Fans of the long-running game show Jeopardy will remember that one famous contestant stood out from the rest. IBM’s supercomputer, named Watson, beat two human opponents in a 2011 competition that brought artificial intelligence into the spotlight. More than a decade later, AI has taken a giant leap forward. Generative AI, made popular by the launch of ChatGPT, has already made its impact felt on the investment world, and it appears poised to shake up the modern economy as businesses find new ways of utilizing the technology. The next generation of AI presents new possibilities for the way people live, work and invest.

This episode of PGIM’s The OUTThinking Investor explores a future reimagined by generative AI, from its business applications and social impact to the investment opportunities and risks across sectors. Guests David Ferrucci, the former project lead for Watson; Gillian Tett, US editor-at-large for the Financial Times; and Erika Klauer, portfolio manager of the technology fund at Jennison Associates, bring their unique perspectives to the topic of AI and how investors can prepare for what is to come.

Subscribe & Listen: Apple - Spotify - Google - TuneIn

>> In February of 2011, an unusual audience was watching Jeopardy's "Tournament of Champions." The game show is normally recorded at Sony Pictures Studios in Culver City, California. This time, the location was IBM Research Labs in Yorktown Heights, New York. But the most unusual thing was that the winner of this Tournament of Champions wasn't human. It was IBM's supercomputer named Watson. Sitting in the audience was Dave Ferrucci, the project lead for Watson, along with his team members. Ferrucci's goal was to develop a machine learning system that could understand language in a deeper way. Competing against the two top Jeopardy champions would put Watson to the test in a monumental and very public way. For each question, Watson would need to analyze the general knowledge provided in the clue, figure out what was being asked, search for the possibilities, score the possibilities, and then choose one, not to mention buzzing in ahead of its human opponents. Watson made a few errors along the way, including getting Toronto for a clue under the category of U.S. cities, but eventually won the $1,000,000 prize, which was donated to charity. 12 years later, Watson's achievement is almost quaint against the backdrop of recent advances in artificial intelligence. But it was never just about the Jeopardy title. The end goal was something much bigger, to use advances in deep analytics in ways that would help solve some of the world's most important problems. To understand today's investment landscape, it's important to know how we got here. This is "The OutThinking Investor," a podcast from PGIM that examines the past, the present-day opportunities, and the future possibilities across global capital markets. In this episode, three experts look at the impact of artificial intelligence and particularly generative AI. Dave Ferrucci was Watson's project lead and is currently the founder and CEO of Elemental Cognition. Gillian Tett is editor at large in the US for the "Financial Times" and author of the book "Anthro-Vision: a New Way to See in Business and Life." Erika Klauer is managing director and portfolio manager of the Technology Fund at Jennison Associates, an affiliate of PGIM. The buzz around AI has risen to an almost deafening level. It might seem like a recent innovation, but Erica Klauer and the Jennison team have been focusing on AI for most of the past decade.

>> Generative AI is a new way of processing data. The basic way to think about it is that one can make a giant pile of information -- it's actually called the pile -- and feed that pile with everything from the Bible to periodicals to newspaper articles, and then train a model to understand all of that input. Once the model has been trained, you can generate new content with that data, whether it be a song, a drug development, or even a jingle for an ad campaign.

>> We've come a long way with the current generation of generative AI since Watson's Jeopardy win in 2011. Dave Ferrucci witnessed this evolution up close.

>> This new type of generative AI built on these large language models builds these giant statistical distributions of words, so it will look at that context, and it will just generate and say here's the next most likely words having analyzed lots of words and looking at how they occur together, and that's super powerful.

>> The swell of excitement in AI is matched by the degree of uncertainty in its future and the potential consequences. Gillian Tett takes an anthropological perspective.

>> What I argued in my book, "Anthro-Vision," is that all human activity, whether it takes place in the Amazon Jungle, or an Amazon warehouse, or on the Amazon Corporate Board, is fundamentally about social networks and rituals and tribal groups, and cognitive maps about how the world works. And we can always benefit by looking at the social and cultural context and the patterns that drive humans. What you have with AI is very similar to what you had with the financial crisis, or in nuclear science, or in areas of medicine, which is a small group of highly skilled professionals who are in command of a technology which nobody else understands but which has the potential to impact us all. There is a burning need for all of us today to not just get to grips with what AI means, but also to understand the cultural and social context to how it's being developed, how it's being programmed, who is doing that, and how it's being applied in our lives and frankly, how it's going to change all of our lives going forward.

>> Expectations for generative AI are sky high in part because of how today's large language models are able to produce output that looks like the result of smart human thinking.

>> Because large language models are so powerful and how they can take what's written, break it down into the various words and their context and how words tend to be assembled together, and they discern those patterns, that they can, you know, mimic those patterns and actually produce often good, coherent responses. But they're not trained to be correct. They're not trained to be factual. They're generative. They will generate information that is consistent with the prompt, that's consistent with the structure and form of the language. Watson was trained to be correct. There's an important difference there. Humans are very affected by rhetoric, by the sound and power of the language, by how the language is chosen to be aligned with how I think. And what's really powerful about the modern large language models is that they can be easily influenced. In other words, you could prompt them in such a way to say, write this response for a second grader, or write this response for someone who has this political view, write this response from somebody who comes from this culture and make sure it says this, that, or that. It selects the words appropriately. And when I read that, because it's coherent and it's language that I'm used to hearing and it's fit for my culture or fit for my politics or fit whatever, I'm just more likely to believe it.

>> The risks associated with AI could be vast. In fact, the Center for AI Safety recently published this brief but powerful statement, "Mitigating the risk of extinction from AI should be a global priority alongside other societal scale risks such as pandemics and nuclear war." The stakes couldn't be much higher. Gillian Tett also highlights potential concentration risk, similar to what we saw during the global financial crisis.

>> So you had a tremendous reconcentration problem in the financial system, and something like that's happening again with AI because the number of companies developing AI today in the West is very, very small. So if one of them is essentially putting at best, say, a bug into the system, or something malicious, or something biased, that has a potential to have very far-reaching impact, which no one will understand until it's too late because of the opacity. You also have a problem that there's tremendous concentration in terms of the use of the cloud in the financial system as well, whereby a very small number of companies are essentially acting as cloud service providers. And so once again, when you overlay that with the AI risks, you're getting even more concentration. And just to add to the problems, the regulators for the most part, are not set up at all to understand these dangers because, historically, you've had a very marked silo issue in terms of the regulatory system and the financial industry. To put it crudely, the people who are paid to handle money and regulate money have sat in a very different part of the ecosystem from the people who were developing cutting edge computer tools or AI-type platforms or systems.

>> And of course, there is also a risk that AI does not live up to the current hype.

>> Like all technologies, there is a huge amount of hype, and there's essentially a bubble going on. And we've seen endless bubbles before, whether it's the Internet bubble, the financial bubble, or any other bubble. The thing we know about bubbles is that they can explode dramatically in size and then go pop, although sometimes they go hiss as the air comes out. But when they go pop, they always leave an infrastructure behind them, which is valuable and worth using, whether that's the 2008 financial crisis or the 2001 dot com bubble. In this case, what is clear is you have both evolution and revolution going on. You have evolution because much of what is happening with AI is not that new in that it's a culmination of many years of research, but the thing that is potentially revolutionary this time round is the fact that so much have been released open source in a way that anyone can access and potentially deploy within their own operations.

>> Dave Ferrucci sees significant near-term and longer-term opportunities, particularly with the creation of new and even more powerful large language models.

>> I think where they'll have impact is in customer experience, sales support, marketing, all the creative arts, including gaming and entertainment, certainly in search and knowledge management. And that's across industry from retail to law to healthcare, the ability to take large bodies of content, documents, images and being able to kind of quickly navigate and find and synthesize and summarize what's there, which improves the access and application that we have of that knowledge. I think in the longer term, the underlying machine learning techniques that are used in generative AI, these attention-based transformers are what they're called, but this technique is, I think, very powerful beyond its application to just language. So in drug development and areas related to understanding how molecules interact, how proteins can fold, there's sort of big advances coming in the application of this type of machine learning in there, and our ability to program nearly everything and anything by simply talking to it. So as devices have more and more compute power in there and smarts and can be reactive, from robots that are cleaning your floor to robots that are gardening for you, these are control systems that can start to be trained and tuned very, very quickly by showing them some examples and explaining how things work to them. And then all of a sudden, they're operating that way, and the implications, I think, of that are significant. It'll accelerate general robotic solutions of all sorts, from transportation, manufacture, retail, home and healthcare. So there's lots of things beyond just sort of using language like the ChatGPT stuff you see. There's much broader implications of the underlying AI.

>> The opportunities for investors are only beginning to come to light.

>> If we look back at technology as an industry, I would say it has two profound impacts on the economy. One is productivity and one is cost reduction, and those two become very pronounced with artificial intelligence and its flavor of generative AI because instead of having to use and access lots of old periodicals, that data is going to be at one's fingertips and be able to sync itself, analyze itself, and produce itself in a new format in seconds. So the productivity will be there, and then, of course, the cost reductions will be down. The need, for example, to hire someone to write a song or write a marketing campaign, the need to go to low-cost places for data entry, for example, is going to be eliminated with generative AI if those piles of data are trained to supply that information for whatever the ultimate utility may be. One is the overall data center business; we think that that market is going to be penetrated at a much faster pace than what we're seeing on the street amongst our competitors and amongst sell-side analysts. We think that the pace of adoption is probably about double what we're seeing out there, and that's where we see an opportunity for the companies that we've invested in to meaningfully outperform relative to their peers.

>> If much of the expected cost reductions come on the back of lowering labor costs, how might that affect productivity and the global economy?

>> There's good news and bad news about what robots are doing to productivity. The good news is that they can indeed replace humans. And in a world where we have aging demographics in many countries, that could be transformational because most of the economists who are projecting growth and productivity into the future are working with previous decades of data about the relationship between population size and growth going forward. And so Japan is a classic example of this, where you've got a population aging very rapidly, which implies that growth is going to collapse. Except if robots can start to do some things that humans are doing, that equation may start to change, and you can see that already in Japan in that you're getting AI-enabled machines helping to monitor patients in hospitals and essentially providing a lot of high-touch activity that is often quite effective as far as a patients are concerned. And Japan is one of the few countries in the world where trade unions and ordinary people say they love the idea of automation and robots because they know they're short of workers and they need help. But the problem, of course, is that in other countries which aren't quite as desperate as Japan, essentially robots are seen as very threatening, and the risk is that they end up eroding not just a number of blue-collar manual labor, but also middle-class jobs as well.

>> What might that next generation of generative AI look like, and when might it arrive?

>> I think that the next generation of generative AI relates to, most likely, the migration of robotics and artificial intelligence and the union between those two. Those two combined really make for quite a huge stair step, Big Bang effect in terms of the utility, whether it's in all the different factories that make things on a global basis, all the different vehicles, not just cars that people are driving but delivery vehicles. To me, that is the next big thing on the horizon that one can see that marrying of robotics in the physical world to unite that with the artificial intelligence world. That's an incredibly interesting and powerful tool that we have very much our eye on.

>> And with great power comes great responsibility. The challenge for policymakers will be in determining how to regulate AI without stifling it.

>> How we shape laws to address AI, the impact of AI, is critical to be doing. I'm happy to see a lot of governments saying, like, this is something we need to wrap our heads around quickly. Establishing the broad areas of concerns, the role government should play, I think, is important. I think it might be a bit early to be coming up with very specific restrictions. We have a lot to learn. We have a lot to experiment with. We have to keep a watchful eye on that and, you know, adapt and react quickly, but coming in there and anticipating too much and putting too much restrictions, I think, could slow down innovation. So I think we have to be careful. Social scores, revenue scores, the impact they have on hiring, facial recognition, I mean, these are all things that can create undue bias. That's a big category we should be thinking more deeply about, coming out with broad statements I think important, the more specific they got, we'll see. Identifying risk categories, I think, is a good idea. Getting your dream interpretation wrong is a lot less risky than getting an individual medical diagnosis wrong. So where are we going to focus? What sorts of risks and applications are more important than others? This same AI may be used in preventing crimes or enabling criminal activity. Transparency is key to all of this. and I like that in the EU legislation, they're trying to pay attention to that. I still think tricky, but it's, to me, transparency. Any AI should be able to reveal its underlying decision logic in a way that humans can understand and adjudicate. And it should be able to reveal its sources, like there must be some path to accountability.

>> Investors will need to be aware of the regulatory and competitive environment as they navigate investment opportunities at the industry level, even as the excitement around AI seems unending.

>> Certainly from Jennison's perspective, we have seen the portfolios rise at a very, very fast pace so far this year, as we have been overweighting these trends for quite some time and we increased those weightings towards the end of 2022. So we would expect short term that there should be and always are consolidation periods. There are moments of doubt that come about, whether it be restrictions with international markets, particularly tensions with China. Ultimately, those have proved to be buying opportunities when we are so early on in a cycle, and it is our opinion that this cycle of widespread generative AI use is just beginning and will accelerate into '24, accelerate into '25. So while there could be some pause, corrections, in some of the names that we own at Jenison, we still see substantial upside in the shares of the winners, those that are really disrupting their industries by understanding and enabling or using generative AI to their advantage. We have a holistic approach to the opportunities, where we are looking at the underlying hardware that enables generative AI, the chip makers, whether they be the actual GPU makers, the networking processor providers, the memory providers, those companies to us are very, very compelling. We're also interested in automotive technologies, autonomous driving, which really goes hand-in-hand with artificial intelligence and generative AI, because you need to be able to not only analyze every road that has ever been driven, but generative AI is also helpful because you need to generate conditions that cannot be analyzed. That actually, to me, is the biggest question, or thing, that AI has not been able to conquer yet, which is this concept of causation. If there are new variables that are entered into the mix, the drawback of generative AI is that you have to train with a given set of data.

>> Software also continues to be a big area of investment interest with opportunities potentially growing along with demands from households to the corporate enterprise level.

>> We are very heavily invested in software companies, both big ones and then emerging ones that are really looking to optimize the efficacy of big data sets as well. So I would say our approach has been one of embracing both hardware and software. In the software area, we have two buckets of opportunity. The first are the software companies that have extremely large amounts of data that relating to how to deal with customers and or suppliers. Those are the big companies that we all know. And then there are smaller companies that are particularly interesting because they're smaller, they're more nimble, that are related to things like combining different data sets within the enterprise to come up with better understandings of how a business ought to be run. But then there's also, I think, a very interesting aspect of all of this, which is security and using AI to be able to target and instantaneously recognize a varying pattern of behavior that could be some kind of malware. That is going to also be a very, very good business, too. And then the third piece that I think is really important to look at is the value of those big piles. Who has the information that ultimately will make those piles so effective? And those are the companies that have lots and lots of data that is very valuable. It could be a healthcare company that has a lot of genetic information. It could be a consumer company that has an enormous amount of data about what consumers want and like and don't like. Those companies with the access to that amount of data are also going to be very, very valuable over time.

>> Then there's the question of whether AI could make us more efficient investors or less.

>> I think that generative AI will make us more efficient investors because one can get context of new industries very, very quickly if you can overlay onto your analysis of a retail company and understand some of the underpinnings that might be driving the overall sales in a much more effective manner that takes a second rather than a minute, two minutes, three minutes to be able to look up that information. And that data source is so much greater than any opinion that you might find. That, to me, is an incredible tool of efficiency. If I were able to slice and dice my questions to understand where there are problems that urgently need to be solved, then that probably will lead me in the direction of the company that is best positioned to solve those problems.

>> The applications for AI seem endless. If recent history is any indication, AI will continue to shape the world in dramatic and sometimes unpredictable ways.

>> The Chinese have a great phrase, "a fish can't see water." None of us can see the assumptions that drive us unless we jump out of our fish bowl, go swim with other fish, and look back. And so what's desperately needed today is that the people who are driving AI development and research need to jump out of their fish bowls, go swim with other fish, and look back to see the risks, and fish who live in other fish bowls need to make an effort to understand what's going on because we're all in this together.

[ Music ]

>> Thanks to our experts Dave Ferrucci, Gillian Tett and Erica Clower for their insights on artificial intelligence. And if you missed the previous episode, listen to "The New Oil, a Geopolitical Battle for Chip Dominance" to hear about the economic and geopolitical impacts of the microchip industry, "The OutThinking Investor" is a podcast from PGIM. Follow, subscribe, and if you like what you hear, go ahead and give us a review.