Dr. Mark Esposito: Rethinking the GenAI Value Proposition Towards a Human Centric Society

Thank you for having me. This is the second time that I make an appearance in an event organized by your university. And so I'm very happy that we have an opportunity to talk about a big conversation, which is where are humans in the loop when we are dealing with the conversation related to GenAI? And the question will be about, Can we think about introducing more human-centric societal implications in what we do? If I look at time, I'd love to have this conversation for a little bit and within reason, also some questions. That would be incredible if we could do that. So allow me to move my slides and hopefully you can see me moving the slides as well. So I take it as a yes. I think it was important for me to start with where I'm coming from, from a perspective of defining artificial intelligence. I think it's important that I tell you why do I research, write, teach, train in this field. One answer to that is my academic work across my career. Technology has become more and more important. And although my training was originally in economics, I started to integrate more work in the impact that technology has on a global level. From my non-academic side, I have co-founded an AI company in 2016, which went from small to big. And I'm now launching two spinoffs from the same company. And the idea is really to start understanding from the practice how this powerful architecture of technologies is really fundamentally changing us. So when I think and talk and research and to some degree present works on artificial intelligence, I kind of look at this definition as the one that is closest to my own experience. I see artificial intelligence fundamentally as a powerful computational model inspired largely by statistical structures that mimics cognitive function. Now, clearly, if the future of artificial intelligence will likely integrate physical and digital, which is very likely, in which robotics will simply become smart or smarter than what it is right now, it is still important to create a clear demarcation line on the fact that we're dealing with a technology that is designed to mimic the brain. Now, it might feel special if I tell you that we're mimicking the brain, but a calculator is mimicking the brain and is doing calculus better than us, and it does it in a fraction of time. So the fact that we have computational capacity that is able to emulate what the brain does, it doesn't replace humans from a perspective of being the center of life on Earth. On the contrary, is the question about have we done enough to empower humans with the technologies we have to address some of the major challenges we're currently facing? You'll see from some of the conversation that we will have that the driving, and I would say dominating narrative that we are coming from is economic efficiency, cost saving. So like an incremental/decremental binary model, where we were defining our value system based on some form of measurements that is quantitative. A large part of this is the fact that we are using technologies to improve what it would take us longer to do. There's nothing wrong with that. But the assimilation of technology that's so powerful to intelligence is far-fetched. It is far-fetched in 1956 when Andrew McCarthy brings the conversation for the first time to the table about artificial intelligence mimicking cognitive functions. It is far-fetched today where a lot of sensational media propaganda is fundamentally trying to create an equalization between humans and machines. So I'll start from where I'm standing from. It's my view, but it's the view of the work I do. There is no intelligence in artificial intelligence. It is nothing more than advanced statistics and computational power that tapping on data is capable of replicating an incredible amount of content generation because it's working on the ability to narrow down a specific scope into extremely accurate outputs. And so there is a clear fascination about what the technology can do. But look, an airplane that is flying in the sky is equally fascinating, isn't it? It's not that we consider it to be smarter than us. It just has a function that is about mobility. So in this case, artificial intelligence does an incredible amount of work in really redefining the industries in which we operate. Likely we see this coming into the public sector very soon. One of the reasons why I'm in Taiwan was presenting a paper on the governance of technology based on how do we protect ourselves from mitigation of— actually from risks that are coming from the use of generative AI in governments. So clearly we are at the very beginning of a societal and epochal shift. But it doesn't necessarily mean that humans are going to be removed from the conversation. The conversation I'd like to have with you today is the opposite. How do we basically wire them back? Now, again, where do we stand? It's clearly shifting from what we consider to be the overarching idea about artificial intelligence to now where we see more and more multilayer neural networks which are empowering the deep learning revolution to become much more applicable. For a decade, from 2010 until 2020, the dominating logic of artificial intelligence was either machine learning, the majority of that, but it used to be on a binary structure. So it used to operate mainly in what you hear often, supervised, unsupervised. Reinforcement models. And we already had deep learning, but deep learning was a much more complex function to perform because the conversation about AI has always been about the barriers of entries have always come by the representation of the industry into very few players. And so it wasn't possible for a small-medium enterprise to simply think about artificial intelligence. It was too prohibitive, it was too far, and there was a clear cost discrimination. Now today, it doesn't necessarily feel any better. I know that you might feel that because we have access to generative AI, our life has changed. Well, our life has changed up to a point, but we're now facing a different set of challenges, which is how much can we trust this to integrate it into some of the function we are going to see redesigned? And there's a conversation we'll do together, but Fundamentally, it's important to look at the historical evolution of artificial intelligence first as an architectural technologies, at first coming from mimicking— well, at first it was a rule-based, then it started to shift over to neural language processing. Then we started to look at the brain as a way to replicate trial and error. Then we got into machine learning, major breakthrough 2010 in Canada. And then today we are now dealing with the deep learning in which multiple forms of conversation on the data, large language models, big data, small data, scattered data, generative adversarial network, retrieval data. All this narrative is now within a larger application of deep learning. Now, the reason why I'm telling you this is just to make sure that when I'm using specific terminology, we tend to have agreement on what I'm referring to. At least you see where I'm starting my conversation from. When you're looking at this slide, this is largely coming from the US military compound called DARPA. We consider today artificial intelligence right at the very middle or to some degree in a blossoming phase related to statistical learning. Today, what we see is an incredible ability to use machine learning or deep learning to address many different challenges that can be positioned into a statistical equation. And we see that when we are able to integrate this in more than one industry, we tend to see the distribution of this technology to millions of people. So today, the larger pilot of data labeling is not coming from research companies or from surveys with primary data. It's coming from users who are prompting the machines every single day as they are using the different kind of open-source GenAI portals to facilitate some of the tasks It did happen a lot that we have seen significant integration of generative AI into the workplace. Although to this day, the challenge is to understand what is the right modus operandi, what are we using it for, where do we eventually incur into possible risk, where are we creating a fabrication of information that are going to keep on labeling the data incorrectly. How do they differentiate information from truth? So there are many questions that the use of generative AI has opened. We can say, as is common in English, it became a Pandora's box. But this is really one of the challenges we're facing these days is integrating the value proposition into everyday life. What you see on the right-hand side of this screen is where the future likely will lie. This is where a contextual adaptation of systems that are constructed around models that are explaining real-life events will likely continue to be proposed in the future. It means that when technologies are integrating not only a software but equally a hardware, and so that's what I meant before, the future of AI is fundamentally robotics AI. We might see more contextual use of technologies that are going to enhance even more the understanding of what humans do naturally. If I tell you, what do you see in that picture? Everybody will see a baby with a toothbrush. The machine will require significant amount of data for that conversation to become clear. And so while machine vision is very, very advanced these days, it continues to be imperfect in understanding context. This is one of the things humans do very well. It's almost actually innate. We are easily able to understand context. Machine requires enormous amount of resources to be trained to understand context. Very likely we'll get there. But also bear in mind, ladies and gentlemen, a conversation we hardly have unless we are reforming our energy portfolio. These technologies are very energy intensive. And you know this already in the last few years, unfortunately, with the crisis in Russia and Ukraine, the rise, the rise of electricity prices has been impactful for regions like Europe. You started to realize that the oil and gas companies are starting to diversify their options. Sometimes they electrify, sometimes they integrate solar, and sometimes they remove themselves from the petrodollar. So reshaping the macroeconomics of the regions that are producing oil and gas. All of this to give you a sense that the energy market is still very volatile. Unless we are able to largely conform, we'll see that the rise of technological opportunities might be hindered by the access to power, which is unfortunately not evenly distributed these days in the world. Now, if this is the past, the present, the future of AI from a perspective of what the American compound DARPA looks at in classifying the evolution of artificial intelligence, this is also where we might want to see what is going to be ahead of us. Now, right now we are into the most proficient narrow AI era. This is where we have used artificial intelligence across multiple applications. And we do see this now integrated across. It might be in transportation. It might be in logistics. Most of our consumer behavior is now directly traced by data that becomes consumer insights. All of marketing today, it is digital. It would be unthinkable to have a campaign on a poster just in the middle of the street. Most of the technology solutions that are now facilitating collaboration like the one we're having right now, where you are in Poland, I am in Taiwan, and we are talking to each other, the entire ability of collaboration coming from the online tools is clearly where we have seen an incredibly extensive use of artificial intelligence. What you see in the middle is the promise of artificial general intelligence. This is where we think in the next few years we'll probably be seeing the application more and more widely distributed. But there's always sort of like a catch-22. Sometimes when I talk about technology like AI, I happen to do it always in places where people have access to infrastructure, data, technology. And so, yes, it's a matter of time. But there's equally almost half of the world in which we don't have the right infrastructure. And so it's still unthinkable that we'll see an integration of technology like artificial general intelligence that requires scaling for us to see this largely apply at the price that we'll be able to afford. But more importantly, where we're going to have enough use cases where we are going to be able to fundamentally justify that we have moved into what we call an AI-native society, a society in which AI will become the norm. It won't be just an exception anymore. We see also on the top right corner is this entire speculation that comes from movies where superior AI singularity. Look, we don't have to wait that much longer for the entire computational power of machines to be superior to the one of humans. If we're measuring transactional thinking, if we're measuring analytical thinking. So we don't necessarily need to wait for this to happen. It already is happening. The point is this: defining life and intelligence only in analytical ways It's like if I tell you that your IQ score defines the value of your life and you would know that I would be wrong, right? There are many more sides to life that would define the experience. Now, when you're comparing humans to machines, often what we're doing is nothing more than an equalization between the analytical side of humans and of course, the computational power of machines. So it's a very, very skewed, narrow, to some degree speculative comparison that we shouldn't be doing. Now, where do I want to go with our conversation today? Well, I'd like to tell you what are the waves of AI that we have experienced so far. We have 4 that have preceded the current one. The internet AI that started at the end of the '90s with the internet. We started to have the business AI where we seen especially driving Driving models that were coming from China have largely defined the field. We started to see that the perception AI, which was an AI that was no longer just about computation, but it was, for example, camera, motion sensors, fact that we can now define energy flows or water flows in a house by having smart sensors. The idea of Internet of Things, that was integrated different kind of sensor to the physical world is a wave that we've seen in 2011 as we initiated what we call the digital transformation era. It was originally a term introduced in 2010. And then we have seen in 2015 the fourth wave, which is the one driven fundamentally by autonomous AI. This is where we were delegating large part of decision-making processes that used to be made by humans to technology. The next wave likely will be, as I mentioned before, what we call the phygital, the integration of physical and digital, where technology will become smart also in the hardware. So expecting, for example, more smart robots integrated into our value chains, expecting more security systems that are gonna be integrated with smart robotics. We see some inklings of this already, for example, in airports where some of the detectors that we are going through right now do not require us to remove computers or liquids anymore. And that's simply because they're using smart processing to identify the content without us removing them from what we used to do back in the days, right? So there's a lot of different application that we'll see running. But the fifth wave will be the integration of AI into the physical world. Now, what has really also been an incredible lesson for all of us was that in no time, one subset of machine learning, which was large language models, has become an incredible success story. In fact, if you want to consider what the generative AI revolution is all about, it's very comparable to the dot-com. The dot-com, if you remember, was a period of our history where suddenly going online became the only way to do business. Today, we notice, for example, that venture capital fundings, they are comparable to the one of the dot-com in terms of the level of exuberance that we see in the market. We see that if the '90s, the conversation was about the web, in the year 2024, the conversation is about data. And so if you think about it with the games or words, we are dealing with data estate. So basically the data become in some form of assets. This is why generative AI, more than anything else, is the beginning of the data economy, an economy where data will have a price. That might be equally separated or independent from a transaction of supply and demand from pure economic terms. And today, what has really created a lot of opportunity around GenAI was the transformation at the industry level. We have noticed more operations becoming autonomous. We started to see that healthcare through the COVID crisis was largely automated to a number of different processes. That had to rely on smart technologies. We started to see hybrid work, hybrid conferences, hybrid engagement integrating itself into a normal way of life. Our logistics today are about traceability, blockchain, the ability for us to know where things really are and where they come from. So all of this is to give you a sense that we saw already in 2022 when we could aggregate the data that the market size of the innovation investment around GenAI, we're now totaling roughly $1.6 trillion. This number is going up right now. You have to consider that total global economy GDP is about $100 trillion. So we are still early, but we have a significant rise of the events in the first place. Now, when we're looking at what has really been happening, The Gartner cycle tends to be always the easiest way to exemplify. If you're looking at the curve, different kind of technology were piling up in the last few years. We were defining this technology most likely B2B or technology to technology. They weren't necessarily commercialized. We were integrating already some of these technologies into some pilots. Take, for example, Alexa or Amazon Go or Siri. They were examples where we started to see AI in everyday life that were collecting data. We started to see the need for us to rethink the data collection with either the GDPR compliance or the lack of GDPR compliance. But fundamentally, the conversation on data was already happening. What simply occurred is like in any accumulation of energy, Eventually, this concentration has tipped into the generative AI, fundamentally empowered by what OpenAI as an organization has been able to do, largely endowed by a large technology company like Microsoft. And at the right time, at the right place, with people willing to try what still to this day is a bit of a mystery. It's a bit of a black box because the the utilization of generative AI is still very amateur. We are at the early stage. We don't have yet either literacy or fluency in this. And it's a question that we'll have to have towards the end of my time with you. So from here, I think it's really about where are we going from here? Well, I told you this slide is intentionally very busy. AI is right now everywhere. So we see this especially in the application at the industry level. So clearly, if that is the case, will generative AI continue to revolutionize the business world? I know many of you are equally coming from a business background and you might be interested into that. Here is where I think the conversation will keep on being interesting. What we see in this slide is that GenAI has already been impacting a number of different core functions in industries. Most corporate operations today, they tend to be determined by some form of autonomous system that can be largely labeled on large language models. Marketing, as I mentioned before, is fundamentally a digital function. Coding for the development of programs, even simple website, even reservation system management. Tends to be today already largely defined by automation of tasks, which implies, of course, GenAI. Virtual assistants from chatbots to you calling a phone number of, for example, an airline or a credit card, and for a good part of the conversation, you are guided through automated voice. This is already an application of GenAI in which we are using the information coming from the customers for a constant training of the algorithm in terms of the efficiency to recognize the nature of the query. But you can also see in terms of the speech, today we use a lot of GenAI in what used to be translation services. And engineering and design has clearly become a digital design where we have not only the ability to define content ourselves, but anybody today with some simple prompting can generate PowerPoints, can generate reports, can generate visuals. So the coding of design itself has largely changed. Now, to go a bit deeper into this, what I find fascinating about GenAI is what we consider to be the ambidexterity that is allowing us to use this technology not only in what historically used to be the left brain, where we used to think about delegation of analytical tasks to technology, but equally the right brain, which was associated with creativity and strategy. Today, what we see more and more is this fusion between both the analytical side, because nonetheless you're dealing with the statistics and the non-analytical side, the content creation. Which is clearly something that was not belonging to technologies before. So one of the things that we have to recognize about the GenAI revolution is the ability to have largely revolutionized the application also for those kind of tasks that used to be delegated exclusively to humans. So this is quite normal that we're dealing with a degree of confusion. Why is it so difficult to know exactly the right and the wrong in the use of these technologies? It's 'cause we see the permeability of this technology across so many different sides of our life. And so it's clear that if you are an artist and you can now build the campaign for your next gallery by using a GenAI tool that creates a beautiful design and you print it out and you put it outside, you kind of are like meddling a little bit with what used to be historically a piece of art. Today we can see about symphonies, musical symphonies that can be coded through AI and become basically wonderful musical harmonies that were generated not by humans but by machines. So we're clearly in a space where I would say in between, you know, the idea of a limbo, right? We call it the liminal phase, a phase in between. It's like between day and night, right? There's a moment during the day where the day is coming to an end, the night is coming, but it's not night yet. We are in a liminal phase right now in the conversation. But I equally find it to be still an exciting time for us to explore. In fact, this is one of the major challenges we'll have to have when we are dealing with an increasing level of ambidexterity. How will we eventually have conversation about how do we explain this technology? What is the governance that we're going to apply to this? And where's the provenance? So where are— where's the data coming from? So the sourcing of the data from a perspective understanding how the data was collected, redacted, aggregated, and eventually structured will become a fundamental movement towards what I call a forensic approach to data like we do for taxation, where we have an auditing and forensic capability to make sure that people pay taxes. In the same way, we'll start developing forensic capability to determine the quality of the data inputs. Because as you rightly know, whatever we factor in has a large impact on whatever we factor out unless we filter it, guardrail it, vet it, probe it, test, stress test, testing for viability, testing for reliability. There's a whole language that is now coming live in how do we use the right level of governance for the kind of challenge we're going to have. Now, as I try to move you towards the next few slides, clearly GenAI will continue to transform business, but we equally will start seeing the rise of incredibly numbered dilemmas. And so we have to start considering the very likely in the next few years, we'll prototype, we'll sandbox, we'll have opportunity for us to start testing hypotheses and understanding, for example, where a technology application is good and we see the benefit distributed across where a technology application is less good because it might make economic sense, but the societal implications are important to be considered as well. So I'll try to guide you a bit in the thinking that is required. One of the things I'd like you to start taking from my conversation is that we are absolutely experiencing an incredible time in terms of the evolution of the technology. But the beauty of that is not the technology itself, is what it will allow us to become. It's much more about us rather than the technology itself. So the key questions about how do we use this transformational power of technology to redefine societal constructs, to redefining what is the right use of technology to address the many socioeconomic, environmental challenges we're facing. Identifying where, for example, technology can surrogate where we tend to be imperfect and unable to really compensate. Understanding how do we redefine any risk in times that we were living into mega threats, threats coming from many different parts of the world. So in this redefinition of our life in the 21st century, if the question is oriented towards how do we use technology to improve, our life, then fundamentally we have not removed humans from the equation. We simply have integrated them at the center of this. What we should be very careful about is that by simplifying humans to an economic function, we're going to create an equality between machines that would be absolutely immoral and unfair to do in the first place. Where I want to take you, ladies and gentlemen, as I try to wrap it to a place where we can ask some questions as well. Is really, first of all, likely what we'll see in the next few weeks, months. A lot of different kinds of text, images, speech, video, music, flowcharts, they're going to likely be impacted by the rise of generative AI. This is another very busy slide, but it shows you that across multiple functions in the firms, we'll see more and more applications of GenAI because it's simply becoming easier for us to test, to try to run assumptions, to try to run experiments. And so this is just the very beginning. What you see into ChatGPT is the very beginning of, if you remember, those of you that are old enough, when we started to have the first websites and they were taking forever to upload. When we started to have our first email, and it was kind of exciting to try to reach to friends to the point that you would send an email and then call them to say, did you receive it? It was magical, isn't it? Similar, ChatGPT is the very beginning of the text generation. But the moment that we start having content that is going to become directly from labeling data with millions of users, the possibility ahead of us are just at the very beginning. And that's an important part for you to recognize that You cannot use the mindset of today to understand likely what will happen in the future. It's really about defining the next step to make sure that we're not distributing the cost to the weakest member of our society, which would be number one, but equally that we're not removing us from the question. It's the contrary. How do we empower ourselves with conversation where humans are in the center of the conversation rather than removing them. I'll try to bring you to a point where you can see this to some degree visualized. This is where I think an example where likely I'd like to take you towards the end. You know, as a professor, I work in the, in the academic world. And so education is one of the examples that I personally find very, I'm very close to. There are already quite good examples where education likely will see incredible benefits. And this is where in a report from the World Economic Forum, we started to see and you see the language, right? How do we fundamentally start looking at technology empowering children at school, teachers, institutions? So it's only an example I wanted to take forward, but there's one close to my heart because I work in the field. Now, this is again a question about apply this now. Not only to education but to any field, right? We are at the very beginning of an intersection where education and AI is meeting. But you can say logistics and AI, accounting and AI, banking and AI, any field is currently creating this level of intersection. So how will we fundamentally think about the level of transformation required for us to be able to navigate the challenges in the educational context? There are clearly few learning experiences that are going to be very critical. I'm going to show you some so that you can see sort of where this likely is going to go. Remember, you can apply, as always, this to your own industries. And so we will see, for example, more skills on global citizenship and what it means, for example, to be part of a much wider community. A lot more skills on innovation and creativity using technology. Imagine this applied to any department. How will we start utilizing technology to enhance creativity rather than directly delegating 100%? There will be an entire number of digital skills gaps that will have to be taken care of. How will we eventually do that? And how do we start building interpersonal skills in a world in which technology by design tends to disintermediate the transaction between humans. So this is a clear space for us to be intentional and mindful about it rather than the opposite. And then you see there will be, for example, self-paced learning. Apply this to any industry, there will be a degree of learning coming from the access to technologies. There will be the ability for us to constantly train ourself. And so the idea that we're using training as a way of integrating our skills to where the market is currently going, making sure we're not falling into pretty big gaps that are becoming over time insurmountable. And so we will never be able to fundamentally close those gaps. We'll see more problem-based and collaborative learning. It will happen in every industry in which we'll be able to see the value of collective action, because now, rather than information asymmetry and silos, we can integrate and mainstream information. These are some examples where in education we see already the change. But as I mentioned before, it's just an example, an exemplative model on how this can work also for any industry. Where I find this to be, of course, challenging is that if, for example, education faces an incredibly challenging moment, the same can happen around any industry. So we will see more and more of gaps that will need to eventually be filled. You see here one of the most interesting parts of this conversation. You see the third point, digital proficiency deficit. As we are dedicating ourselves more to GenAI, we'll need to make sure we're not leaving people behind, but we're pulling them into the conversation, making sure that the gap is mainly a gap of evolution from awareness First steps, some degree of literacy, numeracy, eventually expertise, fluency. And so the idea would be how do we start building this so that we are not using technology to leave people behind, but we're using technology to include, to integrate. This is why the title of my presentation is about putting humans in the loop, right? Is a human-centric society. One of the things that likely will happen if we are using AI in education, but the same will be in any industry, that we are augmenting and automating roles. And this is where I really like you to start thinking as, so like I'm wrapping up, open for the question to come. I think there's a slide for this that I find to be better. When we are looking at large language models, take now any industry, there will be fundamentally 3 things we can do. And this is really where we decide. It's not decided for us by the algorithm. The first one will be, what do we automate? Because efficiency is important. We have entire institutional frameworks that are largely defined by the ability to become better, optimized, more efficient. We won't go away from this. So if we have a task that is repetitive, that is incurring into unnecessary cost, it's not very dignifying as well, automate it. Step number 2, what can we augment? How do we make people bigger? How do I expand their roles and tasks? How do I now let them use technology so that they can make a better allocation of their time, their resources, fundamentally their talent and their life. So the second question is recognizing where in the job that we currently have, we can use GenAI to augment tasks. And the two are going hand in hand. Automatable, so you're automating tasks that are obsolete, large level repetition, likely to be disappearing no matter what. Augmentable, the task where you're using technology to augment humans. And the last one is the one you're not necessarily leaving to machines, but you preserve as unique human agency. The ability for us to be humans when we need to, having a conversation where it is necessary, getting together and using the power of discourse and critical thinking to engage in critical challenges. But at the same time, recognizing that we will never live in a society that is going to be decoupled from technology. Technology is us. We have created a society in which the technology we have today is an extension of our brain. And so as we think about the fact that technology is nothing more than extension of our brain, it's not neutral, it is what we make of it, then in this conscious awareness that is nothing more than fundamentally the extension of our own thinking, that choosing what can never be replaced because we want to. Take, for example, a conversation with a child that happens in school with the teachers. You wouldn't like that to be with a chatbot. But there are circumstances in which, for example, a claim in a police station where you don't want necessarily to be exposed to prejudice and biases might be made by a smart robot. Now, it might look like it's very similar, but maybe in one situation, we decide as a society that we want to automate. In another situation, we decide to augment. I have seen an example, for example, of a chatbot that's doing some medical triage so that when the doctor sees you, there's already a lot of information that the doctor can use to give the right prescription. What we had done, we have automated what might be repetitive but we have augmented the doctor to do the right choice with the time that the doctor has. At the same time, we might decide that things that are important about therapy, about the political discourse, political debates, running important decisions, do not have to be mediated by technology. They have to be human functions. The choice is entirely up to us. I think it would be a mistake if we think that everything is replaceable. I think it will be a mistake that we can actually fight innovation and keep things as they used to be. So the balance across automatable, augmentable, and basically what I call the no-fly zone, we keep it for humans, is really the balance of the conversation I would like you to have as you move forward in the conversation ahead of you. So look, ladies and gentlemen, I think the presentation I have continues the conversation about an example of the Grok Academy. I invite you to look what they have done in Australia to integrate AI into education. They have fundamentally rethought the entire curriculum. I don't want to spend too much time on this because I'm more interested in maybe questions coming from the audience. And then what I think is my final takeaways, and that's where we're going to close it for questions, is that when we're looking at technology in education, but in general in any industry, The promise that we have is huge because we are going to use it to basically prepare people in the future workforce. The same can be applied to any industry. We will continue to have challenges about privacy, data security, job displacement, not only for educators but for any job. So the question will be what will be the governance systems and the safety measures that we need to have to guardrail? When we're thinking about the equitable access, Unless we're making this a distributed technology, we will never really create the level of literacy that we need to have. You don't want to have a system based on pockets of knowledge depending on where you are or depending on what country you were born. And fundamentally, rethinking the relationship from a competition between machines and humans to collaboration is really where we want to elevate humans to be the focus and the scope and the purpose. Rather than necessarily just the operators of technologies. This is where I want our conversation to finish. Thank you very much.