 Let's now take a moment to talk about AI. Since the field got started in the 1950s, the journey for AI has been a long and winding one. It's been both promising and confusing, and at times, underhyped as well as overhyped. The grand challenge of AI has long been to create a system that can truly understand human language. AI may not mimic, but understand, reason and learn from human language. It is an arduous task, and one that still challenges us today. But the pace at which we have made progress has been vastly accelerated through new methods, such as deep learning. But what if we turn these powerful new AI tools to look at languages in a completely different way? You see, languages have been the cornerstone of human progress since the beginning. And although we're not the only species communicate through audible languages, we are, however, the only species that has taken the next step of abstracting that language into symbols. All around us are the symbols of language. From the typographic characters of our spoken language to the numeric symbols representing the properties of mathematics, to the diagrammatic maps displaying the chemistry and arrangement of molecules. Languages surround us. They help us think and reason, then allow us to share those thoughts with others. Now, we have seen the power of AI applied to human language. Its importance and pervasiveness is undeniable. But at IBM Research, we are also exploring how to teach AI two new languages, and their exciting implications, the language of code, and the language of chemistry. This is where AI's journey is about to accelerate. We are now going to talk to IBM Fellow Maya Vukovic. She and her team are spearheading great work. Let's go meet Maya. Hi, Maya. How are you? Hi, Daria. How are you? Maya, let's talk about what we are unveiling, and what do you see as the power of applying AI for code? So our AI for code technology is going to fundamentally change how we think about coding. Let me first give you an example of some of the work we have done, and then let's talk about how we did it. One of our clients came to us with a problem they couldn't crack. Imagine their mission-critical application has ballooned to over 1.5 million lines of code. Decades of adding, migrating, combining different systems. Moreover, this evolution of the code happened by multiple development teams, some of which moved out to a different role or out of organization. Some of them are not even in the organization anymore. Correct. And there may not be even any documentation left. So imagine that, and imagine how this kind of impacted the operations of this application over time. So the client put a team dedicated to understand how this code works and understand which parts could be made leaner, which parts could be better built to take the advantages of the cloud's agility. And it took them over two years of trying without a result. And why is that? Well, we as humans, we are not built to go and look through 1.5 million lines of code and understand what business functions are buried in there. But luckily, AI is there, and AI is very good at this. That is a gnarly problem. So how did we apply AI for code to actually solve this problem? So we built an AI model that helped us in a very short amount of time to come through all the code in this application. So what the AI model helped us not only identify just which parts of the code are obsolete or no longer in use, which parts of code are redundant, and also which parts of code can be grouped in a better, more manageable groups of code or rather microservices. Not only did AI help us recommend what are the suitable business function-driven microservices, but we can also use AI to help us generate the code for these target microservices, further simplifying the time. That part of it is automatic, is AI is helping us write the code that is a target microservice. Correct. Yeah. So further it saves the time and effort for the developers. It can also tell you where the gaps, what else needs to be done to make those microservices fully executable. So as you can imagine, this simplifies and accelerates the entire application refactoring process tremendously. Not just when you think about one business applications, but look at our clients. They have thousands of applications in their portfolio. You were giving the example that just one application, a million and a half lines of code, two years. So imagine if you have to modernize thousands of applications, that's kind of the power of the technology that you and the team have developed or being able to compress that time from a multi-year effort to something that you can do in months or weeks. And this is, as we said, where the AI is very good at. It's amazing technology and I'm just very excited about the impact that this is going to have for our clients' application modernization efforts. Very often we think about software and the role of software in business and it's really becoming the language of business. So tell us a little bit about the broader implications of AI for code and what's next. Well, I'm very excited that we are launching today Project CodeNet and making the over 14 million samples of code available as part of the open source data set available on GitHub, right? If you thought that 1.5 million lines of code is a lot, think about it, 14 million code samples that we have derived out of half a billion lines of code, our team has extracted most representative code samples that can help us or help AI train and better help the developers write the software. Yeah, so in some ways many are familiar with ImageNet and the implications that that had for the AI field in terms of visual recognition efforts and so on and the explosion of the utilization of these data sets with deep learning to be able to propel the state of the art of image recognition. So what you and the team are doing here is something similar, but now for the world of code, is that right? Right, I really hope that this becomes a benchmark, a data set benchmark that can be used for the source to source translation. Yeah, I just cannot help but think again of this productivity impact of if we look in our clients and everybody's enterprise, how many software developers out there and the role that AI can provide to be able to help with modernization efforts, write better code, debug code, deploy code faster. I mean, I just think that the implications are sort of boundless and just so excited about this announcement and it's really beautiful work that you and the team are doing, right? Really, thank you. Thanks, Daria. Thank you. So from learning about the implications of teaching AI, the language of code, we're going to now have the opportunity to learn how AI is learning a different language. Let me share with you another piece of tech in which AI is learning a different language. This time, the language of chemistry. Leading much of this work is Theo Laino and his team in Zurich, Switzerland. Hey, Theo, few in the world know or associate the world of chemistry with IBM, but now what's very exciting is that we are training and teaching AI the language of chemistry and you've been a pioneer in doing that. Can you tell us a little bit how that works? Indeed, Dario, we have been able to learn how to teach the language of chemistry to the AI architecture. And the result has been a way of accelerating discovery for designing new materials that instead of taking years and millions-dollars budget can now be designed and synthesized in weeks or months. So what is Theo, the technology behind when we say AI, what kind of technologies behind the scenes are we using to do chemistry? Let me go a little bit deeper into the heart of the technology. The very first thing is that we use the AI and more specifically natural language processing architecture to curate all the chemical records from publicly available chemical unstructured records. Yeah, and the parallel perhaps for everybody to understand is that in the context of human language, let's say translation between, say, Italian and English, we would have historically curated large amounts of human translated documents and then we would have trained these neural networks to be able to do that mapping for us. But here what you and the team did was curate large bodies of patents and publications that contain the chemical translation language, right? The diagrams of chemistry of how do you go to a reaction as an example? The initial effort was made in 2018 when we made available all the trained models and also the core of the architectures, the molecular transformer to the scientific community through a portal that we call IBM Arrex and for Chemistry. The portal has been a great success. We have been gathering the attention and constructing fast growing community of roughly 25,000 users that have been using the AI models almost four million times in a slightly more than two years. So this gives a little bit the idea of how interested is the audience? How interested is the scientific community in the journey for the digitalization of chemistry? Yeah, Theo, I think the work that you and the team have done on the core AI, the digital platform IBM Arrex and the digital robot that then helps you synthesize has really been remarkable, is the pioneering of a whole new field. And I think, Theo, what you and the team have demonstrated in the example of chemistry, which is a notoriously difficult and challenging language, is really remarkable and it's illustrative of the potential. So thank you very much for the first for the fantastic work and for spending some time with us today. Thank you, Dario. You just saw two powerful examples of how IBM research is advancing this state of the art in AI and reimagining how teaching AI new languages will accelerate software productivity and scientific discovery. Language, automation, and trust are the three pillars of IBM's AI strategy. With innovation from IBM research now being integrated into IBM's Watson AI portfolio faster than we've ever done, this is the time to scale trusted AI for business.