The COVID-19 pandemic has accelerated artificial intelligence adoption across industries, demonstrating there is clear confidence that AI has the potential to solve critical business challenges. However, as with any emerging technology, there is a lot of hype surrounding AI. As artificial intelligence barrels toward ubiquity, it is increasingly difficult to separate puffery from reality.
Our webinar, AI Trends that Matter in 2022 (and Beyond), offers perspectives from super.AI CEO, Brad Cordova, and Intelligent Automation author, Pascal Bornet, on the artificial intelligence trends actually worth paying attention to.
Pascal Bornet [00:00:00] Hello. Welcome to this discussion on the AI predictions for 2022. Today, I'm joined by Brad Cordova. Brad Cordova is the super.AI founder and CEO. Myself, Pascal Bornet, I'm the Chief Data Officer for Aera Technology and the author of the intelligent automation book. Brad, very happy to have you with us today. How are you?
Brad Cordova [00:00:29] Same, same here. I'm always excited to talk about A.I.. So excited for our conversation.
Pascal Bornet [00:00:36] Excellent. So let's deep dive into it. Brad, thanks for that you sent me a few days ago a few predictions for 2022 in AI. A few predictions, not a few. A long list of predictions, but you can read of them. And of course, we won't have the time to go through each of each and every one of those ones. So what I've done over the last few days is extract from those predictions, the key themes of 2022. And I would say the overall overarching idea for 2022 is AI focusing more on business, on the use case, on the value that we create on building applications and not models. I think we will come back to that soon about in one word, maybe about scaling, basically enabling scaling in companies. So the key points of the discussion that I would like to have with you today. Let me know if you're happy with that. The first one is starting with new platform capabilities that help to bring a higher business value. The second is combining technologies to create more value and especially involving unstructured data. The third one could be the importance of humans in the loop so, so a centered, human centered transformation. And finally, we can talk about the changes in people roles that are implied by business technology trends. If we have the time, we could maybe talk about the impact of automation and AI on jobs. What do you think about that?
Brad Cordova [00:02:25] I think I think that's a really good summary, and I'll be happy if we get to all those points.
Pascal Bornet [00:02:32] So let's try and deep dive into it right now. So some of your predictions, how about new platform capabilities that helps to build higher business value that are more focused on use cases on on the business values? Can you tell us a bit more about that?
Brad Cordova [00:02:55] Yeah. So you see this a lot with any technology that comes out. Typically, new technologies start as an interesting R&D science experiment, and the ones who survive that phase eventually get used to drive economic value. And that's where we're at today with AI. So how I made the prediction is we're going to see the commoditization of what we call A.I. models and the rise of AI applications and and this follows in that same trend. And so what I see is instead of an isolated R&D group sitting in the back of the office, the AI projects are connected throughout the entire organization. And instead of just being something flashy or the deliverables, a great demo, it's really about the the ROI in the system. And and this is really important. I think this is this is the core of that and I think underlies a lot of the things you mentioned. So I think that's a big one. Yes. And I don't know your thoughts are on that.
Pascal Bornet [00:04:08] No. And I completely agree with you, especially on the collaboration part of its where as you as you said, the AI Centers of Excellence or R&D teams shouldn't be isolated from the business so that they can really focus on bringing the highest value, starting by identifying the right use cases together with the business and then build those AI applications with the business, hand in hands programing and then testing with the business to make sure that this is aligned with the needs. And and I feel that today we are in lack of such platforms on the market and and I really appreciate the points that you that you raised here.
Brad Cordova [00:04:56] Yeah. And I mean, I think in terms of infrastructure and platform. I mean, we'll get into hyper automation, and I mean, you wrote a book on that, literally, but I think before that quite a technical point that's really specific to machine learning. I mean, is this point of of quote unquote AI safety. So the as we know, the power of of machine learning is that it learns from data the algorithms implied through some kind of gradient descent. And so if you train it on some data distribution, it's going to learn the patterns of that data distribution. And if you run it on your test set, it's going to say, let's say, 95 percent accuracy. But then if you change the distribution, what what surprises people is, it's no longer 95 percent accurate because this problem doesn't plague software. And so stemming from that, you have all these problems with machine learning. It's very powerful, but its greatest strength is also its greatest weakness. And I think another and this is more of a technical point, but another trend of twenty twenty two is is air safety. And so this is, can you build a AI with guarantees? Can you handle covariate shift? Can you handle governance explainability, A.I. uncertainty estimation? And I think I think this is this is really important and it's still in its research phase. But this is something we focus on a lot at super.AI because we're making real applications and you need real guarantees, and we figured out a great way of paying off Henry Ford of decomposing complex problems into simple problems. And this old idea, this 100 year plus idea turns out to take us a big step along that direction.
Pascal Bornet [00:06:50] Excellent. I see it, I mean, listening to you, I see it like a like a platform that would, first of all, focus on, I mean, clarify the the needs and the objectives of building a certain application and monitoring these goals throughout the building process of these applications. Is it in order to, as you said, build safely these applications, meaning making sure that we that we reach those goals at the end. And that you I
Brad Cordova [00:07:24] I agree, and I think that's important in every application, no matter what you're building. But it's particularly important when you're building AI machine learning based applications because when you when you write software, you can write integration tests, unit tests because it's deterministic. The the algorithm comes from the programmers head and you write if and if statements in for loops and it's deterministic logic for the most part, where as machine learning is probabilistic and statistical. And so if your unit tests fails, well, is it because something's wrong or is it just if your algorithms 95 percent accurate, what you would expect it to fail five percent of the time? And does it happen to be in that five percent? So there's this whole new modality around quality that is very unintuitive for people, and machine learning models can silently fail. And because it's probabilistic, it makes this whole monitoring thing take on a totally different importance. And so I agree 100 percent,
Pascal Bornet [00:08:31] yes. And so hearing also from you here, I can understand that this platform would also efficiently monitor all those processes throughout the building of those applications. And it makes me think of of a trend that I really like to talk about, which I really believe in. Basically, it's a. And it comes from a fact, basically that is very ironic today, most of the artificial intelligence transformations are human intensive, they are manual. Basically, what we do currently when we build any application is it's done manually, programing manually. Finding the use cases manually interacting with the business users manually. But what if technology could help us do this easier and faster? So typical process of building an application, I start with identifying use cases, then designing those use cases into into an AI model and then and then designing test and going into production. And and so some some applications, such as process discovery that helps to identify more in the process automation space without with other use cases that can be automated. Or data discovery that can help to identify the relationship between data so that you can build models we've looked evaluates that that helps to to improve the accuracy of of the predictions. Those type of of of applications, you can think also of auto-ML auto machine learning that helps to to automate some of the time consuming tasks that data scientists are doing, such as testing, testing different models on this on the same data. Is it is it also something that that for you is part of this AI safety application platform? I would say
Brad Cordova [00:10:44] Yes. So I have a very particular prediction about this. So when I look at the history of intelligent applications, let's say going back even to the 1950s and or even in the 1980s when deep blue beat Garry Kasparov or when we kind of beat Lisa Dole in Go. What I see machine learning AI excel at is what I call closed systems. So whereas Go and Chess has a common rhetorically, expansive and almost unsolvable amount of moves. It's still a closed system. And in these closed systems, I think if if there is closed systems that exist, AI will quickly become better than humans at that. I think the last thing AI will conquer is open systems. These are these are really difficult and I think a lot of these process discoveries. I think the question of having an infinite number of things to do and narrowing that down to a finite number. I think that's going to be a very long time before AI can do that, I think AI is going to solve most closed systems in the next 10 years. But I think humans role will be elevated to these these open systems, and I think a lot of these problems you mentioned, not all of them are open systems, and I don't think AI will solve those. There's you mentioned auto-ML, auto-ML is solving only really simple tasks right now, and it's going to get more advanced, but I think to solve open systems and a lot of my research I was doing during my Ph.D. at MIT it was just it was it became even more clear to me that that that's the case.
Pascal Bornet [00:12:39] Yes, I agree with you that we are not yet. They are fully, you know, more technical.
Brad Cordova [00:12:47] Oh yeah. One thing you did mention, though, is that in a lot of these AI systems, it's still very human, dominant, even in a fully automated syste it's humans labeling data essentially programing the machine learning model, I mean and supervised learning, and most data on the internet is generated by humans anyways. So even if you're doing self-supervised learning, the models are still essentially being programed by humans. And I think a lot of people try to reject this as it's somehow a bug, not a feature, especially if we look at a lot of my friends who are at Google or Microsoft, they're just trying to create purely automated systems. And our thesis is rather than treat that as a bug, we treat that as a feature. And the observation come came from from chess. I love playing chess and the best chess players aren't human chess players. The best chess players aren't machines. Actually, the best chess players are human plus machines together. And when you look at a lot of case studies in the real world, for example, PayPal, they were struggling to be profitable because there was so much fraud. Any they tried the most advanced machine learning models, they tried was humans, and they couldn't get profitable. But when they merged human and A.I. together, working together, because they're very orthogonal, and when you have two orthogonal systems you can have, when you combine them, you get maximal power between them. And so we took that idea and over the last three years made it mathematically rigorous and built our system around this. We have this router that can route between humans, AI, and software and then combine them in a smart way. And then we have one hundred and fifty quality control over one hundred and fifty plus quality control mechanisms. And then and so I think instead of trying to deny this fact or try to hide this fact, I think it's really important to see this as the reality and see this actually as a as a benefit. And I think when you do that, it becomes really powerful.
Pascal Bornet [00:15:05] Yes. Yeah. So thanks. I think it was it was a it was a great discussion around the first theme, the new platform capabilities to help us build more business value with the more business focus. And a second theme that I wanted to to to discuss with you. That is extremely hype for the moment and and and that I'm also passionate about. It's about combining technologies to create more value. So basically to be able to address more complex use cases and this involces as well unstructured data. I remember I mean, I read in your email the capacity to have different models interoperate between themselves, connect between themselves. The capacity also to leverage unstructured data to to get to get more value out of those those those applications. Can you can you tell me more about that?
Brad Cordova [00:16:11] Yeah, and and I I definitely also looking forward to hearing your opinion on this, but I'll kick it off. Yeah. So if if we look at the world today, 90% of the world's data was created in the last two years. And if you double click on that, 80% of that data is unstructured data, as opposed to structured data. So as you know, but if someone's listening. Structured data is data that can fit into your spreadsheet, has rows and columns, and the computer can understand this because it's structured, unstructured data data such as images, audio, video, text data that doesn't fit organically in a spreadsheet. So 80 percent of this data is unstructured and unstructured data. Sometimes people call it hidden data. You can't directly use it with your computer without adding structure to it. Our human brains can can process that because we take the unstructured data converted into electrical signals, which are then structured into neural activities. But our computers can't do that yet. And so one of one of the most important things you want to do is since most, or 80% of the data in the world, that means 80% of the enterprise data is unstructured. If you want to use all of the things you invested millions and millions of dollars in like RPA, your software infrastructure, your machine learning infrastructure, you need to get into a structured format. And that's what we focus on at super.AI. We do unstructured data processing, but but this this turns out to be really important, and it is a key if you want to unlock the collaboration between these technologies between RPA, between the Amazon's A.I. models, or Google's AI models, or open-source A.I. models, and and between humans. You have to you have to transfer the unstructured data in a format which all these people can understand. And I think that's really the starting point between this collaboration you mentioned. I have a lot more to say, but let me pause and not monologue here, and I'd love to hear some of your thoughts as well.
Pascal Bornet [00:18:28] Yes. Yes, it's so, so so yeah, really critical to to to be able to tackle this, these unstructured data. As you said, it's it's 80% of the of the of the company's data. For the time being, we've only worked on 1% of them. That's what I read. Do you have an idea of the potential value that we could get from using this unstructured data in the world, in companies?
Brad Cordova [00:18:55] Yeah. So you can you can estimate this a number of different ways, obviously top down bottom up. But the the latest estimate I've seen is right now it's about a $15 billion market. So what we're seeing is three multi-billion dollar companies in the RPA space. And what RPA is doing they're using software instead of A.I. to handle structured data instead of unstructured data. And in the unstructured data market, not only is there more data, but there's more problems that need to be handled. And so we're seeing a lot of growth. Some of the fastest growing companies in the world on a really tiny amount of data. And so imagine bigger problems and bigger data. It's a really big market.
Pascal Bornet [00:19:43] Amazing, and it's across I mean, the types of unstructured data are obviously anything that is written like emails like conversations, but also videos, is it videos, what else...
Brad Cordova [00:19:59] Even satellite data? Yes. Yeah, there's there's a lot of different types of of data out there. And I think I think there's even a bigger point here. And I don't want to go too far out in the future. But if you look in the history of the human race where you have, for example, the printing press that allowed essentially created a protocol for more people to communicate, and if you look at the GDP per capita when that came out, there was a massive spike. If you look at the the industrial revolution, another masssive spike, if you look at when the internet came out, another massive spike in GDP. And so what happens during these points where you develop some technology that can connect people instead of us being individuals, we become in a way this collective consciousness where now we can almost access all of the collective knowledge of humanity on the internet. That hasn't yet happened with machine learning, machine learning models don't talk to each other the way humans talk to each other on the internet, and so there's this really latent intelligence that I think is waiting to be unlocked, but it needs a protocol. This something else we're working on at super.AI. We developed the protocol just like there's the TCP/IP protocol, where computers can talk to each other. There need to be a protocol that makes it easy for machine learning models to talk to each other. But I think once you do that, we're again going to see a huge spike in the intelligence of of AI. And then if we can connect that to RPA and connect that to humans, then it's it's going to be another huge spike. And so I think that's kind of another long-term vision of hyperautomation. But at the same time, it's very practical because there's a lot of things that humans can do right now that AI can't do. There's things that software is much better at that you would never want to use AI for in RPA. And there's just things that are much better for for AI that software could never do.
Pascal Bornet [00:22:13] Yes, and we've heard a lot about GPT-3 over the last 18 months. GPT-3, the the the largest and and most sophisticated language programing application launched by Open AI. I think it was in 2020 where do you position super.AI compared to GPT-3.
Brad Cordova [00:22:44] So another one of our theses is along the lines of the commoditization of models, so our goal is not to develop, let's say, the best machine learning model because let's say even if we do, there's maybe going to be a 100 trillion parameter GPT model going to come out the next day, or there's always going to be a better model. What we decided to focus on is the actual protocol. So we collaborate with Amazon, Google, a lot of open-source things as well. And so what we do is we actually use the GPT three model and GPT Neo and there's there's a few other ones that have come out and you can with the click of a button integrate it into the super.AI system, and the router will treat that as another worker to possibly route to. And maybe it'll route to GPT-3 and GPT Neo, maybe GPT-2, and then the combiner will combine those into a single output along with a human and maybe a software bot. So we're actually not in competition with any of these companies who build machine learning models, but instead we use them, and we try to squeeze the juice out of them as much as possible. We give them guarantees. We combine them with human and AI. And this turns out to be very practical, because typically before someone meets super.AI, they want to solve some automation problem. And they're trying to make this choice of do I use AI? Do I use RPA? Do I use BPOs and just humans? We feel it's a false choice because they all have pros and cons. And so we turn that "or" into an "and." Why do you have to choose? Why not use them all? And you've probably made a certain investment in each of these, so why not plug them into a single platform, have kind of mathematical guarantees and use them all?
Pascal Bornet [00:25:00] Okay. Exciting. Yes, yes. Don't reinvent the wheel, but build on this to to to to create more value for for for businesses, for companies. One of them. So. So I think it was very important to talk about unstructured data as a starting point to talking about combining different technologies to build a higher value. Indeed, most as we just discussed most of the data in companies is unstructured. If you can grab those data, you can benefit from them and you can create more value from all the for all the the applications that we built on top. Talking about applications and models, I remember reading as well a very interesting part in your in your predictions that is about connecting and having models that operate together.
Brad Cordova [00:25:58] This is this is what I was talking about about this protocol, and there needs to be what I would call a general AI marketplace. So people still today when they release models mostly still release them to GitHub, and GitHub was designed for software, right? So there's a lot of great tools that support your software development, and a lot of social tools so developers can collaborate. And this was really game changing. This kind of launched the world of open source to where we see it today. And this started a bunch of other things. The thing is, a lot of machine learning is built with software and so to a certain degree, it's appropriate to use GitHub, but with the rise of A.I. comes two additional things on top of software. And that's the data. So you need to track so many different things about the data because the algorithm is implied from the data. As we talked about, if you if you change the data, then it changes the algorithm. So you need to really make sure you know what data you train this on. And then the second one is is the parameters of the model. So we could have software which designs the model. But if you train it with this data set or not, dataset, it's or if you change it with different hyper parameters, it's going to create different parameters of the model. So you need to track these as well. And there's a there's a third thing, which super.AI adds, is we have this schema. So every machine learning model may have as input an image, and an output as classification. Or maybe video, and there may be a time series text output. And so there's things like this that GitHub wasn't designed for. And I believe for this interoperability to be possible, we need another kind of quote unquote GitHub for AI. And there's a few projects that are making some good leaps in that direction, but they're still missing some key features, so I'm hoping someone will make even more progress on that.
Pascal Bornet [00:28:14] Yes. Yes. Thanks. Thanks for that. And I'm very passionate about the capacity of hyperautomation, that we call also intelligent automation, to bring more benefits to companies. So maybe just for those of you who are not aware of what is intelligent automation, or hyperautomation, it's the capacity to automate end-to-end business processes by combining different technologies such as machine learning, computer vision, natural language processing, robotic process automation. And, for example, let's take the example of a pool cue to pay end-to-end process. Every company in the world is using these process, which is about identifying suppliers. Select those suppliers and you can use machine learning based on ratings and history of this, of those vendors to select the best ones. And then it's about sending orders to those to those vendors, leveraging workflow platform, for example. And after receiving the goods or services from the suppliers, it's about receiving their invoices and processing those invoices. Invoices are unstructured data. So you need natural language processing to automatically process those invoices. And finally, the payment of these vendors is made using RPA because it's a very transactional activity. So as you can see, combining those different technologies together helps to automate more than 80 to 90 percent of processes, delivering 20 to 30 to 60 percent productivity improvements to companies, according to the research we did in the book. So that's really a trend that is that is improving as well customer experience because you go faster, improve your experience because you will know the people in the company to focus on the most interesting and value added activities, freeing them up from those transaction and repetitive activities. And Brad, what is your your experience with hyperautomation and how do you use it at super.AI.
Brad Cordova [00:30:32] So the things you said, I agree with and hyperautomation, in the way you defined it, is really the core of the super.AI system. So first of all, we make it really easy to integrate whatever AI model you're using, machine learning model. Whether it's your home grown, your home built machine learning model, whether you're using some open-source model on GitHub, whether using SageMaker or Google AI models, we make it easy to integrate. We make it easy to integrate humans. They're using your own humans. We have a crowd around the world, or if you're using BPOs, and we make it easy to integrate software bots. So the first part is just it sounds basic, but it's really important if you can have the most amazing technology, but if it's too hard to integrate, no one's going to use it. So step one is making it very easy to integrate all these technologies, you said. And then step two is we spent a tremendous amount of time on a framework so that these can all work together and talk to each other again, in a very easy way. We're not talking about you have to be a genius and a Ph.D. in all this stuff to get it to work. Any business user can use it. We made it really easy and it actually works. That's the most important part. That's key because it sounds nice to talk about all this in theory. But at the end of the day, as we talked about, we're moving away from the isolated science projects into a world where this needs to deliver real value. And it's easy to get it running quickly. It's really hard to get this running at scale, at the enterprise grade level. And and so it's very, very key to everything we do. And I like the way you mentioned it. As you were saying that, it reminded me of Adam Smith actually in the specialization of labor. Every time we specialized as humans and then we collaborated, let's see. I was better at fishing and you were better at getting coconuts. And we we traded fish for coconuts. We're all better off. And then when we were able to tame animals, we extended the specialization of labor across species. And then we had oxes farming for us instead of us having to do it. And this is just a natural extension of that in my mind. We're entering, we're allowing other species to enter not just humans, not just animals, not just robots, but AI, but software. And I think kind of from a theoretical perspective, this is what makes it work. But practically this is just the right way to go. And what I've seen, and I've seen similar numbers to what you were saying. But there's another way I like to think about it. So let's say, let's say the difficulty of a task is how long it would take. If it's a very simple task like what's one plus one, you could say two very quickly. If I give you a very challenging, let's say, differential equation or integral to solve, it'll take you longer. So the time it takes you to solve something is correlated to the difficulty of it. And so what we saw is around 2019 automation was powerful enough to automate things that I would say took humans around 10 seconds to solve. And again, this is a rule of thumb. This is simple, but it's a good way to think about it. When we see the merging of humans, AI, and software in 2022, we're seeing it be able to automate things that take humans around five minutes to solve. And so this is tasseled, say, over 98 percent accuracy. And I think what's more impressive to me is the trend of this. And so, I can imagine in the future it's going to go to 10 minutes, to 20 minutes, and it's going to continue to automate more and more. And I also like what you said because we see this as well, like sometimes there's this fear that the AI and robots are going to take our jobs. And that is true, but I don't necessarily in all cases, I think there are some bad cases, but see that as a bad thing because what the AI is doing is solving these mundane tasks that we as humans don't want to do or shouldn't be doing. We humans love being creative, interacting with other humans, being empathetic. And so I think the almost paradoxical thing is AI makes us more human, and I think that's another interesting thing we're seeing.
Pascal Bornet [00:35:13] Completely aligned with you on that point. And that's a good transition to the human aspects of those technological transformations. And we talk a lot about human in the loop human centered transformations. We talk a lot about democratization of artificial intelligence by using technologies like low-code platforms that that require limited skills to build AI applications, that make AI accessible to most business users basically. Even those who don't know how to program or could. And so to make it very simple, those are programs that helps you to drag and drop parts of codes. For example, use it on user friendly screens using wizards. So basically, the benefits of using those type of technologies is, first of all, to accelerate the speed of the transformation because you have obviously more hands, more people in the company using those those programs. But from my experience, the most important is because everyone in the company is empowered to change their day to day work, this drives higher ownership and acceptance of AI and allows the shift of the company's culture to more digitization and automation and AI. What's your take around this and how do you work on this at super.AI?
Brad Cordova [00:36:49] When we first started, and because I'm technical, we built a more technical platform that was designed for developers. We actually pivoted pretty early on because we saw a much more valuable and better market, and we thought we could add more value by making this a low-code or no-code platform. And I think in general, the rise of these low-code, no-code, AutoML is being driven by this massive surge in adoption of AI. And AI is being adopted so fast that it's creating a really big talent gap. And what low-code, no-code, AutoML does is it solves this talent gap. And as you said, it allows you to harness and maintain your growing AI machine learning momentum without getting blocked by this absence of talent. And it really empowers people who aren't coders to get stuff done. And there's this old adage that getting things done is better than getting it perfect, and low-code and no-code and AutoML, it doesn't get it perfect. Developers are going to do it better, but it gets things done fast. And in my opinion, it's about not letting perfection be the enemy of progress. And so I think it's a really great thing, and I'm excited to see the explosion of these tools. And we saw this firsthand, and we still do offer a full code platform. We're actually seeing the adoption of our of our no-code and low-code platform really take off.
Pascal Bornet [00:38:25] Excellent, and I see yes, I see yeah, I see it as a key train for the future. Imposing a maximum of people in companies to participate in those transformations is vital to get to get this transformation to scale. Brad, thanks for that. Let's talk now about the impact of automation and AI on jobs. We've talked about how AI brings new capabilities, how we can combine technologies to bring more value to companies. Now what is the impact of those transformations on our society and on us? For example, I've read that 30 to 80% of all current jobs could be automated in the next 5 to 20 years. My take on this is there are different schools of thoughts on the topic, and one of them is coming from the economists, which think that in the past automation revolutions, while technology has changed a lot of things, new jobs have appeared and we don't know yet what will be those new jobs in the future. And I keep on presenting this example that 10 years ago, no one would have known that today our world would count more than three billion or three million Uber drivers. Another school of thought thinks that this change is going faster than before, that we won't have the time probably to train everyone, to those new technologies. Brad, what's your take on that and and how is super.AI playing a role in this?
Brad Cordova [00:40:12] So I think a bit of both of those have some truth to them, kind of like anything, it's never usually on the extremes. And so. What I like to do, and this is something Ray Dalio talks about a lot, is if you see something happening in your lifetime, people think it's the first time it's happening, it's the end of the world. But often if you look, this is probably happened multiple times, but it just was before your lifetime. And so if you look at similar scenarios, for example, in the 1800s, something like 95% of people were farmers and single digit people are farmers now. So really close to our lifetimes, there is kind of even something worse than than the 80% you said, it was 95%. And during the industrial revolution, there were the Luddites destroying thimble factories because the machines were taking their jobs. And so, as you said, people still have jobs. It's not that because we're not farmers anymore, or we're not building thimbles by hands that no one has a job because now there's Uber drivers and now there's software engineers. And so new jobs do come. And typically they're jobs that are higher on the intelligence range. But the problem is, as you said, I think there's two big problems. One comes from the concept of telescoping evolution, that each technological advancement happens faster and faster than the other. The agricultural revolution took over 10,000 years, the scientific revolution 400 years, the industrial revolution 150 years. And now we're going to have AI and technical revolutions sometimes happening every 10 to 20 years. And so previously, maybe you and your father and your grandfather and your grandmother and all these people did the same job and then maybe had a switch. Now these changes are happening multiple times in one's lifetime. And so I do think through kind of the second point you made that the speed, the fact that it's happening is nothing new and nothing that really worries me, but the speed at which it happens and looking at the state of our education system. I have faith in humanity that we'll figure this out because I know people personally who are trying to innovate education system and and if these can, if you can merge human and AI and improve the education system, then I think it'll work. But it's not there yet. And so it does make me nervous and I do think that's a problem. But I think the solution involves accepting, like we talked about, whether it's hyperautomation, unstructured data processing, is that humans and machines need to work together to solve this problem because they both have strengths and weaknesses. And it's the faster politicians and decision makers figured that out, I think the faster we'll be in a good place.
Pascal Bornet [00:43:25] Yes. So it's so if I hear you well, it's about probably reviewing the way we educate, or at least the subjects that that we that we teach to our kids, to make them to make sure that those topics are brings kids that are complimentary to to the to the capacities of technology so that we can work well together with technology as a [unintelligible].
Brad Cordova [00:43:49] And and I think even more importantly, rather than than teaching them the subjects, is teaching them how to learn subjects because everything, maybe even our kids, the next generation, probably everything they learn in school is not going to be useful. I think the only things that are going to be useful are like emotional intelligence, maybe how to use Google, and how to learn, how to reeducate yourself. It's going to be a stressful world, but I have faith.
Pascal Bornet [00:44:20] Exciting, exciting world, ever moving. And I agree with you learning how to learn is a key skill in the future. Thanks. Thanks, Brad. That was really great discussion. Very exciting points. Thanks for your time. Thanks for this. Any any small word for the end that you want to share with with our audience?
Brad Cordova [00:44:49] I enjoyed our conversation. I think as as people approach their AI journey, I see a lot of a lot of hype around AI. And I think what I like to say is: AI is just another technology, and if someone's promising that this can solve cancer and do everything, I would second guess that and I would stick to the basics. Does it offer an ROI? Can we implement this? And nothing changes. And I think that's really important because sometimes when it comes to AI, people think it's magic and they throw all their intuition out the window and it's just like any other technology. It's a very powerful technology. It's changing the world. And but I I think the sooner people realize it's just another tool in the toolbox, the more it'll get adopted and the more it will improve our lives.
Pascal Bornet [00:45:46] Thank you Brad. Very exciting and very inspiring. Thanks for your time. And I hope to talk to you soon. Thanks to the audience who listened to us and talk to you soon.
Brad Cordova [00:45:59] Thanks, guys, for listening. Take care. Bye.