June 26, 2023

Let’s Talk About AI

Michael Carrico, CFP®, CRPC® Wealth Manager

Artificial Intelligence, or AI, has created a lot of buzz in 2023. One could argue it is a major reason the S&P 500 and Nasdaq Composite indexes have rallied so significantly year-to-date. Tech CEOs certainly seem to think so, mentioning AI constantly in their earnings calls with investors this spring.

We’ve all seen AI-generated art from platforms like DALL-E, Midjourney, and NightCafe and it really is incredible so long as you don’t ask it to make hands and faces. Speaking of hands, I’ve tried mine at picking out the AI generated art from a lineup and I’ll admit that I’m not very good at it (can you beat my 501 points?). Chances are you have also had some fun playing around with ChatGPT or another text-generating AI platform. All this excitement has sparked many debates from the potential of this technology to spur innovation and growth, to the future of jobs and concerns about education, to the fear of a Skynet-like AI overlord depending on your news sources. The more concerning topics even drove over 31,000 people including Elon Musk, Steve Wozniak, and Andrew Yang, to sign an open letter to pause giant AI experiments. So just what is going on with new developments in AI and how should we be thinking about it? This is a sprawling subject, and I am not an expert in artificial intelligence. My apologies in advance to any AI engineers who disagree with my understanding of the subject. By no means do I expect to cover everything in this newsletter, but today I would like to discuss what AI is, what the new developments are, and give some perspective on how investors might think about AI. Let’s dig in.

“1.A robot may not injure a human being or, through inaction, allow a human being to come to harm.

2. A robot must obey the orders given to it by human beings, except where such orders would conflict with the First Law.

3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws.”
― Isaac Asimov’s Three Laws of Roboticsⁱ

What is “AI”?

I jest a bit with Isaac Asimov’s Three Laws of Robotics, but some of the conversation around AI these days does indeed reach into this realm. As a culture, we have been steeped in stories of robots and artificially intelligent synthetic beings with human-like intelligence such as Data from Star Trek, Terminators, The Matrix, and Ex Machina. It’s unfortunate that 75% of the AI that first comes to mind from pop culture is malevolent. But these examples from fiction are representations of something called AGI, or artificial general intelligence, and a far cry from the AI that is driving excitement today.

To separate these terms, Merriam-Webster defines artificial intelligence in two ways, 1) a branch of computer science dealing with the simulation of intelligent behavior in computers and 2) the capability of a machine to imitate intelligent human behavior. I would argue that AI-generated art and ChatGPT fulfill this definition. We can spot inaccuracies in an AI generated hand but a human who isn’t trained in art might fare much worse. I certainly would. It does a good job imitating intelligent human behavior. Artificial general intelligence, however, is a sub-category of AI and something far more advanced. Merriam-Webster doesn’t have a distinct definition of AGI, partially because it seems that definition is still being hashed out, but it could be defined as an artificial intelligence that is equal to the abilities and intellect of a human in a broad sense. Rather than being a tool with the ability to tackle a specific objective such as a chatbot, search engine, or art generator, it would be capable of a variety of tasks and perhaps interpretation of the input request. An AGI would presumably need to be able to understand that a requested task may also require several intermediary tasks. For instance, “please mow my lawn” really means “please go outside, locate the lawn mower in the shed, make sure it is fueled or charged, based on the length of the grass determine whether to bag or mulch, avoid my flower bed with all those beautiful spring flowers, make sure my dog is inside because she is scared of the lawn mower, and put the lawn mower back in storage afterward.” Any version of AI we have today requires specific programming to know these steps and still your Roomba eats the rug and gets stuck under the bookcase. Our vision of sci-fi artificial intelligence is yet a step further with the ability for novel thought, planning, and problem-solving. If you have played around with ChatGPT or AI-generated art at all, you know that your prompt must be curated to get the output you want. The first output will probably be passable and can be refined with additional prompts. It does “learn” from inputs, but that process requires the input of a human with a vision for the desired outcome. ChatGPT and DALL-E don’t have intention, volition, or imagination. Artificial general intelligence may conceivably be within reach, but superintelligence that improves itself and surpasses human abilities (like Skynet) is still a long way off, if it is possible at all.ⁱⁱ

What AI Are We Talking About Today?

For an example of current AI innovation, let’s consider chatbots. There are several versions of this technology including Google’s Bard and Microsoft’s Bing chatbot, but I will pick on ChatGPT since it’s in the zeitgeist. I’m sure you know that ChatGPT isn’t Data or WALL-E, but this distinction helps put the current innovation in AI technology into perspective. We didn’t just invent artificial intelligence in 2022 with the launch of ChatGPT. We have been using more rudimentary versions of this technology for years. You have almost certainly interacted with a chatbot on Amazon, a brokerage website, or even your doctor’s web portal well before 2023. Those chatbots are typically used to answer simple questions and direct you to a relevant FAQ topic or connect you with a person when your requests get too complex. Internet search engines like Google use AI. In the world of finance, we could say some form of AI has been in use for many years in the form of algorithmic trading and robo-advisors. Algorithmic trading is a process for executing orders utilizing automated and pre-programmed trading instructions to account for variables such as price, timing, and volume.ⁱⁱⁱ A robo-advisor is a digital platform that provides automated, algorithm-driven financial planning and investment services with little to no human supervision.^iv While ChatGPT is an impressive improvement over old chatbot technology, is it truly something novel?

ChatGPT stands for Chat Generative Pre-trained Transformer and it is categorized as a Large Language Model (LLM). In layman’s terms (from a layman), it is generative in that it generates new sentences and paragraphs that resemble natural language from received prompts or inputs. It is not simply outputting a pre-programmed response like the chatbots of old. However, it also isn’t generating novel ideas. It is “pre-trained” on massive amounts of internet data that were fed into the program from which it “learned” information and was then trained by coders and developers on how humans structure thoughts and sentences through a trial-and-error type of process called reinforcement learning from human feedback (RLHF).^v In this way, a LLM like ChatGPT is a closed system. It can only generate responses based on the information fed into it and the responses which are generated aren’t nonsense because restrictions were added by way of training to refine the output. An interesting way ChatGPT has been described is as a word calculator. In effect, it is a highly complex and efficient algorithm. This isn’t to say ChatGPT isn’t remarkable. I was thoroughly impressed that ChatGPT-4 could describe a meme of a world map made of chicken nuggets and the humor inherent in that concept.^vi As with any technology, it is refined and improved over time and so new advancements accelerate, but this is an improvement on existing technologies and still has limitations.

Among the limitations for the current iteration of ChatGPT is that its “knowledge” (training data) ended in September 2021, so it cannot generate content with the context of new information since that cutoff. Another limitation is that, although the dataset is massive, it is still finite. ChatGPT doesn’t know everything, it simply has a very large database of information from which to draw when generating outputs. The information provided by ChatGPT is not always accurate.^vii ^viii Part of the reason for this is that ChatGPT is building sentences and paragraphs as it goes based on the pre-training provided to it. While it draws on the information in the database, it is not considering the full concept to be conveyed prior to beginning. Rather it builds sentences and paragraphs one word after another based on predictions and algorithms. For instance, an article like the one you are reading was outlined beforehand. There was an overarching theme and message planned out and the article was revised throughout to add new information to ensure thoughts flowed correctly to convey that theme and message. When a question arose as to the accuracy of information, additional research was conducted, and sources cited to provide a foundation for the claims made. ChatGPT doesn’t do this. The problem that arises is that users may implicitly trust the output of a large language model without conducting additional research to verify claims.

How Should Investors Think About the Current AI Developments?

Regardless of any limitations of the current iteration of LLMs or AI art generators, it is still an improvement in technology and presents the potential to disrupt businesses and, as we have seen this year, drive market returns. While the S&P 500 index is up 13.83% year-to-date, the tech heavy Nasdaq Composite index is up 29.68%, and Nvidia Corp (NVDA) is up 195.2% on news that it had a strong first quarter related, in part, to demand for AI processing power.

S&P 500 / Nasdaq Composite / NVDA

Markets are forward-looking in that market participants always try to get ahead of what may come. Any time a new technology has the potential to generate growth and investment returns, the market will place bets on who the winners and losers will be. This can often cause stock valuations to soar. As an example, the price-to-earnings (P/E) ratio of NVDA went from 62.19 on 12/30/22 to 223.50 on 6/22/2023. This means that at the beginning of the year investors were paying 62 times the earnings per share (EPS) for NVDA and that “multiple” has increased by more than 3.5x six months later. This is significant because investors are paying a lot for expected future earnings from this company. By way of comparison, the P/E ratio of the S&P 500 index was 24.43 on 6/1/23^ix. As reported by T. Rowe Price, at the peak of the dot-com bubble in April of 2000, the P/E ratio of the S&P IT Sector was 77, and the P/E ratio of the “AI Basket” as of May 2023 was 31.^x This is not calling a bubble in AI stocks in general or in NVDA in particular. There may still be significant headroom for both, and the debate is lively. This is simply an observation that the valuation of NVDA is historically high, and the company needs to generate significant future earnings to justify the valuation.

There is a straight line from demand for AI processing power to the potential benefits for a large chip maker who can supply that processing power, but each technological innovation creates knock-on effects in the economy and there are many businesses that may benefit from this emerging technology over the coming years. In the near-term, it is reasonable to expect significant investments in technology infrastructure, such as semiconductor chips and networking equipment, that will enable the adoption of AI. While NVDA has soared in 2023, there is the potential for other businesses to benefit indirectly from the effects of this technology in the longer term. Many companies are currently trying to understand how to incorporate AI into their businesses to drive growth.

Any time there is a significant amount of hype about a particular development in the markets, it warrants a reminder that chasing returns can lead to poor performance over the long run if an investor buys in at a high price and a company struggles to justify its stock valuation in the future. Two examples of hype cycles in the recent past in which some speculators benefited greatly from a quick run up in value while other investors got burned are cryptocurrency/digital assets hype and meme stocks. This is not a direct comparison, of course, and there are certainly differences between the companies benefitting from emerging AI technology and the above examples. However, before jumping into the hot topic of the day, it is important to pause and consider the price paid and the value expected. Part of our investment process includes a consideration of value. We look for the potential long-term beneficiaries of technological innovations across the broad market and buy them at reasonable prices. It may very well be that AI improvements will drive growth for years to come and we continue to monitor the market for opportunities.

Conclusion

AI is an interesting and exciting topic, and this article only scratches the surface. I have a feeling that there may be more articles to write on the subject before all is said and done. For now, I hope this article has been informative and please reach out to us if you have questions.

In case you were wondering, I do not use ChatGPT to write these articles. I have tried to use it as a research tool, but the propensity for inaccurate information leads me to double-check the AI responses and I still find that the tried-and-true internet research and use of trusted sources is the best bet. Perhaps someday that will change with the release of ChatGPT-4 or something even more advanced. Until then, can you tell there’s a human behind the keyboard?

[i] Isaac Asimov and the Three Laws of Robotics | SciHi Blog