Microsoft’s Chris Young on What’s Next for AI

So what you just described are two great examples of autonomous systems. A couple years ago we did some work, and we said, you know, this was even pre-generative AI, we felt that autonomous systems were really going to be some of the future applications that would really change how we lived our lives, how organizations function. But when we look at that space, we think it's going to be a combination of self-driving cars or self-driving vehicles, to put it more broadly speaking. I think it's going to be less about transporting people around and a lot more about transporting goods and services around. Ultimately, long-haul trucking would be an example of that, or just think about all the vehicles that operate in small warehouse contexts or even in a shipyard, as you move goods off a ship onto trucks and the like. And then on the embodied side, look, I think humanoid robotics is going to happen. That'll be something, it'll be quite commonplace, I think particularly in categories of work where it's very dangerous for people. Think about offshore oil rigs or Antarctic oil drilling. There's just going to be a lot of different opportunities, I think, for us to bring new types of embodied technology into those kinds of categories, and it's pretty exciting.

We're great partners. It was even mentioned by Jason a little bit earlier. You know, we invested in OpenAI a number of years ago, even pre-pandemic. Over time, we've had an amazing partnership. If you think about all the great things we've been able to do together as partners. If you look back, it's just been a great partnership, and I would say one of the more unique partnerships you would even see in technology. Obviously we have a lot that we're doing together. You even saw at our Ignite event a couple of weeks ago, we announced a number of new innovations. All of our Copilot innovations are built with OpenAI technology as part of it. So we're really excited about the partnership, and we continue to do a lot together.

So three big things at Ignite. There were many, we had many announcements, I think 80 announcements, but three big ones. You just mentioned Copilot Actions, really templates or really being able to bring more agentic functions to Copilot. The second one was agents, which is ultimately going to be the extension of how people do their work and integrating that into a process flow. And the third is really Azure AI Foundry, which is going to enable developers to build a lot of their own Copilot-like capabilities, build their own agents. So we were really proud of all the announcements, but those are some of the big categories that we announced a couple weeks ago.

So first of all, think about Microsoft as a platform company. We are bringing a diversity of capabilities to help developers build their own tools, their own Copilots, their own agents into Azure. And that's really what Azure AI Foundry is all about. And you're seeing a lot of that. I would say if you look back, we were really one of the first companies through GitHub Copilot to bring capabilities for developers to do more code development. Now what we're doing is bringing capabilities and tools for developers to build agents, to build Copilots, and to do that using a diversity of models. We have a lot of different models on our catalog and are allowing developers to use what works best for them.

So what you're referencing is some work that came out of our Microsoft Research organization, and they published a paper on Magnetic-1, which is an agentic framework. And if you think about it, it's really based on a lot of the theory that agents will function in teams, in some ways just like people do, and that you'll have a lead agent that'll direct some of the specific work that will happen by very specific functional agents. Part of the reason for that is as you go through any completion of a task, whether it takes a period of time or it has multiple facets to it, there will be changes, there will be adjustments that need to be made, and the lead agent will direct that work similar to the way a team leader might lead work in a project. So it mirrors that concept. But if you think about it this way, ultimately agents will need to function in concert with one another because there'll be many different agents that work in an organization that might even work on behalf of an individual. So having a lead agent that directs the functionality of other agents is really the core concept of that. And you've seen a lot of this at Ignite. We're focusing on extending the value of what we're doing with Copilots into the world of agents, and the multi-agent framework is really a part of that.

So look, the way we look at nuclear, nuclear is a clean energy source that will become part of our overall portfolio of energy sources that we use for delivering our data center capacity around the world. And we expect there's going to be a diversity of energy sources. We've made certain commitments as a company to being carbon neutral and even carbon negative by 2030. So nuclear is a part of that portfolio, and that's really what that announcement was. It's part of a much bigger strategy that we've got as a company.

So the way we think about sustainability is number one, it's a must. And we expect that there's going to be a lot of different... In fact, there's a recent blog that we published which shares some of the trends that we see. One of the big trends is really the ability to deliver more and more compute capacity with a lot better efficiency. So for example, better efficiency around how we build data centers in the first place, better efficiency on how we do cooling in data centers. That's one of the big ways in which power gets used in a data center context, is that the cooling takes a lot of energy. Being more water efficient in how we run our data centers as well. So we have a series of different methods that we're employing in our data centers. Some of them we've been doing for a while, others are new. We continue to work with our partners, all of our partners like Nvidia, AMD, even the in-house silicon that we're building at Microsoft. All of that is with an eye towards making our data centers more efficient, more effective, and ultimately part of a more sustainable delivery mechanism across what we do around the globe.

Well, I'll tell you what we're investing in right now. Because as was mentioned earlier, one of my responsibilities is for our venture arm, M12 at Microsoft. So a couple of key categories that I think are really important. One is broadly speaking AI infrastructure. But if I were to break that down into some core components, one aspect of that is going to be cybersecurity. Near and dear to my heart, I used to be in cyber for a long time. For example, we've invested in a company called Hidden Layer, and their focus is protecting models against being manipulated by third-party actors, because the models themselves will become an attack surface for malicious actors. So that's just one example. We're also seeing need for data governance and data compliance, data protection. So if you think about it this way, as we bring more AI capabilities into organizations, those AI agents, the AI Copilots, are going to want to access information. You need better governance, you need identity capabilities. So we're seeing a resurgence. We've invested in a number of companies. Reliance AI is one we recently invested in, and that's in the data governance category. We also have a number of our own products at Microsoft, first-party products that are in this category. And then we're even seeing the need for new databases in some cases. One of our recent investments was a company called Neon, it's a serverless Postgres database. And we're seeing a lot of demand for different database models, different data architectures as it relates to developers building new applications that are AI-first in nature. So there's so much happening right now, but that broad category of capabilities that are AI infrastructure enablers is a big area that I think is important to focus on. That's where we're making a lot of bets. And then some of the other ones you just said a moment ago, embodied AI. The other, the last one I'd add is I do think there's a lot of opportunity in generally speaking the healthcare space, all the way from patient care and delivery through drug discovery and development. A lot of exciting companies that are using AI. One example I would share is we recently invested in a company called Inveda. And Inveda, what they do is they're using generative AI to essentially study what happens in nature and look for the natural remedies that nature has for certain ailments. Eczema is one that they've got actually a few drugs in trials to treat eczema using natural remedies that occur in nature. So that's just one example of many. We've invested in companies that are using AI as a better mechanism for diagnosing or assisting pathologists and radiologists in the diagnoses of cancer by being really good at the screening aspects of this. So I think there's a lot to be done there across the board in healthcare, and it should hopefully raise the bar for health outcomes for all of us.

I mean, there's near-term fears and there's long-term fears. I look, my wife would tell you that being married to somebody like me, you have a certain degree of paranoia as your baseline. Same with a journalist, though. Yeah. But I would tell you, look, I think there's a real need for responsible AI. It's one of the reasons why at Microsoft we've invested really heavily in really testing a lot of the use cases around looking for what we think could be model manipulation, misinformation, disinformation. There's a lot of focus on that the last couple months here, and just even in the United States given the election, people were really on the lookout for that. And I think we came through, all things being equal, came through without a lot of too many big alarms on that front. So I think we have to be mindful though that everything that we do with models for good can be flipped and turned to be used by malicious actors. So just making sure we have the right security models, it's why responsible AI is really important. It's one area we've invested heavily in at Microsoft, and we encourage all of our own customers, our partners, to have their own investments in responsible AI for that reason. In the long run, we'll find new things, we'll have new challenges we face. But I actually am pretty confident that we're making a lot of the right decisions and we'll work through challenges as they arise.

Please, thank you.