-?
- Yes, but there were different reasons, to ensure the continuity of the state.
- Exactly. Good, Mikko, did you have anything in mind?
- I'd perhaps return to electricity consumption, that it's not quite as simple as it just being like Vili's or my sauna use being a bit more expensive due to increased electricity consumption by data centers, because electricity consumption is increasing in Finland anyway, or would be increasing if there were enough available, it could be transferred. Now I'm referring to the green transition, if we consider, for example-, well, very recently there was news that an E-methanol plant was planned for Naantali, which means that carbon dioxide captured from the air would react with hydrogen to produce methanol as fuel, and for it to be green, the hydrogen must be produced in a way other than currently. So, currently, most hydrogen used is made from fossil feedstocks like natural gas. But the green alternative is electrolysis, splitting water by electrolysis with green electricity. And these will be very large consumers. The steel industry is transitioning from carbon reduction to hydrogen reduction. It was abandoned in Raahe, but if it had been adopted, the amount of hydrogen needed-, or the production of that hydrogen would have consumed the entire production capacity of the Hanhikivi nuclear power plant, approximately. Of course, Hanhikivi is not happening, and that wasn't the reason why-. But electricity, and specifically green electricity-, there are many other buyers besides data centers, which together lead to the conclusion that it is a cause for concern.
- Yes. And I was thinking earlier, when we talked about paper mills, that at the time, paper companies also produced a lot of energy themselves and built production.
- Actually, a pulp mill produces electricity. A paper mill then consumes it.
- Right. Good clarification. This discussion has so far focused quite a lot on data centers in Finland. Should we dream bigger? Why isn't Finland getting its own chip factory? Now there's a lot of talk around the world about reducing dependence on Taiwanese chips, and efforts are being made to set up chip factories around the world. So, why not in Finland?
- Well, a chip factory could probably be set up in Finland, but when we talk about the most advanced chip factories, it's so capital-intensive, and a good example is the Taiwanese TSMC, when they are currently building a factory in Dresden, Germany, the investment is 10 billion, and 5 billion is public support. So, it might be quite difficult to allocate five billion from the Finnish state budget today.
- Yes. And you mentioned it briefly, but tell us again why they are so expensive and what costs so much.
- Well, yes, the entire technology is expensive, starting with the cleanroom where they are made, but the most expensive equipment there are the lithography machines, which are used to pattern these small structures. And now we're talking about something like EUV extreme ultraviolet 13.5 nanometers, the price of first-generation machines, one machine is 130 million. Now a second generation is coming. It's over 300 million per unit. And all other machines there probably cost millions of times more than the price.
- And who manufactures those machines?
- There are manufacturers all over the world, but this lithography machine, the most advanced one, is manufactured by the Dutch company ASML, which is a rather funny situation, because the semiconductor industry generally doesn't want to rely on a single supplier, but in this case, that's what has happened.
- So, no one else can do it?
- No one else can do it yet, but this goes back to politics, because exporting these machines to China is forbidden, and like all other equipment that cannot be exported to China, the Chinese are building their own. And now the big question is not whether they can do it, but when they can do these machines, and we are not necessarily very far from that.
- And what happens then? What changes?
- Well, then the global semiconductor industry could be completely shaken into a new position, and specifically in China's favor. So, currently, a lot of microchips are already being made in China, but not the most advanced ones, because this most crucial piece of equipment is missing.
- Exactly. Maybe then we'll get cheaper Chinese lithography machines in Finland too, and we can set up factories.
- Maybe, but unlikely. Yes.
- This is the technology struggle between the United States and China, so it wasn't a decision by the Netherlands or even the EU, but on the contrary, the EU didn't really join in, but the US State Department contacted ASML directly and said, could you stop selling these machines, and only after that did the Dutch state itself, and the CEO of the company later complained in the media that it's quite difficult for a single company to resist the Americans, that the EU should have a unified line on this and its own China policy, and not that the US dictates what can or cannot be sold to China. And they also complained that, according to their claims-, it would be interesting to hear an expert's view on this, that according to their claims, companies selling similar photolithography machines in the US were allowed to continue selling to China, but European companies were forbidden.
- Well, ASML is also allowed to sell lithography machines to China, but it's not allowed to sell these EUV machines, and China is certainly a big market for both Americans and Europeans, but the most advanced technology is under a ban.
- Good. We'll talk more about China and the USA soon, but let's stay for a moment or return briefly to energy, because for the ordinary citizen and consumer, it's sometimes visible how much one ChatGPT search consumes energy and causes emissions, and how AI is becoming a new source of carbon dioxide emissions at a time when emissions need to be reduced, so should we ordinary consumers avoid using AI in the same way that others avoid flying or eating meat.
- Well, good question, in a way, what would perhaps-, in a way, the environmental impact of a single AI action-, even if it's to produce a caricature of yourself, which was the hit of the weekend last weekend, is practically minimal. It gets lost in rounding errors, but then when tens or hundreds of millions of them are made, it's a difficult question in that sense, but perhaps the most essential thing would be to be aware that everything you do has some kind of footprint, so perhaps the most useless ones could be left out, or at least be aware that every funny cat picture has caused some kind of damage.
- It's a manufactured product.
- But what about the other perspective? Of course, there are also many utopian ideas associated with AI, and how it will solve the world's problems, so does that weigh on the other side of the scale?
- Well, it's probably the marketers who associate these images most strongly. The world's problems probably won't be solved by making cat pictures or caricatures, and whether they will be solved by this paradigm of large language models, what does a computational scientist say, or is this a situation where, when trying to achieve general human-like AI with this technology, it's a bit like climbing a tree to the moon, that the process is going very well, until suddenly there's a sudden stop at the end and you can't get there anymore.
- That's a good question. Yes, we are still climbing the tree vigorously, and the trunk seems to be sufficient, so perhaps the good question is whether the stop will come at some point. And perhaps, on the other hand, if you think about how much you can already do with these current AI tools, how much you can utilize them in your work, how they help, then on the other hand, it perhaps also indicates that our general level of expectation is not very high. It doesn't require reaching the moon either, but the middle of the tree trunk is enough for most people in most cases.
- On the contrary, it could be detrimental if one were too aware of AI, then one would have to start managing its emotions and so on.
- Possibly, yes, but hopefully, we are still far from that.
- Didn't they develop their own religion when they created that platform?
- Well, that's what the story claims, yes.
- Speaking of utopias, one thing that has been in the news in recent weeks is Elon Musk's utopias again and the idea that data centers could actually be moved into space, where they could operate. This would solve various geopolitical problems and border issues if they were in space and operated on solar energy, so what does that sound like? Is there any merit to it?
- I'm sure Elon Musk will do something like that, I believe. Whether it makes much sense in itself, that's a good question, so there are certain good aspects, these discussions about how data centers affect our spot electricity price, they would disappear. The fact that they wouldn't be located in any country, that problem would actually perhaps be exacerbated. They would still be owned by some operator. Whether it's commercial or some state or similar, it wouldn't remove that dependence, but it's an interesting idea, and there would certainly be interesting research questions about how this terrestrial data center infrastructure, if it were in orbit, could raise interesting questions. I do know that my colleagues who study near-space are starting to talk about space debris and so on, so it's not a magic bullet.
- There isn't infinite space in near-space either. Yes, but it's technically possible?
- Technically possible, yes.
- Yes. Interesting. Good. Well, we've already touched upon China and the United States, and we'll talk more about it. One key perspective that we haven't yet touched upon is which companies and which countries' companies are running this digital world. Your research group has literally mapped it out. Tell us what you've found.
- Well, for example, this image shows the infrastructure of leading cloud service providers around the world. So, the blue spheres are data center complexes of American companies. Green are European, and red are Chinese. And from this, it's clear that, firstly, as we discussed earlier, data processing production is scientifically concentrated in certain areas, it's not evenly distributed around the world, and it continues to concentrate. And secondly, it shows that the US and China have significantly more of this infrastructure, and this is not a completely comprehensive sample, but these European companies are quite the underdogs here, and for example, in this cloud infrastructure market, according to various estimates, these three American companies, Amazon, Microsoft, and Google, have about a 70% global market share, and of the remaining 30%, Chinese Alibaba Cloud, Huawei Cloud, and Tencent Cloud have a significant portion. And these European service providers have only a minority market share, even in their own domestic markets.
- Yes, we all probably use Microsoft Outlook or Gmail and Drive, and so on, so-.
- Yes, and this is about infrastructure on which all other services also run, so to speak. As a user, you don't know, when you use an app on your phone or a government online service, you don't know which cloud platform it's running on, but it's likely running on either Microsoft Azure or Amazon's AWS, so we-. We investigated the hosting of digital services for citizens and businesses in the Finnish public sector, and websites, where the infrastructure is physically located, and we found that 64% run on either Amazon's or Microsoft's cloud. And Amazon alone had about half the market share, so largely, services that people use every day like Oma vero (My Tax) and Oma Kela (My Kela) and apply for a passport and apply for a driver's license, they run on these companies' servers and not necessarily in Finland.
- But has it always been like this? The visible side to the consumer is that American companies have dominated this market.
- Well, there was an interesting intermediate phase, when organizations installed their own-, initially, when the internet first came about, organizations installed servers in their basements and then offered websites from there and managed their own infrastructure themselves, but there were many problems with that. And it was, you could say, an pre-industrial era, where everyone produced data processing for their own needs within their own community, and now we've moved to the era of industrial data processing, where data processing is concentrated in these large computing facilities, from which it is then distributed via cables. Just like during industrialization, we moved from each household and village forging its own tools to them being manufactured in factories and distributed through distribution networks.
- And now we are probably also living in an era of globalization, so as has been stated here, it is primarily these two superpowers, China and the United States, that rule this world. What is the significance?
- Well, what's interesting here is what Mikko has also brought up, that since the capital required for this infrastructure is so large, there is only room in the market for a small number of very large players in many such digital markets. This differs from the industrial era, where we generally had a machine shop in every province at least, and energy production in at least every country. But as seen from that map, not every country even has its own cloud computing complex, its own computing factory, but many countries operate solely on data processing imports. And this image shows, well, in countries where this infrastructure, cloud computing infrastructure, exists, is it American or Chinese? So, each bar represents a country, and the blue bars on the left, the infrastructure is purely from American companies, and the red bars on the right are from Chinese companies, and we see that both exist. In reality, the competitive situation is not quite fifty-fifty as the image suggests, because these Chinese infrastructures are generally smaller in size, so they are not entirely comparable, but for example, Finland is on the left side. In Finland, there is only cloud infrastructure from American companies. Our NATO partner Turkey is, for example, on the right side, with purely Chinese infrastructure.
- And the reason why this is suddenly being discussed in public is surely related to politics and who is the president of the United States, and what new risks might this-.
- This kind of infrastructure can be used as a tool of influence, and the United States has used it as a tool of influence. Countries, organizations, or individuals subjected to sanctions are denied access to technological infrastructure, personnel of the International Criminal Court, judges, and prosecutors have been sanctioned in this way. The biggest-, sort of, what Mika referred to, when the US denied China access to advanced technologies, then of course the consequence is that alternatives are quickly developed. So, it acts as a launchpad in a way, that's when the alternative is developed. So, in a way, the big risk is not that the US, for example, by seizing Greenland, would block access to these cloud services and thus halt the Finnish government and industry, but rather that such a possibility exists and it must always be taken into account when making decisions and presenting positions.
- Good. Let's talk a bit more concretely. One news item related to this discussion has been that in Finland, the Ministry of Justice intended to move Finland's election information system to the cloud services of the American company Amazon. A couple of weeks ago, Uutissuomalainen reported that this is now being reconsidered. For an ordinary citizen, election information, an American cloud service, and data centers located abroad or in Sweden immediately seem like a suspicious combination. But what are the actual risks involved, Jussi?
- Well, yes, many react emotionally for precisely these reasons you mentioned, but then again, if you think about it, the list of voters is public information. Candidates must be public information, and the number of votes received by candidates must be public information, in order to ensure that the elections went correctly. And in a way, what's essential is, votes in Finland are counted from paper slips. At that point, no computer program can tamper with them. So, it's just the system into which the counted votes are entered. And what's essential there is, of course, that the program is written correctly. Well, such a vote counting program, in practice, must be able to add numbers and, in the Finnish electoral system, also perform division to calculate the proportions. It's not a very complex program. Even a first-year student could probably get it right. So, then when we have a recount afterwards, the risk from a technical perspective, that someone would distort our election results, is minimal. Of course, if you want to be really paranoid, it runs in Amazon's server room, data center in Stockholm. We don't know which physical computer, we don't know who designed the microchips for it. Are there possibly some backdoors, if you want to be really paranoid, there are possibilities for influence, but given that there are recounts and so on. What was slightly interesting about this system was that the old system, which Tietoevry used to run in a server room somewhere in Espoo, was the backup system, but now in the last year, this part of Tietoevry has been sold to a British investment firm. So, now both our main vote counting system and the backup system are both in foreign hands. So, that was something that wasn't really discussed much, that we could probably do this ourselves.
- Well, more broadly, why don't we do these things ourselves? It has already been referred to that there are many different reasons, but summarize again, first of all, is it now emerging-, France, for example, announced that they are abandoning the use of Teams and Zoom, these American video calling software in state administration. But do we have alternatives?
- I think this election information system is a good example precisely because the election manager, Atte Jääskeläinen, explained the reasons for the decision very well in public. It resulted in concrete cost savings of several million euros. And such
The election system is a good example of a service that would benefit significantly from being run on a cloud service, because it is only needed once every few years. If it is now run in our own basement or in an Espoo shared data center on a dedicated server, it means that it will lie there unused for four years. The capital is tied to that rack, to that machine, but no one uses it. And then once every four years it is always woken up, and well, now it is being used. It is very inefficient in terms of resources. It is much more sensible to put it into a cloud service where the same resource, when it is not needed, can be used for others, for example, for running Omavero, because the tax authority's services are another thing that people hardly use until just before the tax return deadline, then suddenly everyone wants to use it at the same time and the demand is very uneven in terms of time. So I understand, the justifications are really good economically, but the election system is, in my opinion, a bit of an exception, that even though I am sure that there is no factual problem, in the case of elections it is not enough that experts assure citizens that yes, this is completely okay, but in the case of elections we have tried to arrange things in such a way that citizens can verify for themselves that yes, this system is reliable and works as it should. And now, if the election result is published from a server owned by Jeff Bezos in Sweden, from abroad, then it is simply fuel for various conspiracy theories, and it does not help that university professors announce that you are wrong, that there is nothing to see here. - Fortunately, we also have journalists, so in the last municipal elections my colleagues went - I don't remember anymore - in some municipalities an exceptionally large number of votes were rejected, so they went to photograph the paper votes and see if there were grounds for those rejections, but that is quite labor-intensive. - That calms down the social media groups when the media announces that everything is okay. - Exactly, exactly, but let's talk more broadly about this, a lot has been said about dependencies and risks, so do we have alternatives? Microchips were already discussed, that it is such advanced and expensive technology that the most advanced ones do not have them, but can they exist? - It depends, of course, who we are here. If we talk, think from Europe's perspective, then Europe has launched its own EU Chips Act, the goal of which is precisely to bring chip manufacturing back to Europe, to increase its share to 20% by 2030. I don't think it will be achieved, but that is the goal. In the USA, there is a similar, even with the same name, Chips Act, which has attracted chip manufacturing there. Now, what is essential is that when we go to these more specialized chips, they are not manufactured from scratch. So there are these few companies that know how to manufacture them. And the way to get manufacturing to Europe or the USA is to attract these companies to build factories, and as I already said about TSMC, that in Dresden it will be ten billion, in Arizona, USA, they have invested - well, the first investment was probably 12 billion, now it is over 160 billion, what they are investing there. With that, we get manufacturing away from the risk area, diversified. In addition, expertise is still needed, so for example, thousands of Taiwanese will apparently come to Arizona to start up the factory in the initial phase. - Exactly, and in a way, this shows the same thing that has perhaps been happening more broadly in the last roughly ten years, that states have strongly returned to industrial policy and public money is being put into it to get factories and investments back home. - Yes, also so-called Western countries, because the reason why we are in this situation now is that in Asia, China, Taiwan, Korea have invested enormously at the state level, so they have reached this situation. They have supported these companies. The game has not necessarily been fair at all compared to European competitors. Some have gone bankrupt along the way. And now, both in Europe and in the USA - well, the USA has certainly done some earlier, but especially in Europe - we are now realizing that we need state-level support to get this manufacturing here. - And how long does it take to build such a factory? - Well, TSMC decided to build a factory in Arizona, was it 2020, and the first phase started in 2024. - Yes. - So maybe it could be squeezed a bit shorter. I guess it happens faster in China, when there are probably no appeals against zoning and things are done a bit more smoothly anyway. - But that is also quite fast, in four years from decision to production. - Yes, it is. But in that case, of course, a lot is copied. So the factory exists as is, or almost as is, already in Taiwan, and then a similar one is built. But still, it is. - What about software and cloud services, do we have any alternatives? - Well, yes, the dependence on Microsoft has been discussed, and are there alternatives? The short answer is no, the long answer is maybe, and a more optimistic answer is yes. But if we now think from the perspective of an individual user or a smaller company or something like that, then all existing open-source alternatives for basic office tools - if you wanted a replacement so that I'll start using this and everything works as before, then such a thing does not exist. It would require additional development, that someone, who would then be the actor, currently, as you mentioned, the French state is trying to get rid of video conferencing software, but it is part of a larger project where it is also trying to get rid of, to bring email, instant messaging, word processing, spreadsheets into the French public administration's own use. It will be interesting to see how that project goes, but in principle there is no obstacle to it, but it is difficult for me to see who would be the actor that would bring that option to the market or through actors who would bring that option to the market, and regardless of who it is and how it comes, it would also have some cost. - And Vili, isn't there something more limited, specifically related to, for example, the tax authority or Kela or election information, so is there really something that could be done in public administration? - I would feel that we have ended up in this situation in such a way that in Finland, every authority, ministry, and so on, has been allowed to decide where to outsource its digital infrastructure, and they have all ended up with the same solution for the same reason. And then this has created such a systemic dependency. If this were done more coordinately, then we would have more market power as a buyer, a larger unit, also to dictate more terms and also a volume that could potentially help to grow these alternatives, so that there would not be just these conventional options, but that is definitely not a quick and easy path. And I fear that right now there is a real enthusiasm for digital independence, and this may lead to a small hangover when we realize that it did not happen overnight, but still I hope for the best and fear the worst. - Good. Perhaps, so that the end is as concise as possible, I will ask Mikko again: if we consider these dependencies, perhaps the biggest single or clearest one is still the chips from Taiwan, even though factories are now being built elsewhere, what would happen if China attacked Taiwan and their production and supply chains were disrupted? What would stop working? - Well, wow. First of all, let's hope that doesn't happen. That would mean that if China attacked Taiwan, it would be a declaration of war on the United States, because the US economy relies so much on chip production, which is in Taiwan. And it would mean that we would have two nuclear powers at war with each other. And the risk of escalation is, of course, what comes to mind first in this situation. But if we now focus only on the chip side and think that China somehow destroyed the very small area in Taiwan where chips are made, then that production would be immediately off the world market. The immediate consequence would most likely be various trade sanctions against China. So chips made in China would not be bought. We would be 80% of chip production gone, and then many things would stop immediately and prices would rise, and even though production has been brought to America and Europe, for example, TSMC's core expertise is certainly still in Taiwan and its elimination - well, yes, the world would go quite badly mixed up at that point. - Give another concrete example, what would we not have if we didn't get these chips? - Well, we would lose a lot of different things, because if we think about where microchips are, of course computers, mobile phones are obvious, ordinary internal combustion engine cars have 150-200 microchips, coffee makers, all home appliances, all of this would disappear if, especially China's chip production somehow disappeared from the market due to these sanctions. Here are a few examples to start with, but don't - if you still have slide rules, don't throw them away, they might come in handy. - Good, thank you very much for the discussion so far to Jussi Kangasharju, Vili Lehdonvirta, and Mikko Ritala. [music] [applause]