[Company Representative] (Sobal): [Foreign language]
[Company Representative] (Sobal): [Foreign language]
[Company Representative] (Sobal): 昨年末12月、そして3月にM&Aをさせていただきましたリソー社、プリサイス社、こちらのグループ上場企業に耐えられるような会計基準への変更であるとか、規定、そういったところに関しましての費用がかかってくると見込まれます。そういった意味で、通期に関しましては現状据え置きとさせていただいておりますが、今後もしっかりと下期収益を上げていって、皆様のご期待に応えられるような着地になるように、全社一丸となって邁進してまいる所存でございます。続いてのご質問です。「直近のM&A案件についてや、個別の売上利益への寄与額や投資回収期間についての開示方針について教えてください」というご質問です。先ほどの質問にもお答えしましたが、直近では2社をソーバルグループとして迎え入れることができました。それ以前にも、弊社グループではソーバル以外に4社の子会社がございます。こちらに関しましても、各グループ会社は今、堅調に推移しているというふうに認識しております。M&Aに関しましては、我々はお互いの会社が相互に協力して、全体で収益を上げていこうというふうに考えております。そういった意味では、グループ全体として考えて、ソーバルグループがしっかりとした成長をしていけるように考えております。また、リソー社、プリサイス社につきましては、収益に対してのソーバルグループに対する貢献は、まだ低い関係ではございますけれども、今はしっかりと準備を整えて、来期以降しっかり寄与させていこうというふうに考えております。続いてのご質問です。「配当方針に関するお考えや、手元資金の配分基準および株主優待の方針について教えてください」というご質問です。当社では、今現在、配当を年間で33円ということにしております。優待に関しましても、今現状のところを、今後もしっかりと保持をしていって、株主の皆様に安心していただけるようにしていきたいとは思っております。しかしながら、今後しっかり売上、利益ともに増加をさせていって、更なる優待、配当、そういったところを伸ばせるように、これからも事業をしっかりと伸ばしていきたいというふうに考えております。続きましてご質問ですが、「現在のソーバルの技術的な強みについて教えてください。また、急激なAIの進化に対してどのように対応し、事業戦略に活かしていく方針か教えていただけると幸いです」というご質問です。ソーバルの強みといたしましては、我々ソーバルグループでは、IT事業全般として行っておりますが、ウェブアプリといったところはもちろんのこと、組み込みですとかSaaSという知見のところ、あとはハードウェアやファームウェア、そういったところを幅広くカバーをしているというところがございます。世間一般では、そういったIT業界、ITの技術全般を賄える会社というところは数がそこまで多くはないというふうに認識しております。したがいまして、お客様からはどんな質問があったり、要求があったとしても、ソーバルグループ全社で全ての技術に対応できるというふうに自負をしております。そういったところが我々の強みというふうに考えております。また、AIに関しましては、やはり最近生成AIがここ数年台頭しておりますが、当社といたしましては、AIに関する技術というものは、さらに昔、7、8年前からずっとAIに関する開発というのに携わっています。生成AIが出てくる前の機械学習、ディープラーニング、そういったところから、我々はずっとAIに関する技術を作り上げてきました。そういった意味では、他社より先駆けてAIに関する技術者というところを育成してまいりました。そういった意味では、今後も他社とは比較にならないぐらいの技術力を持って、生成AIを使って生産性を上げるというのは当たり前ですが、今後AIに関するコンサルティングのようなところ、そういったところも我々はできるというふうに自負しておりますので、他社との差別化ができるのではないかというふうに考えております。続いてのご質問ですが、「主要取引先だったトヨタグループがランク外となりましたが、今後の車載や自動運転分野への取り組みについて方針をお聞かせいただけますでしょうか」というご質問です。ご質問の通り、今回、主要取引先の中から、トヨタグループ様が除外ということになりましたが、こちらは、やはり皆さんご周知の通り、自動運転関連というよりは、EV関連に関する各社設備投資の抑制というところがございます。そういった影響の中で、若干ですが、我々にも影響があるというところがございます。しかしながら、今後の自動車業界という中では、自動運転というところは絶対になくならないというふうに考えております。したがいまして、今後も、いわゆるトヨタグループ様はじめ、日立グループ様を関連とした自動運転関連というところは、今は若干の停滞がありますが、今後は増えていくというふうに考えております。続きましてのご質問は、中期計画として売上高100億円を目指されていますが、達成に向けた主な課題や想定リスク、もし可能でしたら、その対策について教えていただけますでしょうか、というご質問です。当社といたしましては、中期計画として、売上高100億円を今期中に達成するという目標を掲げております。中間期、第2四半期では、売上高50億を超えることが無事できました。そして、通期におきましても、この100億円というところを一つの目標とは考えております。しかしながら、リスクといたしましては、第2四半期にもありましたように、検収ずれというところがございます。そういった意味で、最終的に我々、2月決算の時の着地の状況によっては、万が一ですが、検収ずれ、そういったことのリスクにより100億円を下回ることがあるかもしれませんが、我々はそういったことがならないように、これからしっかりと売り上げを積み上げていって、今年の2月の決算時期にはしっかりとこの目標を達成できると信じております。続いてのご質問です。売上高以外の利益目標や資本効率指標の具体的な数値での目標設定について、現在の考えをお聞かせください、というご質問です。繰り返しにはなりますが、先ほど申し上げました通り、まずは今期の売上100億というところを目指しております。しかしながら、今後は、100億を達成した先には、今後しっかりと収益というところも上げていく必要があるかと思っております。現状6億7,000万円という数値ですが、こちらをさらに上げていく、増収増益を続けていくということが必要になるかと思います。そのためには、我々が得意としているAI駆動開発、そういったところや、AIのコンサルティングによる高付加価値な案件、そういったところで収益性をしっかり上げていくということが、今期、さらには来期以降必要となると考えておりますので、今後もしっかりと来期、再来期、増収増益を目指していきたいと考えております。続いてのご質問です。株価や株式の出来高に対する認識と、情報開示の充実、並びに株主アンケートの活用についての取り組みをお聞かせください、というご質問です。現在、ソーバルの株価に我々が満足しているかというと、そういうわけではございません。しっかりと、今後も収益を上げて、今の株価以上になるように、グループ全体を上げて邁進していきたいと思っております。やはり、収益を上げるということが、一番株価を左右する要因だと思っておりますので、しっかりと利益を上げていくということが今後必要かなと考えております。また、株主アンケートの活用ですが、冒頭申し上げました通り、今回1,000件を超えるご意見をいただきました。その中から、今回このようなご質問やご意見について、この場でしっかりと回答をさせていただくというような形で、株主アンケートの方を活用させていただいております。今後も、しっかり、もし株主の皆様がご意見やご質問等ございましたら、お遠慮なくアンケート、並びにIRの方にお問い合わせをいただければと思っております。以上、お寄せいただいたご意見やご質問について、お答えさせていただきました。今回用いた決算説明資料は、ソーバルのウェブサイトに公開しておりますので、ぜひダウンロードしてご覧ください。なお、後日となりますが、質疑応答の要約テキストも公開いたしますので、併せてご覧ください。本日は貴重なお昼の時間帯に、ソーバルの決算説明会に最後までご参加いただき、誠にありがとうございました。今回の説明会の感想や今後に期待することなど、ぜひ皆様のご意見をお寄せください。それでは失礼させていただきます。ありがとうございました。
[Company Representative 1]: at this point, but there is an important sovereignty discussion here, and that means building as much as we possibly can in Australia, whilst recognizing we can't do everything. Ultimately, it's about preserving our options and agency in areas of strategic importance, and AI is as strategically important as anything the world is looking at at the moment. I would argue it's the most strategically important technology in human history. As I say, it means doing as much as we possibly can here, but for those things that we can't do here, we want to form reliable partnerships with countries and partners that can actually provide those. Obviously, we are in an advantageous position in that our chief ally is also the leading country in providing these technologies. The benefit of building training compute here in Australia is that it meets both of those objectives.
[Company Representative 1]: We are doing something here for ourselves. We can build an industry around that. That will improve our domestic capability, and we are also securing access to frontier capabilities from a reliable partner, but with some assurance underpinning that in the form of our making a contribution to that relationship. To come back to the question from a national security perspective, I would say that that ticks all of the boxes that we want to be ticking from a national security point of view.
[Company Representative 2]: Chair, if I could just respond a little to one of Richard's points. I agree that there is a distinction between what the intelligence community needs and what the public needs. Where I fundamentally disagree is on the issue of public trust. Completely agree we need public trust, but we can't wait for public trust when we're dealing with something that is a geostrategic issue, and is a national security issue. It is government's responsibility to build the trust and not wait until the public has that trust. No other field of defense or national security should we be waiting to get the social license before government acts, and I think that is shown in history. If we look at World War II, the Americans had to wait to enter World War II until there was social license, and it took Pearl Harbor.
[Company Representative 2]: We don't want to do that again.
[Company Representative 3]: Sorry, can I just add?
[Company Representative 4]: Yeah.
[Company Representative 3]: Just to add to the agreement. Look, to be clear, I don't think the intelligence community needs to wait to get public trust because they'll never tell us what they're doing. I'm okay with that, but that is not the proposition that is generally being put. I'm not trying to put words in ASPC's mouth, but generally, it's if we don't say yes to everything that the AI industry wants, the intelligence community will miss out.
[Company Representative 3]: I don't disagree with a word Justin just said.
[Company Representative 3]: But I think that the needs of the intelligence community for access to compute are unrelated. Unrelated to the claimed productivity benefits and other benefits for saying we can't regulate this technology, either in terms of untrammeled rollout of data centers or restrictions on the use of fabricated content. I could not agree more. I just think that is a small part of a much bigger conversation, and the FOMO is being used, I would suggest, to railroad an important public conversation where trust is key to where data centers are built and how they're built and for what.
[Company Representative 3]: And again, we've always separated the intelligence and defense establishment from those decisions. Imagine nuclear technology was just invented. Just wind the clock back. We've just split the atom for the first time. Would we say, "Well, let's just let everybody play with it, and we'll see how it goes, and we don't want to impede innovation?
[Company Representative 3]: Of course we wouldn't. But would we let the Australian nation-state play an active role in it? Yes.
[Company Representative 3]: We have been here before, and I think the biggest threat we face in regulating this is conflating all these different elements
[Company Representative 3]: of what is an important and complicated issue.
[Company Representative 4]: Yeah. I would agree. I think I have made the point consistently through some of the questions that I have been engaging with previous witnesses about the critical nature of social license, around how do we bring people, how do we bring the Australian public with us in terms of how it is directly affecting their lives. Distinctly from obviously what happens and the importance of having access to the frontier technology for our defense, which is also fundamentally about building trust for Australians as well, knowing that we have got the capability to keep ahead of our adversaries and making sure we are not missing out on that. But to your point, how do we make sure, and I think that is part of the conversations we are having around what do the settings look like, what should they look like?
[Company Representative 4]: How do we make it, as we consistently have said or previously said, around shaping this and not just let it happen to us. With all of that in mind around how do we build that social license, because fundamentally, I do not think we actually get anywhere near the benefits proposed, whether they are real or not, if people do not want it or trust it or understand how it is used. I really respect that and take the point and the sort of juxtaposition between the two, but also the extent to which they are concurrent as well. I might trade my Deputy Chair.
[Company Representative 5]: Sure. Thanks, Chair. Dr. Dennis, you drew a distinction between national security and defense and then sort of the rest of society. What is your view on, for example, organizations, whether they be government or private sector, that are custodians of data, being able to access and use AI?
[Company Representative 5]: What is your view on that?
[Company Representative 3]: Oh, I think most of them already are, and if they don't think they are, they probably don't understand how their data's being managed. I think AI, it's such a broad term. It's being used by organizations often when they don't even realize it.
[Company Representative 3]: To be clear, I like technological progress. It drives a lot of productivity growth, and in turn can significantly improve the wellbeing of individuals and countries.
[Company Representative 5]: Just on that, look, because the point's been made around trust, of course. If I'm an individual who's got my health data with my private health fund, and that is storing my data overseas as opposed to onshore, for example. I'm just wondering how we align what you said around it's okay for security and defense to be utilizing it, that necessitates compute build out here in Australia, data centers and all that goes with it. But for example, when it comes to a bank, a health insurer, the Department of Health or Social Services or whoever it might be, because in order to do the same for them here in Australia, we shouldn't be building things all over the place, your contention. We would then be sending things in a big cable underwater to the US or to China or to wherever it might be stored.
[Company Representative 5]: I'm just wondering how we deal with trust in that context.
[Company Representative 3]: Oh, sorry, I misunderstood. Yeah, look, again, I think it is very important. Data centers, they exist now. There never was a cloud. People have just figured this out. There was a giant network of sheds.
[Company Representative 3]: Anyone who has been talking about the cloud for the last decade has been talking about large sheds somewhere else.
[Company Representative 5]: Yeah.
[Company Representative 3]: I am 100% in favor of some of those sheds being in Australia, and I would prefer that information that is significant.
[Company Representative 5]: How many? Do we have a cap on how many we should have in Australia?
[Company Representative 3]: Well, yes, but I was going to go to category. Do I think that defense and intelligence should have access to the highest, the best models and the fastest compute? Yes, absolutely. Would I prefer Australian citizens' tax data is stored here in Australia, if possible? Yeah, absolutely, I do. But again, we need to think categorically about what are the uses to which things are being put. Because I put it to you that every data center proponent is effectively suggesting that without it, there will be no national security, and all of our tax data will be stored overseas when actually maybe that's where the cat videos are getting made.
[Company Representative 3]: I think it's actually really important to build trust and to build public understanding about what is important new infrastructure for us to be far more categorical and admit some of this compute is getting used for national security, some of it's getting used for research, some of it's getting used and some of the storage is being used for what we would think of as confidential, if not top-secret information. A lot of it is storing 10 years of old emails and whipping up fake political images for ads. I think it's okay for us to be upfront about that and say,
[Company Representative 5]: Yeah
[Company Representative 3]: You know what? We've got enough data centers to store the ATO data and the health data and the math, and if we want to build a lot more for those other reasons, we'll get in the queue." There's competing uses for land, there's competing uses for labor, there's competing uses for electricity. Those cat videos don't have to be made here in Australia.
[Company Representative 3]: Again, the debate is so overwhelmingly, it's so totalizing. If you don't like data centers, you don't care about national security, or you don't care about the confidentiality of Australian information. I think a lot of sensible people can see all of these categories quite differently, and I think the parliament and the public would be well-served if we broke it up like that.
[Company Representative 5]: I am struggling to think of a witness that has actually made the same point in this entire hearing around the fact that we do not need to try and catch the train and do what a lot of proponents from the national security witnesses we have had through to the tech providers, through to news organizations, academia. The Academy of Sciences themselves said exactly the same thing. To that end, if your contention is to be pursued to its conclusion, I am wondering where you draw the line and say, "No, you can not have that data center because it is going to be used to make videos of cats.
[Company Representative 5]: Versus, "Oh, you store tax data." How do we recommend to government how we delineate between cat videos and tax data according to the Australia Institute?
[Company Representative 3]: Yeah. Fair question. I am surprised that no one else has attempted to make these points.
[Company Representative 5]: All right.
[Company Representative 3]: I do not mean to be flippant when I say this. I actually think it is kind of the job of the committee to think about how you might draw those lines. I am not here with a concrete proposal on how you should, but I think unless we can, then people will not only lose faith in AI, they will increasingly lose faith in the ability of our governments and our parliaments to stand up for us. Because people do not actually believe that all the data centers are mission-critical for national security. They do not. If we
[Company Representative 5]: Not when we tell them they are being made to make cat videos.
[Company Representative 3]: Well, where the lines are drawn, I agree, is difficult.
[Company Representative 3]: Unless some lines are drawn, people will not accept the Well, let me phrase it in perhaps more concrete terms. There is a real risk the intelligence community will miss out on what it wants. There is a real risk that the banks and the insurance companies will miss out on what they want unless the public believes that there is some form of stratification going on here. I am not suggesting it is easy, but unless we are honest with people that not every data center is a critical piece of national infrastructure with defense and confidentiality significance Unless we can find a way to have that conversation, I fear that the lack of public trust will actually harm what it is that ASPI are here telling us we really need to do. I am not suggesting it is easy
[Company Representative 5]: Is it your view that data centers are used for one purpose exclusively, or do they have multiple applications through their compute?
[Company Representative 3]: Look, that's a very good question. A lot of people with far more knowledge of it with me will say they have no idea because we're not allowed.
[Company Representative 5]: I'm asking you, though.
[Company Representative 3]: Well, but I guess I'm saying that there are people that own data centers that I've spoken to that said, "We have no idea what's going on inside our own data center.
[Company Representative 5]: The point you have been making is that we can have them if they are used for certain things, but should not wave the green light around for them if they are not used for those purposes.
[Company Representative 3]: Yeah.
[Company Representative 5]: A lot of evidence has suggested they are used for multiple purposes.
[Company Representative 3]: Yes.
[Company Representative 5]: You might have 50 clients doing different things through that center. One might be a cat video maker, the other might be ASD. I am just trying to figure out how we deal with that.
[Company Representative 3]: Well, again, at a technological level, I am the last person to ask. I am not pretending I came here with an answer to that question. Again, that I think raises exactly the point that I am trying to make, that no one is in charge of this if no one even knows what is going on inside the centers that they literally own. Is it possible that the cat videos and the ASD computer sitting side by side? Of course it is. Again, if we are going to tell the public, we are going to need to multiply the number of data centers tenfold, the electricity demand tenfold.
[Company Representative 3]: If we're going to take people into that project, we might be able to get away with it today, but as this rolls out, as the electricity price rises, I think if we don't confront these categories now, we will be having much more difficult policy and political discussions in three or five years' time.
[Company Representative 3]: But I agree it's difficult, I really do.
[Company Representative 1]: Mr. Rowe.
[Company Representative 1]: I'd like to just make a point, thanks Deputy Chair, that no one I know who believes in Australia carving out a strong role for itself in AI has the faintest interest in data centers for the purposes of storage of cat videos or emails or 10-year-old emails or anything like that. The interest is in data centers that are optimized for artificial intelligence, and particularly for training. Now, yes, there are different types of data centers. There are different types of chips. Again, I'm not a super expert, but there are CPUs, there are GPUs, there are TPUs. GPUs, the high-end things that NVIDIA designs, they are very sought out. They're about the most in-demand products on the planet at the moment.
[Company Representative 5]: Mm-hmm. Yeah.
[Company Representative 1]: which is why NVIDIA is the most capitalized company on the planet. TPUs, tensor processing units made by Google or Alphabet, they are what we are talking about here, and that is really what we are talking about. Yes, they can be used for different things, but you would be wasting a GPU by using it for storage, for instance.
[Company Representative 5]: Yeah.
[Company Representative 1]: We are very much talking about attracting a very specific type of data center investment here with a very specific purpose. One is to build local innovation clusters around that, and I think history is pretty kind to the idea that if you build an anchor point, then an industry tends to flourish around it. Look at Silicon Valley, for instance. The other one is that we build those partnerships. Some people are talking about it as leverage.
[Company Representative 1]: Some others prefer to avoid terms like leverage because I do not think it should be a perpetually adversarial position where we are constantly saying to the Americans, "If you don't give us the latest version of Astra GPT-6 or whatever, we will cut you off." Just as with Pine Gap, we don't tell the Americans every second year we will close Pine Gap unless you give us the latest whiz-bang radar, for instance. The ideal framework is to think about it as an evolved partnership where we have an understanding and we share those two things, and we each make a contribution to the relationship. That is very much what we are talking about.
[Company Representative 1]: The broader public concern around data centers, for all of the local reasons that people have concerns about them, power prices, they are aesthetically displeasing in some places, they can be noisy, they can consume water, all of those things. Those are all legitimate, and those should be managed. I think it is very much this is an AI data center proposition that we should be focused on.
[Company Representative 5]: Yeah.
[Company Representative 2]: Deputy Chair, if I may. I do agree. I think what Richard was saying around data centers and critical infrastructure, I think it is incumbent on us to know what we view as critical infrastructure versus just infrastructure. I think one of the lessons we can learn from the last 20 years is that we did not have those lists. We did not understand what we viewed as critical, whether it be technologies or infrastructure. I think that empowers us to be able to separate a building that is being used for a particular purpose and a building that is being used for something that is nationally sensitive or nationally critical.
[Company Representative 2]: I do think that it is incumbent on us, whether it be Parliament or the different sectors, to be able to say, "This is what we mean by critical infrastructure." I do think a subset of AI data centers should be viewed as critical infrastructure, and then to have the flow-on impact of that it shouldn't be owned or operated by a foreign company or a foreign adversary, those types of issues. It would also mean that if you are not classified as critical infrastructure, then I think part of what Richard is saying is you then have to prove why you might be proposing to set up a bit of infrastructure if it is not going to be determined to be critical.
[Company Representative 5]: Sure. Thanks.
[Company Representative 2]: Thanks, Chair.
[Company Representative 6]: Thanks, Ms. Cheney.
[Company Representative 6]: Thank you very much. First, Richard, I am interested in just understanding your perspective on this more because it seems like you agree that there are some zero-sum domains where small capabilities make a difference, such as national security, cyber, where enduring access to frontier models will be important. I guess I am interested in then testing, do you also agree that we do not control those models and we cannot replicate them at any feasible cost? So we have a broad national interest in ensuring that we have enduring access to those. Then do we then need to have a bargaining position or, I think as ASPI put it, a partnership where we both bring something to the table. Wouldn't holding a key part of the supply chain give us that bargaining chip?
[Company Representative 6]: Isn't the binary nature of access to frontier models in the context of potential recursive self-improvement a big enough win that means it justifies building a part of the supply chain in Australia so we continue to have a seat at the table and also have a seat at the table on determining global standards so that we have a little bit more power to actually control the direction that AI might be heading in?
[Company Representative 3]: Yeah. I guess, thank you. There's a lot in that question. I might not get all of it in my attempt at a short answer. To be clear, do I think that Australia should facilitate the construction of some data centers for some critical reasons? Yes, I do. Do I think there's benefits to Australia in trying to enmesh itself, in the AI industry? Yes, I do. Where I think there's a lot of, and I'm not an expert, but I've been reading a lot for a long time about this. Where I think there's a lot of confusion, which perhaps the committee members don't have, but I do, I think there is a lot of confusion about what this supply chain is. I think there's a lot of confusion about what AI investment is.
[Company Representative 3]: I think there's a lot of confusion about what it is to train a model and what the lasting benefits of those models are. Now, again, if the committee doesn't share my confusion on those key terms, then that's okay. We can-
[Company Representative 6]: I think we've heard evidence on all of those issues.
[Company Representative 3]: I've heard lots of evidence.
[Company Representative 6]: And there is uncertainty around them as well. They're not all settled conclusions.
[Company Representative 3]: Yeah. The reason I say all that is because, and I understand the question you're asking, but I'm limited in my ability to answer it because it's not at all obvious to me that the ownership of a concrete slab and a shed on which some GPUs that may or may not be owned by Australians, it's not at all obvious to me that the ownership of that piece of physical infrastructure gives us really anything at all. Now, it may or may not, but the mere act of laying the slab and installing those GPUs, it might feel like, and we might describe that as placing ourselves in the supply chain of AI, but without far more information than I'm able to get my hands on about the nature of contracts and the nature of the economic relationships between all the people in this supply chain.
[Company Representative 3]: I think the metaphor of supply chain, if we think of it in terms of how a fighter jet is made and what the role Australia plays in the supply chain of a fighter jet, as an economist, I understand that quite well. But we're using metaphors like supply chain in AI when it's not at all obvious to me that the economically significant parts of the industry are in any way linked to who owns the slab and where the GPUs are. That doesn't mean we can't do it.
[Company Representative 6]: Yeah, I suppose, there's the geostrategic and the economic arguments which most people have treated as being separate. And the position we find ourselves in is we are able to impose some conditions on that happening. So if we've got something that someone else wants, we can actually say, "You can do it on these terms," and determine what are the terms on which it is good for Australia too. And I think that's what the committee is trying to get to is what are those conditions that make this a good move for Australia, both economically and strategically. I'm interested in, maybe this is a question for ASPI. And you say, you're talking about hosting giving us a stronger bargaining position or an evolved partnership, to get frontier model access.
[Company Representative 6]: I'm interested in your views on, with the June export control that happened, the extent to which you think hosting would have protected a country from being subject to that export control that the US imposed in June. Just to understand what is the bargaining power then, and can we imagine it in real terms if the thing that we're concerned about is the US government saying we're withholding access to this model?
[Company Representative 1]: Thanks, Ms. Cheney. Yeah. Look, the answer is, common sense tells us that if we are instrumental to the development of current and future models, we are in a better position to say to both the frontier companies and the US government that we are entitled to privileged access, early access, and some sort of parity position with US users. I think that is just a pretty logical position. The exact mechanism through which that would be done, I think that is a discussion that a lot of people are having at the moment. ASPI is actually doing some research precisely on that at the moment, which, unfortunately from the committee's point of view, we will report in early December.
[Company Representative 1]: But we are looking into precisely what those government-to-government agreements would look like, what the commercial agreements would look like to make sure that there is some substance behind that. But I think, it is a reasonable, logical statement to say that that's the case. Can I just come back to your previous question, Ms. Cheney
[Company Representative 6]: Yeah
[Company Representative 1]: if I may? My answer to your previous question was just a resounding yes in the sense that the role of frontier, and you mentioned recursive self-improvement as being relevant here. I think the role of frontier is going to become immensely powerful in a pretty short window ahead of us. We're talking about 12 to 18 months in which. That's the prediction window at the moment coming from the frontier about when recursive self-improvement might become possible. There is new data coming out.
[Company Representative 1]: There was something that came out just this week that showed what is called research taste in which frontier models rather than just writing the code and doing the mundane routine stuff, but actually having the new insights to say, "Hey, what new ideas can we apply to iterate current frontier models to design the experiment to test the efficacy of that and then to analyze the experiment results afterwards?" Frontier models are getting much better at that, and that is just in the last 10, 12 months that that has occurred. 2026 has been a remarkable year for AI progress, particularly at the frontier. I think that in 12 months from now, we are going to be seeing models that we would think of as science fiction superpowers this year. I do think that is immensely important. We have talked about FOMO.
[Company Representative 1]: I could be wrong about this, but let us say there is a 50% chance that I am right. I think we should feel FOMO about missing out on the opportunity to be part of something that is going to be that powerful in such a short window from where we are sitting today.
[Company Representative 6]: Can I just add-
[Company Representative 3]: Are there Yeah, sure.
[Company Representative 3]: Sorry. I largely agree with all of that. But again, to your question, should we try and get as much out of it as we can? Of course. I guess conceptually what I am trying to suggest is who is it that we would be having this negotiation with to get the best we can? Is the Commonwealth of Australia talking to Anthropic? Is the Commonwealth of Australia talking to the US government? Are we assuming that all the frontier models in 5 years' time will be coming from the US? Are we assuming that Anthropic won't be broke in 3 years and a company we haven't heard of yet will be the company that we will wish that we had? So if your question is should we try and get as much out of it as we can? Yes. A resounding yes.
[Company Representative 3]: I just think it is harder, conceptually harder and technologically harder and politically harder to answer the, is having the slabs and the GPUs going to get us our fair share? Because we have to have a really sophisticated understanding of fair share of what and from who. And if the US can renege on its free trade agreement with us in a heartbeat, what deal is it we think we are going to write that will lock the most powerful country in the world after it has got the most powerful technology in the world? What piece of paper is it we think we are about to write that is going to protect us and ensure that they share fairly with us in 5 or 10 years? It is just bigger and harder than we are talking about.
[Company Representative 6]: I take your point. It is pretty hard. Unfortunately, we do not have an alternative ASPI in terms of the circumstance that we are trying to make decisions in. Another question for ASPI. You talked about funding the AI Safety Institute to do independent frontier evaluations, cyber and bio-assessment, incident analysis, and pre-deployment access in your recommendation 2. What sort of capability uplift do you think is required for it to fulfill that role?
[Company Representative 1]: In dollar terms, how would you like me to quantify it?
[Company Representative 6]: Whatever terms you are happy with.
[Company Representative 1]: Sure. Look, I have seen some of the other figures that have been mentioned. Look, the UK is the gold standard here. I think matching the UK in some sort of proportional GDP terms is a pretty good way to think about it in dollar terms. Look, I suppose the people are going to be absolutely critical here. It is a very select set of skills. The reason the UK Institute has been so effective is that they have been able to convince people fresh out of the frontier companies to come and work for them. People of that sort of level of experience and talent are really crucial.
[Company Representative 1]: I think we should be aiming to get some of those Australians who are currently working in Silicon Valley, and there is a bit of an Australian mafia at the moment, I mean that in a kind way, working at the frontier labs. Some of them are possibly about to cash out with pretty significant shareholdings if some of the frontier companies do go public. Perhaps out of a sense of national duty, they can be persuaded to come back and work for something like the Safety Institute. I think focusing on people is absolutely going to be critical. Then what can we do?
[Company Representative 1]: In terms of what particular work they do and what the uplift is there, I think some of the things like agent governance is something we have already established and runs on the board for, and that is only going to become more and more relevant. If you are working in agent governance at the moment, then you are doing really, really important work. There are organizations like GovAI in the US that have done some great work on this. I think the Australian AC should look at partnering with organizations that are doing similar kind of work. That should include the UK AISI, and it should include organizations like GovAI as well. That is where I would like to see them put their priorities. Like I say, I think something roughly comparable on GDP terms with the UK Institute would be useful.
[Company Representative 7]: Sorry to cut you off just a bit.
[Company Representative 1]: Yeah
[Company Representative 7]: We are right beyond time now, and I know Senator Sheldon has one question that he would like to be able to ask. Perhaps a response short and sharp, but perhaps the rest, a more fulsome response on notice would be helpful. Senator Sheldon.
[Company Representative 8]: It is good to see you, Richard. Probably to Dr. Dennis, and by all means, others might want to make some comments as well. I might just go to, first of all, that last question just briefly. There is a lot of the question you sort of raised as a figurative question to the committee. A lot of the evidence we have received actually has answered to a degree that proposition you put. I would really strongly suggest to have consideration to that evidence we have received on the committee, because it does go to a lot of those questions. I just want to go back to what you initially said right at the beginning of your presentation, and not only looking at I am saying it in my words, but not only looking at the technical questions but looking at the humanitarian questions, the impact on society.
[Company Representative 8]: Workplaces, of course, as you know, amongst others, have a deep interest in. But can not we walk and chew gum at the same time? This is really a matter of the data centers have their place. The important issues we have got with copyright have critical questions that we have to resolve there. Data centers have critical questions. Workplaces have critical questions.
[Company Representative 8]: Next year, we are going to have, supposedly next 2028, supposedly driverless vehicles being tested more substantially in Australia. Critical questions about 125,000 taxi drivers and rideshare drivers. How does the economy possibly absorb that? I think the question you raised, and that is that are we preparing ourselves for where there are impacts on parts of the community? Have we got the training, retraining? Who is paying for it? Do we want these companies here in the first place? I think most Australians do not want to.
[Company Representative 8]: Some of the surveying that I'm seeing, most Australians don't want to go into a driverless vehicle. They don't want to see all the profits and taxes go over to the US whilst we lose all those jobs, opportunity, income, and taxes out of our economy. That's a personal view. There are lots of questions to be answered, and I agree with you, but isn't every piece of this has its own particularly innate importance. The part that I particularly liked about the fact you're highlighting the fact there is a critical piece that we need to be thinking about and constantly thinking about what impact is it actually having, not on building a data center, as important as that is.
[Company Representative 8]: Not only about what happens with copyright, critically important as that is or no change, but what does it happen to the rest of the community, whether we have data centers or not, or whether we have copyright or not. Is that the point you're also making to us? That don't get fixated on lots of issues all at once.
[Company Representative 3]: Yeah. Look, absolutely. Again, to be clear, I'm not anti-AI. I'm a user of AI. The Australia Institute's initials are AI. I'm not anti-AI. I think that the public, not just in Australia, but around the world, are being railroaded through this sense of FOMO. That doesn't make the technology bad. But I do think that, as Bill alluded to at the beginning, we are going to have elections where fake depictions of real people pretending to say things they've never said before are going to dominate our analysis of what's happening in an election. That's a national security risk, I put to you.
[Company Representative 3]: Our democracy, and it's the AI industry saying, "Oh, there's a 10% chance we might end humanity." If I came in here and said that, people would suggest I'm being hysterical. I have no idea. Maybe it's only a 5% chance, right? I'm not anti-AI, but I'm deeply concerned-
[Company Representative 8]: I'm not putting to you that you are. I'm just saying that you're raising, in my mind, a legitimate series of questions that the committee should come to grips with about what we do about it now in any interim report and also what we recommend for future
[Company Representative 3]: Yeah
[Company Representative 8]: steps as well. I think the broad subject matters you're raising is that question about we need to walk and chew gum at the same time.
[Company Representative 3]: Yes, absolutely. And we already are. Again, we've been rolling out AI for five years. We're talking about speeding it up. We've been rolling out data centers for 10 years. We're talking about speeding it up. The way the debate is
[Company Representative 3]: often projected as a binary. Are we in or are we out? Should we start or shouldn't we? When the reality is we've been doing a lot of this, it's already having consequences. Understanding the limitations of what any one committee can do, I would urge you to look at those kind of big questions that I put at the beginning. I think part of it is to say to other parts of this parliament, other parts of the parliament need to be getting their head around this pretty fast. Because again, I'm taking the AI industry at their word. If these companies are worth JPY 1 trillion, it's because a lot is going to happen really fast. If they're right, a lot is going to happen really fast and we're not ready.
[Company Representative 3]: If they are wrong, then we shouldn't be rushed into doing everything they are telling us. This is a real challenge for our parliament. Again, it is a lot harder than is it good or is it bad? But literally, those valuations of those companies suggest a lot is going to happen in a very big hurry. I don't think that most of the public are aware of that. I don't think most of the parliament are aware of that. I don't think most government agencies are aware of that. I don't think you will be able to answer all these questions. But I do think that what you are in a great position to do is to tell everybody that they got to get their skates on.
[Company Representative 3]: Not coming out with the answer or decide whether we are pro or anti.
[Company Representative 3]: But to start literally thinking if the projection, if Anthropic is going to be that big in 5 years' time, if their revenue is that big in 5 years' time, where's the money coming from? It comes from not paying workers. The only way the Department of Health can spend JPY 1 billion on AI is if it spends JPY 1 billion less on humans, or if you guys increase their budget by JPY 1 billion.
[Company Representative 3]: I don't know what the Department of Health's plan for AI is. Maybe there is one. What is the ATO's plan? What are the state government plans?
[Company Representative 3]: But again, all that money that Anthropic is betting on is coming from somewhere, and that is going to have very real consequences for individuals, for communities, for the economy, for the way Commonwealth budgets are drawn up. At the moment, most of the people talking, sounds like most of the people giving evidence to this are people who are really interested in the tech. Well, I am really interested in the macroeconomic consequences.
[Company Representative 3]: I am really interested in the democratic consequences. Again, thanks for inviting us along.
[Company Representative 8]: Sorry, it is that thing about scaffolding.
[Company Representative 3]: No, that is all right.
[Company Representative 8]: So it is this thing about scaffolding and about how we deal with the ongoing challenges that scaffolding should give us the capacity to do whilst we are trying to predict, prepare
[Company Representative 8]: and respond.
[Company Representative 3]: Yeah.
[Company Representative 3]: Very much so. If you guys do not do it, the question is who is?
[Company Representative 7]: Who is? Yeah.
[Company Representative 2]: Commissioner Sharon, I will promise to be 30 seconds, Chair.
[Company Representative 7]: Okay.
[Company Representative 2]: For me, it is not so much a FOMO. I do not think we will miss out. It is a question of what we will get anyway. If we do not do it, in my view, we will get the Chinese technology if we choose to do nothing. I know that is not what Richard is saying at all. We are not going to miss out on this technology. The technology is going to come here. We just have a chance to determine what that technology is and how much control we have of it.
[Company Representative 7]: Yes. I will bring the session to an end. We are quite a bit over time, but really important conversations and important contributions. So thank you very much to both of you. Anything that you have indicated that you would like to provide further on notice, please do so through the secretariat within the next two weeks. But again, thank you very much. I will suspend for about 10 minutes before we start with our next panel. Thank you very much.
[Company Representative 2]: Thank you, Chair.
[Company Representative 4]: Yes.
