ChatGPT and AI Investing

Data is everywhere, but is it ready for consumption? Morningstar developed the language of global investment data so you have the right ingredients to help you shine. Morningstar, where data speaks.

So am I reading this?

Yeah, sure. Since we did the AI episode a year ago, I wanted to have it write a script to see if it's gotten any better.

So this is, so this is AI that wrote this. You didn't write that.

No, I didn't. I didn't write this intro. It's ChatGPT. It's that AI chatbot that blew up on Twitter last week after they released it to the public.

I'm hearing a lot about it.

Yeah, kids are writing their midterm essays with it. It's pretty incredible.

And it's suddenly a lot smarter than when we did this episode.

Yeah, well, if you remember last year, we had AI write the beginning of the podcast episode and it was It kind of didn't make any sense at all.

Let's see. Okay, I'm going to read my AI-written script. Hey Jackson, my favorite audio engineer. How's it going?

Hey Jack, I'm doing well. I'm just trying to keep up with all your ramblings.

Ha!

You know me, always talking about the latest and greatest. But seriously, we have a great episode for our listeners today. It's a rerun of one of our classic episodes. On AI. We're talking about ourselves like it's I Love Lucy.

Go ahead.

I remember that episode. It was a real brain buster.

Yeah, I think I lost a few IQ points just trying to wrap my head around it all. But hey, at least we have you, the resident AI expert, to guide us through it. The AI is making me sound like you're really smart, which is, you know, it's true. But I'm just saying, I don't know.

I didn't—

I didn't tell it to—

you didn't tip the scales at all. Okay.

Maybe a little more.

You didn't put in Jackson equals smart, Jack equals dumb.

I can try that for the next one.

That's it for the read of the AI-generated script. I've got to say, it's not that bad, right?

I mean, it's only going off of one episode.

There have to be some people out there thinking right now that's better than our usual.

Oh no. No, it's worse than our usual. Come on, maybe next year.

All right, but this thing is close. By next year, I'm handing this job over to the AI.

You don't mean that, do you?

Let's play the tape. Listening in is our audio producer Jackson. Hi, Jackson.

Hi, Jack.

So you volunteered to use artificial intelligence to generate a few lines of speech to get this episode started. How'd you do that?

Yeah, so I fed some dialogue from past episodes into an AI system, told it the topic I wanted it to talk about, and it did the rest.

How do I know I'm not already talking to the program? Tell me something that only Jackson would say.

My favorite song is "Obsessed" by Mariah Carey.

Well, that checks out. Will this program sound like my voice?

Yeah, so I didn't have enough samples to make an AI version of your voice, but I was able to make one of my voice. For your voice, I just used the action movie commercial guy.

Okay, let's hear it.

Welcome to the Baron Streetwise Podcast. I'm Jack Ho. The voice you just heard is Eric Schmidt. Listening in is our audio producer Jackson. Hi Jackson.

Good morning, friend. This is the guy on the show. Do you have a dog or maybe a cat? Why don't we talk about food?

Cool. Now, we don't have a lot of time. We've talked about nuclear war and totalitarian governments, but did you see what AI can do this week? This could be giving us robot cops, or robot doctors, or giving AI the ability to drive cars, or giving it the keys to the kingdom.

I gotta tell you, it doesn't make a ton of sense, but what I'm wondering is, if we did a whole episode like that, like, it wouldn't be the worst podcast out there, right?

I'd say second quartile.

All right, well, that was great work for your first attempt.

Thank you, friend.

Okay, so artificial intelligence enables machines to act like humans, and it's not particularly new. The term comes from a proposal that a small group studied the matter during two months at Dartmouth in the summer of 1956. In the 1960s, AI attracted a big tech investor: the U.S. military and its R&D unit called Defense Advanced Research Projects Agency, or DARPA. DARPA has provided funding for all sorts of technologies that went on to see widespread civilian use, like GPS navigation and the internet. Back in 2013, DARPA provided a company called Moderna funds to develop a new way to rapidly manufacture vaccines, which could prove useful in a biological attack or, as it turns out, a coronavirus pandemic. Now, AI has been particularly slow to develop, with periodic booms in research resulting in few real-world applications and leading to long dry spells for funding. But over the past decade, the field has boomed. UBS calls AI one of the top opportunities for investors over the coming decade. There are 3 broad reasons this is not another false start for AI. First, as our lives become more digital, we're creating vast amounts of data, and that data is valuable, but there's too much of it to make sense of with traditional computing. By 2030, there will be a projected 660 zettabytes of data worldwide. Zettabytes, Jackson, or zettabytes?

I'm feeling Team Zetta.

Of data or data?

Data.

Zetta of data. Got it. So if you want to know how much that is, picture an iPhone with 128 gigabytes of storage, all of it full. By 2030, we'll have the equivalent of 610 full iPhones for each man, woman, and child on Earth. The second reason AI is taking off is that computing power is rising to the occasion. And I'm not talking about the central processors that have traditionally powered our personal computers. I'm talking about specialized chips that grew out of video gaming and can handle the highly parallel processing needed for AI. NVIDIA is the dominant player there. And I'm also talking about application-specific chips that big tech companies with massive data centers can build on their own. Sometimes with help from companies like Broadcom and Marvel Technology. And there are other types of designs. The third reason AI is growing so quickly is that it no longer requires government funding because so many companies are making real money from it today. When you open your Netflix queue, AI helps decide your recommendations. That you probably already knew. But it also helps predict whether Netflix should adjust your data rate on the fly to keep your picture smooth, or begin caching the next episode in a series. Maybe you've noticed that the thumbnail image for a movie has changed. Netflix has multiple images for many of its movies, and AI helps predict whether it should show you the one implying romance, or suspense, or an actor you've watched in the past. AI even helps Netflix turn viewing habits into creative decisions when it comes to making new shows. That's just the start, and it's just Netflix. AI delivers your Google search results, helps Siri and Alexa understand what you're saying, and directs self-driving Teslas. It's what allows me to open my photo collection and search for photos with dogs or cars in them even though I've never labeled any of them. Sometimes it makes little family slideshows with music for me without me asking. One the other day was called "At the Beach Over the Years." There are subsets of AI with names you might have heard. Machine learning refers to teaching algorithms using lots of labeled data so that the algorithms can then look at new unlabeled data and make educated guesses or inferences about it. And get better as they go along. Then there's a subset of machine learning called deep learning, which uses computing networks inspired by human brains called artificial neural networks. Those can learn things from data that people haven't taught them. Let me give you an example. There's a 3,000-year-old Chinese board game called Go. And it's enormously complicated, much more so than chess. There are more possible board configurations in Go than there are atoms in the known universe. Alphabet created a Go-playing program called AlphaGo and taught it to play using games with humans, and then had it play over and over again against itself and learn from its mistakes. If you've heard of AlphaGo, you might recall that in 2017 it beat the best player in the world. But did you know that shortly after that, Alphabet announced a new program called AlphaGo Zero? And this time there was no training with human games. The machine was told the rules and that's it. It had to learn to play by itself, starting with completely random moves. It quickly developed its own creative style of play with strategies that hadn't been seen before, and it surpassed all other players. The game, as Alphabet puts it, is no longer constrained by the limits of human knowledge. That's astonishing to me, but does it also sound at all eerie? I read an Oppenheimer report on AI that offered some different ways to categorize it, and the simplest way was by capability. It listed 3 types of AI: weak, strong, and super. Weak AI, according to this classification system, is also called narrow AI, and it consists of everything I just mentioned, plus almost everything else we currently have. These are all things that resemble human intelligence but are designed only for specific tasks. Strong AI, also called general AI, is a subject of ongoing research, but it doesn't yet exist. It resembles human intelligence and can perform many varied tasks. The example UBS gave was Jarvis, the computer that helps Tony Stark in the Iron Man movies. As you wish, sir. I've also prepared a safety briefing for you to entirely ignore. Which I will. Now, super AI is just called super AI, and it is not a work in progress. It refers to a point where machines become sentient with decision-making capabilities that exceed those of humans. This is the type of AI that has been portrayed artfully over the years in movies. There's one that came out in 1968 around the peak of the first AI research wave. It's called 2001: A Space Odyssey, and it features a computer named HAL.

"I'm sorry, Dave.

I'm afraid I can't do that." And of course, you knew I had to mention Terminator, which came out near the peak of the second AI research wave in 1984. It features a self-aware defense system that turns against humans called Skynet. Skynet, by the way, is also reportedly the name of a real system used by America's National Security Agency overseas to try to identify potential terrorists using cellular network metadata and machine learning. I don't know why they called it Skynet, and there's been disagreement over the system's accuracy. I'll mention one more movie, and this one came out in 2014 during the current AI era. It's called Ex Machina. The head of a dominant search company invites an employee named Caleb to stay at his home and meet a beautiful robot he's created named Ava and test whether he forgets that Ava isn't human.

Are you attracted to me? What? Are you attracted to me?

You give me indications that you are.

I do?

Yes. Poor Caleb can't tell whether he and Ava are falling in love or whether she's manipulating him to try to escape. At one point he gets so turned around he wonders if he's a robot too. I think he would have stood a better chance with Terminator. To learn more about AI, I recently had the chance to talk with someone who is the real-life head of a dominant search engine. Eric Schmidt was recruited by Google founders Larry Page and Sergey Brin to run their company, which he did for 10 years before returning control in 2011. Eric was paid largely in shares, and they have of course performed exceptionally well over the years, putting his wealth at an estimated tens of billions of dollars, which is unusual for anyone, but especially for a non-founder. In recent years, Eric has been an investor and philanthropist and advisor to the Department of Defense, especially on the subject of what could go wrong with AI if we're not careful. You heard him earlier talking about an AI response to hypersonic missiles, a new class of missiles that are incredibly fast and can arrive with little warning. Let's pick up there.

I led an AI commission for the Congress, which made a recommendation that we ask China and Russia to forswear the automatic launch of nuclear weapons. As you know, in the United States, the president, the human president, has to make the decision affirmatively to launch a nuclear weapon. There's no other role prescribed except for the president to do that. So you see the problem. It's the compression of time, the imprecision of the recommendations, and they're destabilizing.

Missiles are one possibility for mayhem. Deepfakes are another. Jackson gave that a quick try earlier with audio. Now, anyone with video editing software can paste, let's say, a politician's head on someone else's body and make a goofy satire video. But deep learning can be used to create artificial videos that look and sound entirely genuine. The tools for that are open source and widely available. Here's Eric.

And one of the things that I learned when I was at Google is the power of video is extraordinary. So if you produce a false video, and you actually tell people it's false and you then show it to them, they still more or less believe it. That's a human problem. So we've got a situation where these tools are going to be misused and they're going to be broadly available to anyone, including some evil actors.

AI doesn't have to trigger a war or create political instability to backfire. It can also just make us unhappy.

We also know that outrage is shared, if you will, 7 times more than reason. So what happens in these systems is they get automated with AI and they look for engagement. And the more engagement, the better the system. And the more engagement, the more outrage. So why are we surprised that we're all upset about social media? Social media is designed to make us outraged. And by the way, this is true of the left and the right and so forth and so on. It doesn't pick sides. We have to figure out a solution to this problem which reflects our values, but also reflects that these AI systems are driven by objective functions. If you give the AI the objective function to make me happy and one day I say I'm unhappy and it decides that everyone else should be dead, that's clearly the wrong objective function. So getting the objective function for these systems is going to be really, really hard. It's going to be hard for regulators, it's going to be hard for corporations, and it's going to be hard for citizens.

Eric has a book out about these challenges that is co-authored by Henry Kissinger, the former Secretary of State, and Daniel Huttenlocher, Dean of the Schwarzman College of Computing at MIT. The book is called The Age of AI and Our Human Future.

The book came about because Dr. Kissinger and I and Dan Huttenlocher were talking about all of this, and he was very interested in the historical parallels. Most of us, certainly myself and Dan, are ignorant of the sort of deep history of how societies have changed. And Dr. Kissinger said, This is such a big change that it's epochal. It's a change of the nature of the Age of Reason, where people learn to think independently of what God told them. That was the Age of Faith. And he believes, and we have written down in the book, that this is the first step to a new age, the Age of AI, where we're going to have these intelligences which are non-human, but they're going to be in our lives. We're going to have to learn to work with them. And the most interesting aspects of working with them is what happens when humans coexist with them. What does it mean to be human? What does it mean to be an elderly person? What does it mean to be a child? How do you do diplomacy? What does it mean to have free speech? What does it mean to have all of this misinformation? These are the questions that we explore in the book.

Eric talked about enormously positive uses for AI too. It can discover patterns in how cells work, patterns that it might not understand but which nonetheless can be used to predict, for example, the effect of new drugs. AI-based drug development will save many millions of lives, Eric says. He compares AI to the telephone and the internet, which became popular and essential, and which can be used for bad purposes as well as good ones. It's up to us, he says.

It may be that humans are not really well set up for the kinds of automation and sort of, if you will, the robotic overlords that people talk about in the movies, but we're still in charge. We can still make a decision to stop that. You know, people always talk about, oh, the robots are going to take over the world. Not without us watching. And remember, we can always unplug them if we get really worried.

I asked Eric, As an investor, does he expect AI to benefit large tech companies or small ones? He says both.

The valuations of AI startups are in the billions, and these are valuations of companies that have no real revenue and the hope for transformative platforms. So the market, which is remarkable in the United States, is doing a fantastic job of getting capital to these companies to build them as competitors. The large companies use AI because they have so many signals. So if you think about it, a search engine, any form of social network, you have an awful lot of clicks. You have a lot of data to train against. All sorts of people will say to me, "Why can't you build a system that can predict a terrorist act?" And the simple answer is, thank goodness, they're rare. We don't have a lot of training data as to how these evil people assemble.

Eric says not all companies participating in AI will win, but a few will win tremendously. UBS recommends that investors look beyond mega-cap tech companies to small and mid-sized ones, which might face less regulatory risk. Oppenheimer lists leading companies developing AI chips. Some we've mentioned, like NVIDIA and Broadcom and Marvel. There's also AMD, which is buying another chipmaker called Xilinx. Qualcomm, and of course Intel. There are ETFs that focus on AI, but exposure there varies. For example, Global X Robotics and Artificial Intelligence ETF, ticker BOTZ, BOTZ, has plenty of makers of industrial robots like Japan's FANUC and Switzerland's ABB. If you hold an S&P 500 index fund, you already have plenty of exposure to AI giants like Amazon, Microsoft, Tesla, and NVIDIA. If you do look for artificial intelligence funds, be sure to distinguish between ones that pick AI companies and ones where AI does the picking. Eric says those aren't that big yet, but they're coming.

I'm familiar with about 3 such projects which have not yet succeeded, but they're in development and there are probably more that I'm not aware of. And here's what they do. They take all of the signals. If you set up your own investment firm, you can purchase from companies like Reuters and Dow Jones and so forth all of this essentially traditional data for a fee, real-time data. And then you combine that with other signals. And the most favorite signal that's used is Twitter. But there are hundreds of such signals. And you put it all into one big system and you say, Let's see if we can get the system to predict what will happen next based on what happened in the past. Today, those systems are not very accurate, but most people believe that with more computation, more signals, and better algorithms, they will get better. So I think that eventually the majority of trading will be AI-enabled. So it'll be a huge transition. Now, will it take 10 years, 15 years? Something like that.

To learn just a bit more about using AI to try to beat the market, I spoke with a company that does just that. That's next after this quick break.

Data is everywhere. When orchestrated properly, it sings. At Morningstar, we analyze and enrich data, making it actionable and powerful for you. Morningstar, where data speaks.

Harvard Business School Executive Education delivers breakthrough learning for leaders, stimulating classes led by faculty at the forefront of their fields, topics that will define the future of business, discussions that transform perspectives and ways of thinking, and access to the brightest business minds on the planet. Learn more at hbs.me/breakthrough. That's hbs.me/breakthrough.

Welcome back. There is this notion in finance— it's not new— that companies become less profitable as the business gets more and more sophisticated and digital.

Thank you, RoboJack. Take a break, and I'll handle it from here. Welcome back, everyone. Meet Chris Nativedad, Chief Investment Officer of Echobot, which has a fund called AI-Powered Equity ETF, ticker AIEQ. Chris calls it the first of its kind. It uses IBM's Watson and large troves of data, some of it from traditional financial sources and some from other places like social media and news reports and message boards.

We're taking all of this structured and unstructured data and looking for patterns associated with the companies and the stocks. In the case of the AI EQ, which are the stocks that are going to give us the highest probability of market appreciation, market price appreciation.

Just to be clear, Chris is not telling the program which patterns to look for. The program is finding its own patterns. I asked, what if the program comes back and tells you something that doesn't seem to make sense? That every time Elon Musk tweets about cannabis, you should buy semiconductor stocks. Chris says the program favors different types of data depending on the stock. For a bank, it might rely more on structured data, whereas for GameStop, it might weight comments on the WallStreetBets forum on Reddit more heavily.

We anticipate that the explosion of data will continue. So like, how do you— what you're saying is, how do you deal with a new form of information, And so we've got IP surrounding the data flow to say, hey, is this coming from a credible source? Do we see this coming with frequency, regularity, where we can see historical market impact across different horizons? The analogy I like to give is that of a search engine, right? If you type in something or look at a particular topic, those that are the most hits or most frequency or the greatest amount of impact relative to user search those will appear at the top of your search engine. For us, those are identified at the top of a trading and recommendation list.

Chris says trading is active with daily turnover equal to 1 to 2% of the portfolio. He says if the program senses rising volatility, it can shift to a bit more cash or take other actions he says can reduce risk. The fund launched just over 4 years ago. The expenses are 0.8% per year. Morningstar compares the ETF with large company growth funds and says the performance has been unremarkable. Chris says he and the CEO have all of their retirement assets in the fund and that he believes the program is getting better.

Yeah, so far, you know, it's been mixed. The first year we underperformed, the second year we matched the benchmark, the third year we outperformed. And this year we're still going through— in the past 6 months, we've been up over 300 basis points. So again, we continue to see the system continue to learn and we're almost flat the benchmark. But we believe in the technology, as do our investors, and we believe that it's going to continue to learn and adjust with time.

Thank you, Eric and Chris and RoboJack and RoboJackson. And thank all of you for listening. Jackson Cantrell is our human producer, so he says. Subscribe to the podcast on Apple Podcasts, Spotify, or wherever you listen to podcasts. If you listen on Apple, please write us a review. If you want to find out about new stories and new podcast episodes, you can follow me on Twitter. That's @jackhoug, H-O-U-G-H. See you next week.

Data is everywhere. When orchestrated properly, it sings. At Morningstar, we analyze and enrich data, making it actionable and powerful for you. Morningstar, where data speaks.