Episode 6: (Not Enough) Power of AI | On Your Radar Podcast

Don: Good morning and good afternoon. Welcome to On Your Radar. I’m your host, Don India. I have the distinct pleasure of speaking with industry experts and deep-diving into topics focusing on data privacy, corporate compliance, cybersecurity, and artificial intelligence. Today is no different.

I’m extremely excited to introduce our guest. So let me set the stage for you for our guest who has a storied career in the technology sector. So much so that I actually had to write it down because I couldn’t memorize the entirety of it. His journey progressed him through Deloitte Consulting as an Industry Practice Director in the telecommunications industry, followed by many executive and CEO roles, including the Chief Technology Officer, Chief Strategy Officer, leading to being named the CEO of AT&T Communications, where he was responsible for AT&T’s global telecommunications and video services.

And finally, after spending three years as the Chairman of the President’s National Security Telecommunications Advisory Committee, he is now on the board of directors of Lockheed Martin and Palo Alto Networks and is the CEO of Qudit Investments, where they focus on long-term investments to foster innovation and technology, leveraging artificial intelligence and machine learning.

Don: John Donovan, thank you for joining me today on On Your Radar.

John: Thanks, Don. It’s great to be here. Really enjoy the opportunity.

Don: No, it’s great to talk to you, John. Before we dive into our topic, I really want you to talk to our audience about your professional journey. Can you tell our audience how you arrived at where you are today? Please give us a little bit of detail.

John: Sure. You know, I’m a bit of an outdoor cat. My dad gave me advice when I was young that, in the early part of your career, you should, to the extent you can possibly do it, say yes to every opportunity where you’re going to gain experience. And then, as you progress in your career, you sort of narrow down and become an expert in what you choose to do.

And then, if you get that right, then you really can write your own script for the latter part of your life. And I took that to heart. So I’ve done a lot of things, a wide set of diverse activities, and they served me very well up to this point. I’m really blessed and fortunate to have seen some pretty remarkable changes during my career.

Don: John, that’s a wonderful setup for our topic of conversation today. We’re going to talk about technological advancements because we are in the infant stages of one right now. And we know we talked about artificial intelligence a lot over the past few months, artificial intelligence is an emerging state.

Yes, it’s not new, but it’s now available to the masses. Everyone can use it. So what I’d like us to talk about right out of the gate is your career; you’ve seen a tremendous amount of technology advancements, disruptions, and the applicability of new technologies into the lives of everyone in the world. 

Can you walk us through some parallels to what you’ve seen and experienced to what we are seeing right now with the emergence of generative AI and the use and consumption of artificial intelligence?

John: Yeah. Yeah. You know, I think anybody who would answer this question who has, you know, 30-plus years of experience would have remarkable stories of how technology has impacted them. I think in the early part of my career, being in telecommunications, you know, the first element of it would be deregulation, which really caused a proliferation of startups and advancements of technology in telecommunications. And then, and so as networks moved to IP, we started to get a whole lot of people that were building franchises on top of networks that didn’t require excruciating detailed experience and development work down inside the network.

John: And so, IP as a protocol really synergized with the evolution of the internet, which really allowed an abstraction. And so the analogy I always use is when I was studying electrical engineering in college, I used to have to write print drivers. you know, what is the, how big is the X axis, the Y axis, the scaling, the colors, and now we press the print button and no one will ever look back at printing as, you know, a challenge.

John: We continue to do things like that in all aspects. The first big transition was IP and, you know, the internet coming along. The next big one, which I thought was going to be the biggest in my career, was mobile. And, so a lot of people forget. I actually ran the technology group as the iPhone was launched into the world.

And we had an exclusive in the United States for five years. I think the exclusive, if I recall, it was three years for anywhere in the world. So we were uniquely solving problems with Apple as this idea of, you know. touchscreen and an app-based capability sits on wireless. So networks go to IP, and networks themselves go from wired to wireless.

And then we, we embark on the cloud and so the cloud, I thought, well, this surely is going to be the biggest 1 because the one before it set the stage for it. And this one is going to be the thing where, architecturally, technology shifts. And then this last one with AI is probably the most remarkable and the most fun.

I think what makes it a little bit distinct is that it’s in its early and nascent stage, for sure. But if you were a technology user, using the Internet, going from mobile, or moving your enterprise or applications into the cloud was easy because your role was just to be a user. The tools were robust and happened pretty quickly and developed.

And AI is sort of… you have to figure it out. You have to figure it out in your own context. And so it’s almost a first-principle type of revisiting of technology. And so I think that’s what makes it exciting.

Don: No, John, I appreciate that. When we look back at mobile cloud moving into AI, there are a tremendous number of other technological advancements. But if you think about the value each one of those has provided either to an individual, to other organs, or to enterprises, it’s quite significant.

And it allows for just the advancement, not only of just individuals, but advancement of different careers, advancement of completely different industries. So it’s fascinating to see where artificial intelligence and Gen AI is going to move forward to. 

So, let’s expand upon that. You from Qudit Investments, you invest in new or newer artificial intelligence organizations that are organizations leveraging AI.

Talk to us a little bit about how that is looking in terms of the landscape of investment strategy.

John: Yeah, you know, I started an AI in parallel with my son’s work a decade ago. he came home one Thanksgiving and said, you know, heart rate monitors aren’t very accurate. So I want to go buy some sensors and he started working on signal to noise problems and then he started having to filter the noise out and he ended up building over the Thanksgiving holiday an analog that was run by an iPhone, but he stayed with the field about, you know, signal to noise and the subtlety of trying to extract knowledge from noise.

And he and I have sort of worked in parallel along that. And so he’s been at the machine learning side of things for quite some time. And it allowed me really, you know, when you’re when I used to always say. If you want to know how people really think about technology, ask them what their kids do, because it, you know, it’s different between what you say and what you do, and you’re not going to lead your kids in the wrong place.

So, I put my son into AI, or he put me into AI. However, we arrived at that point more than a decade ago. So, while everything looks revolutionary, you know, revolutions in technology are just evolutions that hit their tipping point. And so that’s what happened. And when chat GPT came out, there was sort of a mad race, but the field itself had been moving along smartly and nicely.

Accuracy of these models got better and better and better and better and better. Then they hit a useful tipping point where you got a nice UI and an easy way to consume it. And voila, you know, we now had a machine that could write love poems better than me for the rest of my life. So that’s sort of awesome.

But I think that there’s, you know, so much that needs to get done. So from an investment standpoint, here’s kind of my general thesis. 

There’s four things that AI Is not today. 

1. It’s not accurate enough. 
2. It’s not private enough. 
3. It’s not explainable enough, 
4. It’s not power efficient enough. 

And so because of how big AI will be from every person’s estimate, one, two, three, five percent gains in any of those four vectors is in and of itself a market.

So, while I think a lot of people now are fascinated by it and moving from fascination to, you know, what practically can I do and implement to make a difference in how I’m doing my business, how I’m conducting my life, these things are starting to crop up. 

So you see things like, you know, I’ll reference a couple of external surveys I just read one said that venture capitalists spent spent spent 50 billion dollars on advanced GPUs from NVIDIA H100s for 3 billion in revenue. And another survey said the average vertical AI startup is spending 40 percent of its revenue. On the compute and large language model operations. So that’s just an example of how efficiency manifests. 

And then you can hear stories all day long of someone who says, I was deploying AI to clean up old, I heard one day, the other day, they were using AI to clean up old files and they gave the AI the permissions that it needed to get the old files Sort of clean up, their data and they were restricted.

The AI was restricted from getting files appropriately that it didn’t want erased. And so its next step was not to stand down, but it went and started to find a way to get itself the credentials. So that it could go erase the file. So, you know, these things that are small error rates, you know, it’s, it’s an error in a poem, is not a problem, but an error in something like that could be, could represent, you know, a leakage of critical customer data or, you know, any other, you know, catastrophe of, proportion, to the business that would be really, debilitating.

Enterprise Risks of AI

Don: No, John, that’s actually a great segue into a question I was going to ask you as we expand upon artificial intelligence. There’s tremendous benefit and there’s tremendous risk. You started to address the risk by talking about it’s not private enough. It’s not power efficient enough. It’s not explainable, nor is it accurate enough from an enterprise perspective.

But put your put your CEO hat on. What are the enterprise risks that you see a large corporate or medium size or small corporate enterprises of leveraging artificial intelligence equally. What is the risk of not leveraging artificial intelligence?

John: Yeah, you know, I think, you know, Don, that’s a really insightful question because the only really bad strategy here is to wait. Because there’s either in almost every case I’ve witnessed, there is either existential risk or unparalleled opportunity. So, if you’re not looking at how you would redo your business from an AI perspective, then someone else is.

And so the longer you wait, the more likely that a disruptive force can come in and fundamentally rethink how a process runs how, you know, a company serves a need in the market. So I think that’s the really bad outcome. 

I think there’s an outcome of sort of foolishly rushing ahead, which is also not a great outcome, but I think, you know, the, the word I, I like to gravitate around is be deliberate, which is neither fast nor slow.

Just be appropriately at pace with where the opportunity or the risk is. And I think with that as a guide, probably you’re in pretty good shape.

Don: Yeah. I think from the organizations I speak to, they are being very deliberate on how they use, especially in the world of privacy and compliance of which I find myself in, there is a deliberate use and a deliberate not use or lack of use with respect to artificial intelligence to be certain that there’s decisions being made. That is accurate decisions. That is the right decision. So there’s a lot that we have to look at in terms of where this future goes. 

Now, if we’re in the consumer, the B2C business, that’s a different ballgame than it is B2B and large enterprises. So we have to look at it objectively where any one of our audiences in the, in their types of organizations, where are you and who are you serving is really how you need to look at your deliberate approach to leveraging Gen AI and artificial intelligence. Let’s go backwards. We talked about the risks. Go ahead, John.

The Benefits of Leveraging AI for B2B

Let’s talk about the benefits, right? So we talked about risks, risks of not doing, risks of being deliberate, risks of someone really outthinking you and outpacing you, but there are tremendous benefits, too. So give me some of John Donovan’s benefits of leveraging artificial intelligence for organizations.

John: Sure. I think that we have a new opportunity to generate growth. And I think that, you know, there’s a number of forces at large that make growth at sort of the country level, really difficult to achieve. 

I just noticed this morning, the forecast that the G7 put out for each country and, you know the growth rates when the population was growing in our country, as we were, you know, innovating, you could, you could get 3 plus percent growth rate and, you know, 1.5 is the new 3 and we’re saying 1.8 is a banner year. 

I think that we’ve got an opportunity to go out and restore growth. And I think that’s a fundamental that people can really be focused on. I think that the second benefit is, I think in terms, Don, there’s, there’s three types of ways that you could apply AI to efficiency.

Three Ways to Apply AI to Efficiency

1. The first one is to take the same workforce and do more. And that’s not terribly risky. That’s pretty much an asymmetrical reward scenario, because if you’re going to pick up growth and you don’t have to hire more people, that’s great. Because most businesses are saying I have a people shortage, or I have a, trained people shortage.

So there’s some consequences we need to think about in how we do displacement and restoration, but you know, that’s a pretty low-risk thing because the humans are aided by the AI, you’re getting more done with the same people, but the processes themselves, In that scenario are replications of the things that people are doing today. So it tends to be a pretty high reward, low risk. 

2. The second thing is to take the same things you’re doing today and do it with fewer people. And I think that’s the scenario most are worried about, but I’ve yet to really buy into the idea that the short-term pain of productivity improvement is a bad outcome for a company. 

An economy or a country, and I think that one of the ways you grow is to keep pushing, displace and then replace in another area, you know, those units of whether it’s capital or labor and so the second type is the one I think that people are most concerned about, as it relates to people 

3. And then the third is to use AI to do things that we haven’t done before. And that, you know, is areas like medicine and some of those where people say this can really be transformational, but it’s also spookier because these are things we don’t know how to do. 

In the first two scenarios, where we do more with the same people or we do the same with less people, it still is people-driven processes, and the third scenario, these are more machine-driven processes, machine-driven outcomes, and that’s where You know, reliability, robustness, robustness explainability become far more prevalent.

So, on the plus column, you can fill those up in a quick hurry. You know, one is, you know, look, shortage of labor. I can overcome it. The next is I can increase my profits, put resources out elsewhere, put more into R&D, and the third is I can go out and create brand new markets. And so I, I sort of take those 3 columns and those 3 check marks, and I sum it up and say.

I think if we do this the right way, we’re going to end up enhancing growth in the, in the U.S. economy. And I think that’s going to create room for a lot of people to be successful, not only to continue success but to be successful. In either way they weren’t or finding classes of, of participants in the economy who may be, were, under-benefiting from it can now enter into the workforce or enter into a more prosperous part of the workforce.

Don: Any technological advancement has created efficiencies. That’s actually the nature of why they keep evolving themselves. It has created repurposing compositions, as you talked about in your second item, the redoing of existing processes, but it’s also repurposing those resources, as you mentioned. So it allows for growth in many ways, and new jobs and new industries get created because of it.

Scaling and Data Access for Large Language Models

Don: You and I had an opportunity to Dario Amodie back in February speak, on his building of his Clyde LLM. And now I’m reading in the news that all these large language models are truly running out of consumable data. What are your thoughts on that? And how, how do you, how does a Sam Altman or a Dario or other LLM builders look to create or find incremental data to allow for these LLMs to be further educated and further refined, which gets to your earlier point of the accuracy of their overall models.

John: Yeah, well, I think that, you know, the, when you, when the quantity of data is the most important element of building the model, then you run into the threshold of there’s not infinity data. Yeah. You know, I think some emerging economies who, you know, use low wage rates. As their appeal to bring work in eventually there’s not infinity people. So you run out of workers prices go up. 

I think that we’re going to move from large language models to logical language models. You can don’t have to change any of the vernacular and keep your label on everything. But yeah, but the logical language model was really about at some point. I’ll use an educational parallel.

I think the most applicable models, of the future to do most of the enterprise AI that’s necessary is going to gain its efficacy by reducing model size, getting the foundation model to be something more like a bachelor of science and bachelor of arts, a bachelor degree. And then you bring that model in-house and then you use specific experience, knowledge, humans, and data inside your company. And then you don’t wait for the model to outrun. 

And the simple example I always use is most United States-based enterprises don’t need a model that knows who the social media influencers in Japan are to make their business better. Yet in some way, you’re paying for the model to have been educated in things like, like that.

So how many variables or parameters or how much data is going to build large-scale models that can do a bunch of things that are, I think, really useful, but I think they’re more useful for exploration, consumers, creative processes, and I think logical language models are going to create a second wave.

And in the meantime, there’s a bunch of innovators out there trying to innovate chips that are specific to the task at hand. They’re trying to innovate on the foundation models themselves. 

And so while most people would say foundation model is a spent area and the infrastructure is already invested in I think that when we move from large language to logical language, we’re going to follow behind that with a new set of infrastructure, whether it’s the big guys evolving or whether it’s the little guys innovating, I think we’re going to come up and ultimately we’re going to end up where we must be with models that don’t have data leakage and jailbreak and privacy problems that they’re not, they don’t cost an arm and a leg to run. 

You don’t have to, you can explain how you got results. You can deactivate data sets. You can, you know, shape a model to do something that’s more consistent with what you want done. You know, there’s a, there’s lots of cases of a model being, empirically correct, but disastrous from what you really intended or wanted.

Don: Definitely. I do have a question for you. Did you trademark logical language models yet?

John: you know, no, no, I haven’t,

Don: One of those, one of those things. You heard it here. You heard it here first.

John: no, I don’t know if I, you know, I, I don’t know where I sourced it. I think it actually was my son who, who, told that to me because he said, he said LLM and I, and I said, I don’t think you’re, we’re talking about the same thing. He said, oh, no, no, no, I don’t. LLM, for me, means logical language model. So he picked it up somewhere.

What’s On Your Radar?

Don: Very good. Last question for you, John. As we close on your Radar podcast, I always ask this question. As you look forward to the remainder of 2024 and into 2025, what’s on your Radar coming up?

John: Yeah, I think that 2024, the back half, I think we’re still, as an economy and still. Within the enterprise sector, I still think we’re working on constrained budgets. I still think there’s a lot of uncertainty. We have an election coming up. And so I think we’re in a year of maybe flat technology budgets at the same time, innovations exploding.

And those things usually, you know, end very productively. So I’m really, you know, bullish on where sort of the back half of ‘25 is going to go because productivity improvements are around the corner. There’s some tech adoption that is sitting in backlog, if you will. And I think a lot of the trepidation of buyers in ‘23 has bled over into ‘24 that says I’ve been spending too much on technology. I’m not getting enough top-line growth, and I think that’s why that earlier comment I made. My optimism is really centered around revitalizing a growth-led economy. It solves so many problems for the U.S., and I’m so, you know, that’s what’s on my radar.

If I drill down just slightly, I think that I’m starting to see the green shoots of problem-solving in AI in those four areas: explainability, efficiency, privacy, and accuracy. And as those start to happen, I think we’re going to get better and more results. So I’m very optimistic about what ‘24 has to offer compared to ‘23 but really bullish on what ‘25 has to offer versus ‘24.

Don: Well, John, thank you again for joining us today on your radar. And thanks to everyone for listening to On Your Radar. On Your Radar podcast is made possible by Radar First. Radar First automates intelligent decisions for all of your privacy and compliance regulatory obligations. You can learn more at RadarFirst.com. And if you liked what you heard today, please follow On Your Radar podcast and look for us for our upcoming episodes. Our next episode will be forthcoming next month. Thank you.