New Frontiers in Artificial Intelligence: DeepSeek and Beyond
Tony Roth, Chief Investment Officer
Anshumali Shrivistava, Founder ThirdAI Corp, associate professor of computer science, Rice University
Tony Roth: This is Tony Roth and you're listening to Capital Considerations. I am the Chief Investment Officer at Wilmington Trust and we are launching a new season. So I'm thrilled to be talking about a very timely topic, which is the advent of DeepSeek and its impact on the artificial intelligence or AI ecosystem.
And the title of our podcast is appropriately enough: Artificial intelligence and the rise of DeepSeek. Today, I've got a perfect guest Anshumali Shrivastava, who is an associate professor of computer science at Rice University. He's also the founder of Third AI Corp. He is renowned for his pioneering work and large scale machine learning, deep learning and randomized algorithms. His innovative research has led to advancements in AI, including the development of sublinear, deep learning algorithms that do not require GPUs, or Graphic Processing Units, which are essentially what NVIDIA sells, which are the very powerful chips that I guess initially hadn't been designed for AI, but it turned out that they were the best thing for AI, if I have that right.
Anshu has received numerous accolades, including the Charles Duncan Achievement Award for Outstanding Faculty, and the Young Alumni Achiever Award from IT Kharagpur, which is a university in India and your work has been widely recognized and implemented in various commercial applications, including Amazon's product search engine, which I love now because it has an AI component.
I recently had my barbecue, the light bulb went out inside the lid. And I went online and there were probably 15 different light bulbs, it looked like, that might work. And I normally I would have bought all 15 of them and I would have figured out which 1 worked and returned the other 14 half of which would have had open packages.
But now I can go into into Amazon, maybe this is because of you, and I can type in what is the right light bulb for a whatever grill and it gives me the exact light bulb. And so, you think about what this is going to do for the Amazon return rate, which has historically been almost 30%. And if they can get it down 1%, it adds about a billion dollars of profit to the bottom line.
So thank you for that. So Anshu, thank you again for joining us today.
Anshumali Shrivastava: No, I'm, I'm really excited, Tony. And, uh, thanks for, uh, this exciting discussion. I get this a lot these days. I like the topic that you're putting here because what has happened is people have started to think.
Tony Roth: Yeah.
Anshumali Shrivastava: And that actually changes a lot of things. And hopefully, uh, we'll delve into why it makes a big difference in, uh, what, what we should expect.
Tony Roth: Well, we may not get to sub linear deep learning algorithms, but we're going to get to 10 levels higher than that. Let's start with DeepSeek. When I think about AI, I think of three fundamental pillars that enable AI. One is computing power, which essentially are the GPUs that NVIDIA sells. Two are the algorithms. So, you know, what is the software or the programming that you, that you stick into those GPUs to do something?
And then third is data. The more data you have, the better empirical starting point you have when you apply the algorithm to figure out what you want to do or what the answer is or what the optimization is. So when you think about DeepSeek, what was it about DeepSeek when you think about those three pillars that is a game changer, if there is a game changer?
Anshumali Shrivastava: So, that is the most fundamental question and I think it's great that we are starting with that because it will set the pretext for a lot of things that we can talk about later.
So, you talk about three pillars, right? Algorithms, compute and data. Right. The way I view is you can among these two, right? So we teach computer science and in computer science 101, there's always a tension between compute and algorithms, right? So they are kind of complimentary, right? I mean, if you have less compute, you can do it with algorithms.
And if you have crappy algorithms, you can do it with compute. So they interact a lot in a complex way. Data, you can put a little bit on a orthogonal track because, for AI, whatever data you have, you have, right? I mean, obviously you can collect quality data, you can make it better, but that is slightly an independent pillar when compared to compute and algorithms.
Now, over the past few decades, let's go back to eighties when AI was developing. At that point of time, compute was what it was, but it was mostly about algorithms and people would believe that you will need very smart algorithms to make, you know, AI work, data started growing, all the things flipped around in ‘11 and ‘12 and in ‘11 and ‘12, we still remember like the ImageNet competition, right?
I mean, it was about, about like people could collect a very large amount of data and to process that data, there were GPUs that were available, which really changes the game because hardware provided a huge lift to something called deep learning and it started making sense. And since then, there has always been tensions about increasing hardware, increases the processing capabilities of your AI and processing capability means it will get better. Improving algorithms also improves the processing capabilities.
For example, what would require a hundred cycles can now be done in 50 cycles because you are smart about a lot of things that also will result in the gains. Over a lot of iterations, people came up with an algorithm. They used to call it back propagation, deep learning, whatever that is. And people believe that that is it.
And most of the next wave of hardware will come through hardware. You must have heard of Moore's Law, GPUs, hardware all going why AI took the boom. But over the last few years, what we have seen is that the hardware has kind of gotten stagnant. Every NVIDIA GPU also doesn't come out, like, they come out every 2, 3 years.
But algorithms, if you see, for example, OpenAI can reduce their cost by 10x every year. So algorithms have started coming into the play even more. And what you see with DeepSeek Is this big shift in what you can do with even smarter algorithms, right? So if you read some of the DeepSeek papers, there are things like mixture of experts and sparsity or DeepSeek did use some of these ideas that have been developed in the community. And they also used, you know, the better algorithms, maybe even good data to offset the hardware. Their goal was, okay, their goal was, okay, this is this capability that the whole world is saying is, is crazy hard and it will require hundreds of millions of dollars investment.
Tony Roth: So the answer is that the advancement or the breakthrough of DeepSeek was the algorithms, and because they were able to have a better algorithm, they are able to accomplish something similar to what others were accomplishing with less elegant or sophisticated algorithms, just grinding through using more computing power, but because they instead had a smarter algorithm, they could get to a similar result with much less advanced chips.
Anshumali Shrivastava: Yes, I will say that that is 80 percent there, but there is one subtlety that I would like to highlight.
Tony Roth: Please.
Anshumali Shrivastava: So remember, when let's say OpenAI brings an AI model and they create it. Creating the first model is very hard because you never know where is that going to come. But once you are there, computer science practitioners, especially the AI folks, they know that there is something called distillation.
Once you have an access to an AI and you want to create something that reaches there, it's much more cheaper. There are algorithms that if I am seeing something, if I know the capability that I have to reach, then I can do it much more efficiently. And on top of it, DeepSeek also uses this sparsity and other interesting ideas to reduce the compute and the investment.
Tony Roth: And so if you had to look at the advancement that was made, now that we've had the benefit of a couple of months and say that 50 percent was attributable to the fact that they were a follower and not a leader as it relates to the actual function and 50 percent was attributable to the fact that they were a leader, an intellectual leader in algorithm development, what would the breakdown be? Would it be 20/80?
Anshumali Shrivastava: Yeah, I work in the field of efficient algorithms and all I'll put it at 70/30, that 70 was the fact that it was following something.
Tony Roth: Okay, so definitely some advancements, but not so profound that it impugns the idea that if you are commercially in the business of selling chips that, hey, the world can operate with, even if we want to just sort of stay at the same level, you know, a fraction of the chips, right, we really need to continue to pump out these chips because there are other kinds of problems that need to get solved, same level of complexity, where you don't know where you're going. So you're still going to need to have the chips.
And then secondly, and in my mind, more importantly, when I think about what algorithms can do, what AI can do, there's all kinds of applications, but let's just, continue the example that we used, it's great to be able to type into AI on Amazon, which light bulbs should I purchase? And while that may be a commercial game changer for Amazon, it's not a changer for me in my life.
I need a much more robust architecture of computer chips, the next level of AI, right?
And the level after that, and the level after that, you know, they call it GAI, general artificial intelligence, which is the idea that you can go to a machine and ask it to do something that sort of replicates the appearance of having more generalized intelligence, that's going to require so much more than where we are today.
I mean, we're still in the first inning relative to where we need to be, whether we get there or not, I don't know, but where we need to be to really create something that appears to be a black box that can put out general AI.
Anshumali Shrivastava: Totally, Tony, you put it in a very right way that what we are observing is still, let's say A.I. in its very early phases. So what we can expect is that the capabilities of AI will improve and they will improve drastically over time. So companies like OpenAI will continue building GPT5, GPT6 which is better and better capabilities of AI. And with DeepSeek, what will happen is once a capability of AI is in the market, following it is going to be easy, which is what, again, not just DeepSeek, DeepSeek is an example, but we know that once an ability of AI comes in the whole enterprise ecosystem will jump on it and iterate, make it useful for their use cases.
And for them, the investment won't be at the level of open AI. For example, if I want to go to the next level, let's say, let's talk about some of the use cases. So let's say the light bulb example, right? If you're asking an AI, maybe the AI doesn't right now, it's an iteration two and it doesn't know a lot about your habits that you are probably like you really mess up something.
Tony Roth: Yeah.
Anshumali Shrivastava: You’re not smart enough to figure out. So they may still recommend you a light bulb and maybe the next level of AI will even be able to figure out. So you still have to wait for the iteration of AI. But even today, maybe out of 15. We can still only show you 10, right?
Tony Roth: Right.
Anshumali Shrivastava: And you will save on five. Tomorrow you'll probably even save on five more, right?
Tony Roth: Yes.
Anshumali Shrivastava: So it's going to be a very iterative process. And in the iterative process, people who are moving the frontier, they will still need to figure out a lot of compute, lot of chips. And every, every, you know, every trick in the game, but once they show a capability, which is the next level. To reach there is going to be incredibly fast and cheaper. That is what we can expect.
Tony Roth: We've all heard now this idea of Jevons paradox, which is as something becomes much more functional, rather than thinking that it's going to decrease the demand or cost for that type of deliverable, it actually is going to increase the demand because if it's something that actually, adds value to human civilization more people are going to want it. And by creating that access, it's going to become a bigger business. So, you know, computer might be an example, right? With Moore's law, you might think that, oh, well, it's a horrible business to be in. It's going to get cheaper, cheaper, cheaper. But in fact, it's sort of the opposite.
The more powerful computers get, the better business it is to be selling computers because people want it even more intensely because you can do more things with them. And that’s sort of what we’re seeing with AI. Does that resonate with you?
Anshumali Shrivastava: That is absolutely what is going to happen, right? I mean, I'm very much a believer that, uh, Jevons paradox will be directly in play in AI and chip space.
Because, look, I mean, the technology that we are seeing now is so phenominal that non experts will be able to accelerate to a capability that you could not have imagined before.
Tony Roth: Mmhmm.
Anshumali Shrivastava: And right now what we are seeing, estimate says there's only 1 percent of what we are imagining. What we are imagining is only 1 percent of what is possible.
We have not seen how this plays out and when it starts playing out, what we might realize is that right now we are cluttering our mind with like, there are these 10 things I have to do, but let's say AI now magically takes that away and you have 90 percent of the bandwidth to think about what else. So imagine if I'm running a customer service, processing some documents, and I was employing 10 people.
And now with AI, I can do away with only two.
Tony Roth: Right
Anshumali Shrivastava: But then what I will realize is, look, I still have a capability to manage 10 people. So what can I do with eight more? Maybe I'll open up a new service that I was not even thinking of because I have a lot of bandwidth to do. You know, a new industry will emerge, right?
I mean, there are a lot of industries that we are not even imagining will emerge once a lot of things get simplified. Effectively, this will end up just, you know, consuming everything and the demand will grow.
Tony Roth: Yeah, I mean, the applications are just so, so broad. If we're three years from now, and you're thinking about what's happened in the next three years, as opposed to where we are in six years, will the constraint be moving forward that will be limiting that develop the speed of development?
Will it be? The hardware, the algorithms, the availability of data, or will they all move in rough parallel with each other, sort of pushing each other to the next level? How does that work?
Anshumali Shrivastava: So I think, uh, definitely they all will work together, pushing each other, because look, the fundamental goal will be how we can take AI and democratize.
And democratization means when non experts having no idea about AI, I don't even know what is running behind it, like whether it's an algorithm or hardware, I have access to an API or a capability that I can massage it for my needs and improve productivity in my organization. There will be an ecosystem that will make that better and better and cheaper, because that's where everything will run on and there will be a lot of competition there.
But once that capability is there and people start using it, then it's, the evolution will be on what next. Right. So I would imagine all three has to come in.
Tony Roth: If you think about daily life and I think about what are the problems that we need to solve, I would sort of put them into two categories.
One is the, there are the problems that are really hard to solve. That the fact that we can't solve them are hurting, hurting us, right? I'll give you an example. That's really relevant right now, which is we can't solve what would be the optimal set of tariffs for the U.S. economy in order to maximize some particular objective function, whether it's economic activity, whether it is revenue or wages for a particular demographic, whatever it may be.
And, you know, the whole idea of reciprocal tariffs is extremely complicated, you know, the biggest reason that the market is selling off in my view is not because the tariffs are necessarily evil or bad. It's because nobody knows what they're going to look like. And the reason they don't know what they're going to look like is because it is such a multidimensional problem to figure out what set of tariffs would optimize our economy or certain elements of our economy that there's a real fear that these tariffs will get thrown around and cause a lot of harm before they create a lot of good because they're going to be used very inefficiently and it's going to create a lot of uncertainty.
So that's one bucket. And the other bucket is what are things that we can clearly solve, but you can have a computer do it instead of a person, to your earlier point, and to save some money, what are some of the things that go in either one of those buckets that you think in 5 or 10 years, we will have computers doing that today people have to try to do them?
Anshumali Shrivastava: Well, that's actually a great question. Everybody is thinking around that direction. But what is getting apparently clear is that all the non-blue collar jobs, right? For example, that requires a lot of somebody thinking, doing analytics. data processing. So let's even look at the case you mentioned about how will the tariffs pan out, right?
So imagine what people are thinking about. Let's say if you are investing, you are taking a bet into five scenarios, right? And then you're working out those five scenarios to figure out what is your probability of a particular world and that how you take a bet. Those kinds of things could be automated, right?
Because whatever you're doing, it's basically pulling out certain data, running some scenarios, and figuring out some numbers, right? I mean, so these are like non-blue-collar jobs. Again, what we would see, it's not going to be fully automated to be honest, because there are still going to be a lot of scenarios, right?
Somebody will bet on a scenario that somebody will act in a particular way.
Tony Roth: Right.
Anshumali Shrivastava: Right? And AI does not have a control over, in fact, nobody has control over how some people act.
Tony Roth: Right.
Anshumali Shrivastava: Right. So the AI cannot solve that problem. But, everybody will be taking bet and then they will be refining bet based on what happens.
But, the whole process, if you remember, there are a lot of moving piece in it. There are analysts sitting somewhere. There are people putting data. There are people, uh, analyzing market sentiment. Other things going on, a lot of those things put at an automated, right? So non-blue-collar jobs, people are imagining there will be a lot of assistance, right?
And what we need—should be thinking about. And I, I, as an educator, for example, we, we also keep thinking about this is I think a good example would be, so imagine a world where everybody has been learning themselves. So think about doing calculations because being able to calculate makes you good with accounting.
And suddenly you have been created the whole ecosystem where you graduate by learning certain things and you work at the top of chain if you're comfortable with certain things. Suddenly calculators come and what calculator is saying is wiping out an entire skill set, which was uh, very useful for the society, but then what is the next thing?
The next thing is always going to be whatever this technology is, is the base. Whatever AI can automate, it can automate. Now, what after that, as in the smartest will be, or the people who will be most valuable will be who can use whatever AI can automate…
Tony Roth: Right.
Anshumali Shrivastava: …to do something that AI cannot do. So maybe mind readers, who can read how some people will act will be more valuable than people who can calculate.
Tony Roth: Hmmm.
Anshumali Shrivastava: You see what I am saying.
Tony Roth: Yeah. Yeah. Because you think about what, what is happening now within the administration. Part of what's happening is, you know, we have an administration that wants to achieve certain things within the economy.
They want to bring back reshoring. They want to generate revenue by taxing imports because that activity is not as valuable to us as if it happens within the country, et cetera. And in order to do that, they have to put on these taxes or these tariffs, right? But they're also trying to guess how people are going to react to those tariffs, right?
There's a game theory going on here. If the president, you know, or the administration had some kind of computer that said, well, if you put in 18 percent in France and 15 percent on Germany on automobiles, you're going to get the same reaction because, or better way to say it, if you put 15 percent on Germany, that's as high as you can go without having them do something that would be detrimental to the U.S. But for France, you can go up to 20 percent on them, right? Or whatever it may be, that's the kind of thing that would be great.
Anshumali Shrivastava: No, absolutely. In fact, I would say that the example that you're saying could be scary because a lot of people don't trust AI and definitely people who has a lot of leverage should not be blindly using AI right away.
But you see this a lot these days in conversation, a lot of the C suites, right? So there are a lot of Gen AI technologies that are now, uh, you've been imagined that that that's going to work with C suites, analyze a whole bunch of these things and make decisions like running scenarios. And in that sense, you should be not focusing on like capability.
Let's say you have to imagine scenarios and if you really capture the right imaginations, you will run the right scenarios. But obviously if you are wrong in imagining what the scenarios are then AI cannot help you…
Tony Roth: Right, so…
Anshumali Shrivastava: …because you still have to input what those scenarios are.
Tony Roth: So that's one area where AI has application.
What about in the blue-collar arena? So I think of blue collar is ranging from a plumber, to a construction worker, to somebody on a manufacturing line. And so, in the first two, I can't even think of a possible world where AI could ever teach a robot how to hold a toilet plunger and plunge my toilet. That's one thing.
Another thing is, you know, to say nothing of, by the way, soldering in a thermostatic valve into a wall or something like that, right? Um, the delicacy with which that needs to be done. Then you have Construction workers, right? Building buildings and stuff. Clearly, you know, people need to do that work, but then when you get into manufacturing, I could see robots and stuff. More broadly in the blue collar space, how much efficiency or job a substitute could we get from AI? Do you think, or is most of the application really in the white-collar world?
Anshumali Shrivastava: Look in the blue-collar world, obviously, um, you know, like right now it all depends on what the next technological disruption is. There is a lot of investment going on in robotics.
People believe some part of that blue collar job can be automated, but it is not proven. Right. It's not proven like chatGPT yet. Now, when we look at what will happen with the current form of GenAI in the blue-collar world is they definitely will be easier to educate, right? There are a lot of videos, text content, right?
I mean, imagine I want to train myself to be a plumber, so I can fix leaks, right? You know, I can watch a lot of videos, do, you know, stuff that we are already doing on YouTube. So those things will become faster, but where there is actual manual skill, I think that is still at this point of time, technology is not there.
Tony Roth: So lots of application in the white collar world, more limited application in the blue collar world. And the reason this to me is so important right now is that when I think about what the President is trying to engineer, which is the reshoring of mainly manufacturing activity, while there's some application of AI, it's more limited, potentially.
So, AI is going to sort of essentially obsolete a lot of roles in the white-collar world…
Anshumali Shrivastava: Yeah.
Tony Roth: …and you would think all else being equal. Oh, well, we're going to need to have more jobs in the U.S. because these people are going to be obsoleted and that it's good that we're onshoring. But what you're doing is you're taking the overall society and you're essentially taking people that would have been in the white-collar space and you're moving them to the blue-collar space.
Um, and you're creating a bit more of arguably wealth concentration in the, in the smaller number of people that are able to stay in the top of the pyramid in the white-collar space. And I'm not sure that's good. So what are your, what's your reaction to that? Do you think I have it wrong?
Anshumali Shrivastava: Look, I mean, uh, it's hard to say how that will pan out, but yes, focusing more on like, look, the white-collar thing is not going to be fully automated.
Like there is a lot of repetitive nature and every white-collar skills, and that will definitely get automated and you will need a less and less work there. But to even orchestrate a lot of that, it will still need like at a top level, we’ll still need human to like do the, you know, checks and reliability.
The blue collars, obviously like right now AI cannot do much. But what is still a void is we don't know what really will happen if most of the white-collar jobs are like automated? Where will things go? For example, let me just give you something. Maybe AI policing will become an industry. You see what I'm saying?
Tony Roth: MmHmm.
Anshumali Shrivastava: Another way to think about it is, for example, content creation, right? So content creation now is becoming very, very automated, right? You can have AI bots write poetry better than people. They can do a lot of stuff. But imagine if all the content creation is being done by AI, then it's easy and you will get content a lot of time, a lot more, more than what we are getting.
Then maybe people will get bored.
Tony Roth: Right.
Anshumali Shrivastava: Then content moderation and creativity, when AI is available will be a new creativity. You see what I'm saying?
Tony Roth: Right. Yeah. If there's a certain set of roles today as AI absorbs a lot of those functions, you're going to have essentially a fewer number of, if you will, kind of super important directors and less players than you do today.
Anshumali Shrivastava: No, no. By the way, one thing that we should always take it for granted, AI is not going to displace the top 10 percent of the food chain.
Tony Roth: Right.
Anshumali Shrivastava: They are better than the average. Definitely better than the average. But they are not better than the top, top, uh, like, 10 percent performance.
Tony Roth: Right.
Anshumali Shrivastava: AI is not. Like, for example, even in the legal space, like paralegals and all, you don't need a degree to know, like, a lot of stuff. You can ask AI. It will do reasonably well. It even crosses the bar exam. But, you really need skilled and experienced lawyers for a very tricky situation.
Tony Roth: That's right.
Anshumali Shrivastava: Right? And that AI cannot take away, right, even right now because it requires another level of creating.
Tony Roth: So when you think about the overall economy, the level of productivity that I can bring to the overall economy, whether it be taking over the redundant tasks like summarizing research or writing the next incremental British crime drama that I've seen all of them and they all seem the same and the AI can write another thousand of them and scripts, right.
And I'll watch them all probably, or the tasks that we can't even solve today, but with AI, we may be able to solve and solve much more quickly. Like what's the, what's an optimal set of a starting point for reciprocal tariffs to try to optimize our, our environment, right? Which if I were, you know, in the administration, I would be asking smart people like you to develop those algorithms for me, right?
So how much can AI contribute to the increase in productivity across our economy? Are people that you interact with thinking about that question? And what, what, what kind of numbers are they thinking about?
Anshumali Shrivastava: Obviously everybody's thinking about around in this direction. I think what people imagine is that about 60 to 70 percent of the tasks are very repetitive.
So that is where everybody is happy if that is automated. But I don't think it's going to be an, in an equilibrium. It's not that all of that will be automated. It will be something like it is automated, but because there is an AI, you will need some policing, right?
Tony Roth: Right.
Anshumali Shrivastava: So you will create a new ecosystem of people about moderation, policing.
Plus we also have to worry about the dynamics, right? For we as a humans, whenever, like, as you mentioned, content creation. If everything is off the same script, I'm bored, but AI doesn't know anything because it will create the similar scripts. So now I am bored with AI, so the winner is not going to be an AI, because average is AI, right?
So where the market is about, like…
Tony Roth: Right.
Anshumali Shrivastava: So for example, entertainment and all, where average, by definition, don't make money. You have to be significantly above average to gain, gain traction. That is where you will need more skill set. So that is where more jobs will get added. But having said that, because let's say, for example, even for me, if 60 percent of my workload of, let's say, reviewing, doing a lot of paperwork and other stuff is taken care of, I have 60 percent more time to think about what new avenues we go and create
Tony Roth: Right.
Anshumali Shrivastava: …a new tariff optimization system.
Tony Roth: So let's say that in 10 years. One third of the white-collar space will have been automated, something like that, or is that,
Anshumali Shrivastava: Yes.
Tony Roth: Does that sound like too little or too, or too long?
Anshumali Shrivastava: 10 years is a long time, right? Especially with these technologies.
Tony Roth: Right, but in five years, a third of the white collar space will be automated.
Anshumali Shrivastava: Yeah, slightly less than that.
Tony Roth: If you look at the cost of running the economy, let's say it's two thirds white collar, one third blue collar, a third of two thirds is you're the professor two ninths , right? So let's say it's roughly 25%. So in five years, if that, if what I said was true, that would mean 5% productivity increases per year on an average annual basis for our entire economy, which would be just for AI, which would be totally unprecedented.
Anshumali Shrivastava: Probably like, I think those numbers doesn't sound unreasonable to me by the way. I mean, it's already happening, right? I mean, I routinely meet people who are saying that, look, now they are running the same thing that they were running with non experts in the loop. The training time is probably like less than a week,
Tony Roth: Right.
Anshumali Shrivastava: And they're quite happy. And in fact, there are less problems that used to be there before. And if that is happening, I would imagine then people are having more time to think about what next. But there are factors that I would still imagine, you know, when there is more of something, then there comes a point where you need to create an additional ecosystem, maybe some sort of a moderation or, because we haven't seen how this pans out at scale also, right.
Tony Roth: Yeah.
Anshumali Shrivastava: And it says, for example, self driving car, if you imagine there is not a lot of self driving car, but then there are almost all of them are self driving car. There will be a lot of transition phase in between.
Tony Roth: Well, self driving…
Anshumali Shrivastava: right?
Tony Roth: …car is interesting because Elon Musk can say, well, my self driving car is more reliable than a human being.
And I believe him when he says that, I believe that he can get to a point where that's true. I don't necessarily believe him today because I've been in a Tesla that has full self driving
Anshumali Shrivastava: Yeah.
Tony Roth: …and it has not been as reassuring as I would have expected. But, the Tesla needs to be a hundred times more accurate than a human for anybody to actually accept it.
Anshumali Shrivastava: Exactly. You know what Elon missed? Yes, you're right. Tesla is better than an average driver. Ask how many drivers feel they are worse than average.
Tony Roth: Yeah, that's right. Right?
Anshumali Shrivastava: 90 percent of drivers will say I am a top 10 percent driver.
Tony Roth: You know, when Tesla launches a vehicle that doesn't have a steering wheel, if I'm going to go into San Francisco to a street that the Waymo has gone up and down a hundred thousand times, literally, and then knows how to go up and down that one street, I’ll get into that car. But, would I get into a car without a steering wheel and go from my office to my house? Not, not a chance anytime soon that that's going to happen.
Anshumali Shrivastava: There are pros and cons. I do see one side of the coin in which if you have used the technology enough, right? I mean, people transition, initially people are generally skeptic, but let's say if you're driving on Tesla one day, two day, five day, 10 day, and then nothing happens, you tend to get used to it, right?
Tony Roth: Yeah.
Anshumali Shrivastava: …but the problem comes in again at this point, let's say there are accidents and there are liabilities. The problem is everybody is special in their minds.
Tony Roth: Yeah.
Anshumali Shrivastava: As I said, I do believe I am much better than an average driver, which may not be true. Because how can 90 percent believe people believe that they are better than average driver?
Of course, 40 50 percent of them has to be wrong.
Tony Roth: Right.
Anshumali Shrivastava: So you're right, just being better than average is not going to cut it.
Tony Roth: Okay. Let me ask you about one last topic, Anshul. I am particularly excited about the potential for in medicine and I tend to think most about pharma and compounds and drugs, et cetera, because, you know, there's more than we could ever test in the history of human civilization.
So we've got to figure out which ones could really help us. It seems that the application to develop more effective medicines must be almost unlimited for AI if you have the right data and you have the right algorithms. And that should sort of be like an asymptotic chart where you start off slowly and then eventually you get to the point where you know, you can almost cure anything and you live 150 years at least.
Where are we on that curve?
Anshumali Shrivastava: It's really hard to predict because it's not just about AI working. It's also about the reality where something like that should exist. Right? Obviously the belief right now is what he exactly said, that we do a lot of simulation testing time, like the drug discovery and everything takes a very long time.
And it's very similar to, as I mentioned before, like tariff simulation. Right now, if you look at what, what is going on in the healthcare world, like even a Nobel prize to alpha four, right? People are discovering proteins and structures faster than they ever could because of AI. So they are running these simulation behind the scenes as we speak, and they figure out faster than ever what will work and what did not.
So it is definitely on an accelerated phase. But it is also dependent on how many interesting and magical drugs exist. That we don't know. We know we'll exhaust them faster. So we'll know whether they exist or not, but that will definitely expedite the whole process. And maybe at some point of time, we realize Uh, there is not much here.
Maybe we should look at somewhere else that maybe some running technology. But yes, it will definitely accelerate. Really hard to predict what exactly will that look like. Whether we will live 150 years or not..
Tony Roth: It'll be our children, hopefully, but not, you know.
Anshumali Shrivastava: Sure.
Tony Roth: All right. Well, I'm sure this has been such an interesting conversation.
I want to give you the last word. I'm going to summarize our conversation by saying that the potential for AI really is by the average person just being understood. I mean, for me, I'm at the how many CIOs does it take to change the light bulb stage. But for you, you're you can see so much further into it. It's just so exciting to have a conversation like this.
And as investors, we need to be very vigilant around how AI is going to impact the economy, not just thinking about AI as a source of computer chips to buy NVIDIA or other companies, but we really need to be thinking about how it's going to change the whole economic system. And that's what we are trying to do at Wilmington Trust.
What is the most exciting single thing for you in the next 5 years that AI will, will, will do in, in our lives that you don't see today?
Anshumali Shrivastava: That's an interesting question. The most exciting thing, there are so many exciting things going around. I think even AI in education, if you think about it, right? I mean, right now, what I think is that a lot of people, they have no access to, let's say, not great teachers, not great education.
And I think one thing that might change is the access to education and capability to know, right. I mean, even access to teachers, then that's a huge service to the whole mankind.
Tony Roth: I really haven't thought about the application of AI that much in learning.
And, um, ironically, because so much of it is machine learning, but human learning, the ability of AI machine learning to help humans learn is fascinating. Um, so, um, for our next conversation, maybe we'll focus on that.
Anshumali Shrivastava: Just one last thing I like to add is AI is also going to be an equalizer because we have buried the whole fabric of society based on this is impactful.
This is more impactful. This is superior. This is there.
Tony Roth: Mmhmm.
Anshumali Shrivastava: Once the foundation of this could be shattered with AI, then a new, maybe more equal system could emerge, right? So this would also mitigates one of the fundamental problems of society, right? Who could be a winner when, see when the rules of the games are rewritten?
Yeah. It's the most fair game.
Tony Roth: That's right. Yeah, I think you can really democratize the access to it, right?
Anshumali Shrivastava: Yup
So thank you so much. Um, this has been Capital Considerations and for the rest of our Thought Leadership, please go to wilmingtontrust.com where you'll find all of our latest economic and planning information.
DISCLOSURE:
This podcast is for educational purposes only and is not intended as an offer or solicitation for the sale of any financial product. Or service or recommendation or determination that any investment strategy is suitable for a specific investor. Investors should seek financial advice regarding the suitability of any investment strategy based on the investor's objectives, financial situation, and particular needs.
The information on Wilmington Trust's capital considerations with Tony Roth has been obtained from sources believed to be reliable, but its accuracy and completeness are not guaranteed. The opinions, estimates, and projections constitute the judgment of Wilmington Trust as of the date of this podcast and are subject to change without notice.
The opinions of any guests on the Capital Considerations podcast who are not employed by Wilmington Trust or M& T Bank are their own and do not necessarily represent those of M& T Bank Corporate or any of its affiliates. Wilmington Trust is not authorized to and does not provide legal or tax advice.
Our advice and recommendations provided to you is illustrative only and subject to the opinions and advice of your own attorney, tax advisor, or other professional advisor. Diversification does not ensure a profit or guarantee against a loss. There is no assurance that any investment strategy will be successful.
Past performance cannot guarantee future results. Investing involves a risk, and you may incur a profit or a loss. Any reference to company names mentioned in the podcast should not be constructed as investment advice or investment recommendations of those companies. Third party trademarks and brands are the property of their respective owners.
Third parties referenced herein are independent companies and are not affiliated with M& T Bank or Wilmington Trust. Listing them does not suggest a recommendation or endorsement by Wilmington Trust. Private market investments are only available to investors that meet the U. S. Securities and Exchange Commission's definition of qualified purchaser and accredited investor.
Facts and views presented in this report have not been reviewed by and may not reflect information known to professionals in other business areas of Wilmington Trust or M& T Bank and may provide or seek to provide financial services to entities referred to in this report. M& T Bank and Wilmington Trust have established information barriers between their various business groups.
As a result, M& T Bank and Wilmington Trust do not disclose certain client relationships or compensation received from such entities in their reports. Investment products are not insured by the FDIC or any other governmental agency, are not deposits of or other obligations of or guaranteed by Wilmington Trust, M& T Bank, or any other bank or entity, and are subject to risks including a possible loss of the principal amount invested.
Wilmington Trust is a registered service mark used in connection with various fiduciary and non fiduciary services offered by certain subsidiaries of M& T Bank Corporation, including, but not limited to, Manufacturers and Traders Trust Company, M& T Bank, Wilmington Trust Company, WTC, operating in Delaware only, Wilmington Trust N.
A., WTNA, Wilmington Trust Investment Advisors, Inc., WTIA. Wilmington Funds Management Corporation, WFMC, Wilmington Trust Asset Management LLC, WTAM, and Wilmington Trust Investment Management LLC, WTIM. Such services include trustee, custodial, agency, investment management, and other services. International corporate and institutional services are offered through M& T Bank Corporation's international subsidiaries.
Loans, credit cards, retail and business deposits, and other business and personal banking services and products are offered by M& T Bank, member FDIC.
Copyright 2025 M& T Bank and its affiliates and subsidiaries. All rights reserved.
Equal Housing Lender. Bank NMLS #381076. Member FDIC.
