Revolution of AI: Elevating the Human Experience | RSAC Virtual 2025

Hi, good afternoon, everybody. My name is Dave Ksky. I am a Dell Fellow in the office of the CTO for Dell Technologies.

So, whether you believe that the current state of AI is an evolution of smart, uh, technologies, or whether you believe it is a technological revolution, the state of the, the matter is that AI is part of all of our everyday lives. So what is the question that's on the table? The question is on the table is, is AI a threat to my very livelihood, or is AI a tool that I can use to elevate my human experience?

About 85% of surveyed workers said that they expect AI to impact their jobs over the next several years. 30% of those workers said that they are worried that AI is got, AI is going to actually evolve to the point where it takes their jobs, and they are actively interviewing for new roles. And about 75% of those workers said that they believe that AI is gonna be responsible for less jobs in the future.

That's a pretty dire situation. But let's look at the flip side. 75% of workers said that they use AI in their everyday work in their everyday life.

And nine outta 10 of those say that AI reduces their time to complete their tasks. Well, now that's an interesting dichotomy. Now, how does that translate to the consumer?

Well, consumers have been using AI for decades without even knowing it. Majority of people are using Siri or Alexa, or they have a collision detection system in their car, or they have a smart thermostat. They're using AI in their everyday lives.

But what about the overt use of ai? When you actually ask someone, are you willing to use an AI agent as a digital assistant? About 45% of respondents said yes, but if you go to Gen Z, that goes up to 70%.

So then we broke it down a little bit further and we said, okay, let's, let's get a more, a little bit more granular. How about looking for a job? How about grocery shopping?

How about health and wellness? Would you be willing to use an AI agent to do those tasks for you? And we got about the same re responses.

About 45% said yes, and 70% of Gen Z, they're all in. Whether you're fearful or whether you're embracing ai, you know, the fact is it's a part of all of our everyday lives. And the people who are embracing ai, they may, may be wondering, it's probably most of this room, right?

I would imagine, right? You're all wondering how the heck can such a large percentage be so fearful of AI impacting them negatively? Well, there's a perception that AI just popped outta nowhere.

It's a technological revolution that's gonna evolve to the point where robots are gonna take over our world and they're gonna take over my livelihood. Now, you all in this audience, I'm sure, realize that that's not true. And we all believe that ai, the current state of AI, is a technological evolution that is actually going to positively impact us.

So let's actually see, start, uh, addressing this, uh, question of evolution versus evolution. Where did the current state of AI come from? Now, I'm not gonna do a big history lesson here because a lot of you all already know this, but we've been putting AI based solutions out into the marketplace.

As I said, for decades, we've been using approaches like linear regression, uh, supervised learning, uh, uh, rules-based engines to address things like customer behavior, modeling, recommendation engines, and even hardware and platform optimizations. But then something happened in 2014. So in 2014, we took, uh, adversarial networks, generational adversarial networks, and we applied those to machine learning and Gen AI was born.

Now, of course, gen AI was so different because this AI system actually created new content out of trained data, and then November, 2022 came along. Does anybody know what happened in November, 2022? OpenAI launched chat GPT.

Why was that such a milestone moment? It was a milestone moment because it was the first time that there was a publicly embraced large language model that was paired with a natural language interface. The natural language interface made it approachable for the mainstream, made it very easy to use, made it so easy to use that we quickly realized that the power necessary to scale AI was just not sufficient.

And so we had a problem on our hands. Now, we also quickly realized that the power and cooling that went along with that was also not sufficient. So what does this room do?

When we have a problem, we attack the problem. So while models, and you've heard today throughout the day today, uh, the evolution of models and the progression of models and where we are in models becoming more performant and, uh, um, uh, better fit for, for many different solutions. But we've had to take those models and apply model adaptation.

A lot of you in this room are probably, uh, playing with different levels of model adaptation. Of course, quantization is the big dog in the room. We've been doing quantization many for many years, and we've been having some great successes with quantization.

But you've also seen new model architectures start to be introduced. Uh, deep seek was the one that hit the hit the press. But, uh, you know, being able to introduce model architectures that have test time scaling, um, and other, uh, and other approaches has really made a difference.

Now, what is the difference that it's made? The difference is we now have medium and small language models that are just as performant as large language models for the solution to which they are targeting. Now, that may say to everyone, okay, well that's great.

You know, we've, we've relieved our problem with, uh, compute bandwidth that we need to scale ai. Uh, but that's not the only thing we've done. We've now taken performance language models and made them able to be put on different platforms.

You now not have to run language models on big iron backend data centers. You can actually run them at the edge or at the far edge. And so now the, uh, in just, uh, this, this last month, we've seen the broad launch of, uh, a IPCs, and now we have agentic AI systems running locally on a I PCs addressing things like latency and privacy, agentic ai, AI agents.

I don't think I go more than 10 minutes in my day until I hear the word AI agent. Matter of fact, I was just sitting at lunch and I was counting the number of times that I heard AI agent with the people around me. And I think I, I stopped counting at about 20, and I'm, I'm not kidding, right?

So there is a, there is a boom, there is a explosion of AI agentry being used in solutions. And why is that important? Well, if you look around, uh, you know, there's, there are a lot of solutions that are very, um, closed and, and, and they, they maintain their own agentic solution within, within the bounds of their application.

But there are a lot of, uh, solutions that are being played with that are open agent solutions. And now, the CHA challenge with Open Solutions, it's uncovered a lot of problems that are barriers to us, uh, scaling this quickly. Things like incompatible AI agent frameworks, things like distributed data sets, and being able to have access to that without compromising the security models.

Things like silicon diversity. Currently, every AI model has to be tuned for the silicon that it's running on. Completely untenable if you're gonna have a distributed model.

So a lot of these tough problems are being addressed, but the answer is, we know that we are entering an era of AI, hybrid and distributed solutions. So, Dave, thanks for the history lesson, right? Um, what does that have to do with evolution versus revolution?

So, I, I picked up a couple of, uh, uh, definitions that I wanted to share with you so we can make some determinations in the room here. So, an evolution, an evolution, technology evolution is defined as a gradual continuous development and refinement of technology over time, driven by, by human innovation and the need to solve problems, often building upon existing technologies. So if I look at that, I could say, all right, well, the application of gans to ml, that's an evolution and model adaptation.

That's an evolution, right? So how many in the room believe that the current state of AI is a technological evolution? At least two.

Awesome. Well, but wait, let's look at this. So I mentioned chat, GPT.

So chat, GPT is the fastest growing consumer application in history. Within five days, it reached 1 million users. Within two months, it reached 100 million users.

So if we look at the definition of a technology revolution, it's defined as a period of rapid and significant technological advancement that fundamentally alters industries, economies, and societies, often leading to widespread adoption of new technologies and transformative changes in how we live and work. So, is it a revolution? You know what I posit it doesn't matter.

What does matter is that we're now looking at how we build, deploy, and maintain solutions differently to incorporate AI as a base and incorporate agentic ai, which will change the way that we look at the future going forward. Now, agentic AI is in its infancy. Now, a, a speaker, I believe it was, uh, two or three ago, was just talking about, well, we're past agentic AI already we're moving into something else.

Well, you know, if you really look at how many AI agents that are deployed out there today in, in enterprise environments, scaled enterprise environments, we're, we're still in its infancy. And enterprises are very measured when they come to introducing new technologies that may be introduced into or affect their very critical, um, business processes. Alright, so where are we?

You know, do we have adoption of AI based solutions and agent AI and enterprise? And so I picked just a couple here. So, uh, Dell Technologies has been helping our customers, uh, deploy infrastructure for AI and help them deploy AI solutions for several years now, uh, we just picked these six, I picked these six because it shows a very broad and diverse set of solutions and the impact that they've had.

We have everything from, uh, traditional content creation to, uh, a solution that brings communities together so that they can communicate. And then one of my favorites is improving outcomes for patients through the use of ai. But look at the numbers.

You know, look at the numbers below. This is not a small impact. These are massive impacts.

So that tells us that AI based solutions, agentic AI is being deployed, but it has to be deployed against the most important problems that an enterprise is having. If you're just, you know, picking a low hanging fruit and applying AI to that, you're not gonna have a return on investment, and you're not gonna get funding for the next project that you come along because you didn't have return on investment. So picking the right solutions is important, but the impact is incredibly impactful, right?

So this can be very scary for many who are fearful of ai. Like, oh man, you know, I know AI is gonna evolve. It's gonna come after my livelihood, it's gonna come after my job.

And look at this, it's not going away anytime soon. Now I look at the introduction of AI very differently. It's a world of opportunity.

And why is it a world of opportunity? We have created a brand new landscape that is allows us to provide value to our customers to innovate, and most importantly, to answer and solve problems that we have never seen before. So I mentioned that there's a lot of work going on to try to figure out how we scale the compute power necessary for ai.

That's true. We're also doing a lot of work in how we get more performant models. That's also true, but right on the heels of that are new memory architectures to be able to handle bandwidth, new storage architectures to be able to, um, remove the burden of, uh, distributed data lakes and new communication architectures to not only connect NPU to N-P-U-G-P-U to GPU, but also to connect agents to agents, have agents be able to discover one another and work in tandem, and to be able to reuse agentry that's been developed.

And they're no, no longer one-offs. And of course, who can forget security? We are at RSA ai, uh, AI agents has again, created an entire new landscape for security.

Now, you've heard a lot of it today, but I mean, just off the top of my head, there is, uh, identity and access rights for non-humans, for agents that needs to be developed and, and codified. We have new platform security models that need to be, uh, put in place. We have now distributed data models that can, changes the entire data model security that we may have in place.

Those are just a couple off the top of my head, of course, model adaptation and, and, uh, uh, model progression will continue. But then the thing that is really gonna open this up and allow the adoption of agentic AI is standardization. The ability, as I said, for models to be able to communicate with one another, discover one another, and be able to enable the reuse of agents and standardization work has already started.

So through MCPA two A, several others we're really seeing some advancements in, in some, uh, progress. Um, and I loved, and I loved seeing the amount of companies who are engaging in the standardization work as well. So what does that mean for the fearful?

Well, they should be able to see that there is a technology landscape that's ever, uh, ever evolving and ever growing. We're gonna need technology professionals to drive the strategy and most importantly, drive the implementation of this evolving vision that translates to more jobs. A lot may look at a lot, a lot of, uh, um, enterprises and a lot of the fearful may look at this and say, okay, Dave, you just described this world, this wonderful world of opportunity where we have a brand new landscape and we have, uh, a lot of new technologies coming on board.

And Jesus whole AI agentic thing. I have no history in that. I have no idea what I'm doing.

This might just be a bridge too far for me. Well, that's where the large system integrators, little pitch like Dell Technologies, and many of you in this room can help. And this is a great week.

So if any of you have solutions, if any of you have trough problems, if any of you of you have barriers, come talk to us. We can help. Now, question is, all right, David.

So we, we went from MLDL to Gen ai. Now AI agents, what happens when this startup phase is over? You said that we're generating a lot of jobs, we have a lot of new technologies.

Well, won't that, you know, kind of, you know, kind of enter the, uh, the maintenance mode. And the answer is absolutely not. We have a long road to go.

We are moving from a world of single AI agency into a world of complex reasoning models where we need orchestrators and planners that are able to see, again, discover what AI agents are out there, what capabilities we have, and maybe even spawn new agents with new capabilities dynamically real time, and then put those agents together in a way to solve complex problems. We see AI agents being combined in fleets, communicating between agents to share context and sensor data so that they can navigate complex environments much more effectively. Examples of that self-driving cars, you know, the promise of self-driving cars talking to each other has been on the table for about 10 years.

We're finally getting to the point where that can be reality. And of course, uh, delivery drones and, and many other applications like that will benefit from, you know, simplifying their complex environment. Now, I have drawn here autonomy.

You may say, Dave, why do you have autonomy just drawn on the right hand side of this slide? Don't we have autonomy across the entire thing? And the absolutely the answer is yes.

And don't look at this as a timeline. This is not a timeline. This is a spectrum of solutions.

Just because we're starting to enter the world of multi multigenic, uh, reasoning models doesn't mean that we're not gonna be continuing to produce, uh, ML based solutions, if that's the thing that solves, it solves the, uh, uh, job. I mean, the, the gentleman from meta who was up here early talked exactly about that, that before you start to think about a small language model or a larger language model, see if you can do it a little bit simpler, right? Don't overcomplicate your lives.

But given that in order to get to autonomy, we have to increase trust. And when I say trust, that's not just trust in that the, the right, uh, inference is gonna come back from the language model that's balancing that with risk, um, and importance of, of the solution. But as we, as, as, as trust goes up, we're gonna be able to get to autonomy.

Now, how, how are we bridging that trust gap today? So we're saying that we are just entering, you know, POCs and, and experimentation for multi-agent, uh, solutions. Um, what are, what are we doing for the trust gap and the trust gap?

We have humans in the loop clearly, right? Today, especially, I mean, you look at an IT professional. If an IT professional is using, um, a gentech technologies to be able to, uh, look at his fleet or her fleet of devices, and it comes back that, Hey, we see this problem and here's a recommended, um, answer.

I can guarantee you that it professionals today are not going to let that agent just go ahead and update a hundred thousand endpoints without them knowing what's going on. They're gonna be in the loop. They're gonna be the one that wants to push the button.

Now, they may give that action base back to the, uh, agentic solution, um, but they may want to do it a different way, but they're gonna be look to see, to make sure that that answer is reasonable and it's in line with what they expected. Now, as a trust bar goes up, we're gonna go from humans in the loop to humans on the loop that is monitoring, watching, making sure that we're going the right direction. Are we, are we following the strategic plan?

Then ultimately, we'll get to the point where we'll have full autonomy and we can have humans outta the loop. And that's what a lot of the people in this, in this day have been talking about. Now, when you start to reach the point where you have humans on the loop, and when you have humans out of the loop, security becomes even more paramount.

Security has to precede trust. So a lot of, a lot of the, the fearful will look at the evolution of humans outta the loop for these complex reasoning models and think, oh, that just proves it. You know, I am, I am fearful for my wellbeing.

And, and I say, I don't agree. AI has been elevating the human experience for decades. From, from Siri to home automation, to collision avoidance systems in your cars.

AI has been elevating the human experience. Um, let's look at some examples Now. I was, I had a couple of security examples, but I think after all of the talks today, that would be, uh, a little redundant.

So, um, I'm gonna let those just sit. But if we look at, um, if we look at healthcare, we now have access to robotic surgeries, to accelerated vaccine development, and to genome sequencing. There are now tools available for the disabled to help them handle their challenges in the world of education, teachers are now using AI to generate custom learning plans for better outcomes.

But we ask about what about the mainstream? Well, I'll tell you, AI is going to continue to automate the mundane and the repeatable, which allows humans to move to the more creative, more strategic, and the more complex. And that gives job satisfaction and personal productivity satisfaction.

AI is not a job destroyer. It is a catalyst. And as AI continues to evolve, we have to evolve with it.

Humans have to keep cognizant of the value chain and apply our uniquely human skills to play the human in the loop. So now, after all that you may say, Dave, how do you answer those questions that you, that you put out there in the very, very beginning of this, uh, presentation? I would say that the current state of AI is a technological evolution, which has kicked off a societal and a business revolution.

We have to embrace the most powerful tool that has been made available to us in the last several decades. And it's that way that we're gonna elevate experience. So thank you.