AI may boost productivity, but investing still needs human guardrails

The first presentation of artificial intelligence is often cited as a computer program called Logic Theorist. Allen Newell, Cliff Shaw, and Herbert Simon developed Logic Theorist at the RAND Corp. in the 1950s to mimic the problem-solving skills of a human, cementing the idea that eventually computers would be able to compete with, if not outpace human understanding.

Artificial intelligence has come a long way since Logic Theorist. Now, programs like ChatGPT and Midjourney have made it possible to see computers outpacing humans in the very near future.

According to a recent report from Goldman Sachs analysts Joseph Briggs and Devesh Kodnani, generative AI — the technology that powers ChatGPT — could boost global labor productivity by more than 1 percentage point a year in the decade following widespread usage. Achieving that will require a $200 billion investment globally by 2025, the analysts say.

That headline figure is driving investment managers of all stripes into companies that could play a key role in widespread adoption of this technology. Their bets may come with a little inside expertise — at least on the hedge fund side — where some specialist money managers have been investing in machine learning for decades in an effort to generate a trading edge. And while much of the rhetoric around generative AI is about how it will eventually take all of our jobs, it may be a while before it dominates the investment process.

Hedge fund managers using these technologies say it's important to separate the hype from the reality. These technologies only recently became affordable enough to be considered for widespread adoption and there is still a lot of work to do on improving the capabilities of the technology as well as the validity of the results it produces.

Man Numeric is one such firm. Man Numeric, which oversees more than $41 billion and is part of Man Group, has been developing systematic investment strategies that rely on high-performance computing since 1989. Man Group also runs the Oxford-Man Institute of Quantitative Finance, a partnership with Oxford University focused on developing cutting-edge machine learning technologies for quantitative finance.

Gregory Bond, CEO of Man Numeric, head of the Americas for Man Group, and a special adviser to Man Group's multistrategy funds, said that while the work being done within the artificial intelligence space broadly is compelling, the uptake for investment managers has been more measured. And, there are still a lot of questions to work out.

Bond said that within Man Group and Man Numeric specifically, the approach has been more of a steady adoption of new technologies going back as far as the global financial crisis. "As we've gone through the last 15 years or so it's been a steady adoption. It started on the risk side using AI to help us on the risk part of our process and it has moved into stock and credit selection. It's also now moved into trading. If you look across our entire investment process, it's becoming an important part," he said.

The kind of artificial intelligence Man Numeric is relying on, however, doesn't work exactly the same way as some of the more popular platforms people are trying out today. Bond is quick to point out that there is still a lot of human involvement in the process to make sure that the technology is producing valid, usable results. That distinction, he said, is something investors should focus on when they are looking at money managers that claim to use AI in their investment process.

"We're seeing a renewed interest in AI from investors in part because of the conversation around generative AI," Bond said. "And while there is a desire to use the newest technologies, it's important to focus on what the reality is. Investors and managers should be asking whether AI is actually going to add value to the investment question they want to answer or process improvement they want to make. It may not always be the case that the answer is yes."

Ryan Pannell, chairman of Kaiju Worldwide, an investment firm that uses artificial intelligence as part of its investment process and oversees about $600 million in assets, agrees, and said there are still a number of barriers to entry when it comes to using AI.

"Until recently, people didn't have degrees in artificial intelligence," he said. "If you wanted to recruit a team, you had to understand computer science at a deep level and be able to identify the people who were going to be able to work on this stuff. It took a lot of time and money to be able to put together a team, and then you're building the technology yourself. That sort of naturally made artificial intelligence the domain of specialist firms that have the resources to go do that. The technology is making its way through the investment industry now, but investors should still look for managers that are going to be able to explain their process in depth."

Like Man Numeric, Kaiju has built human guardrails into their process to make sure that the strategies they create work like they say they will and that the investment decisions are sound.

"Artificial intelligence looks for patterns," Pannell said. "So it's not going to make thematic or fundamental research-based trading decisions the way that we think of those terms when they are decisions being made by humans."

Artificial intelligence could identify when a company is oversold based on patterns of mean reversion, for example, he explained. That's a different trading decision than a human deciding a company is oversold based on earnings reports. If a manager is using that kind of an AI-powered trading signal, they should be able to explain that nuance to investors and how it adds value to a strategy, he said.

Developers of generative AI point to its facility with language and its ability to provide information that is easy for regular people to understand rather than just computer scientists who have built a model. Using text prompts to get answers to investment questions could make the investment process more efficient for portfolio managers or analysts, but Man's Bond says it's not that straightforward.

He noted that the first iterations of this technology have been built on open access platforms so there is a potential for any investment-related query to be seen by others.

"There are a lot of practical questions about how to use generative AI in an investment setting," he said. "There are privacy concerns, for example. Where do the prompts live? Who can see them? If you're entering these questions on a quasi public platform, that could be problematic. So managers and investors have to think through the right balance in terms of what is behind the 'firewall' and what's more public. That's a regulatory question as well. And I think we're going to see the answers to those questions continue to evolve but it could limit — at least initially — how these technologies are used in an investment setting until that becomes clearer," Bond said.

Those regulatory questions are already being raised. At the end of July, the Securities and Exchange Commission proposed new rules that would require broker-dealers and investment advisers to take steps to address conflicts of interest associated with their use of predictive data analytics and similar technologies.

Marc Elovitz, co-managing partner at law firm Schulte Roth & Zabel, said the proposals are notable given how early they're being offered relative to when generative AI went public. He notes that it's more likely that these proposals make it to a final rules stage, even if there is a lot of back and forth with the industry on implementation, in large part because of how pervasive the use cases could be for this technology.

"We are seeing questions come up on (SEC) examinations already, and I think managers need to be able to have clear answers on how these technologies are used in business and investment operations," he said. "They may have to do that before we have final rules. To me its analogous to what's happened with ESG, where people said 'oh this isn't my core strategy so I don't need anything (such as a formal set of responses or policy), and even if that is still the case, managers are still expected to answer questions about how ESG is addressed on the business side."

The SEC's proposals are largely focused on increasing transparency in an effort to limit conflicts of interest but it's not totally clear what that means.

"The SEC wants to see increasing transparency, and I think it's important to note that transparency can mean different things," Elovitz said. "There are disclosures made to the government and there are disclosures made to investors. So we have to look at what is specifically being asked for. And, while disclosures to the government aren't necessarily public, there are still risks around intellectual property transfer as the SEC, like other government agencies, has been subject to data breaches in the past. There are also questions about what the SEC is going to do with that information. How do you evaluate quantitative trading models? Are you just looking for coding errors? Who makes that determination? There are a number of areas we still need clarity on."

Bond said with generative AI, there is an impulse to use it for its predictive capabilities but that raises questions of validity.

"I think there is a risk if you're using generative AI to forecast because you have to be very careful about the conclusions you're getting there. Whereas if you're using it to summarize an earnings call or write an investor flash, there is less risk," he said.

Kaiju's Pannell added that recent challenges in the digital assets industry have increased skepticism around technology adoption across the board, "but there are so many use cases, I think it's going to be unlikely that we don't see widespread uptake of AI in the industry over the next few years."

For Elovitz, that point is key. "There are core differences between AI and digital assets. Both have similar hype and interest but the use cases for AI are much more obvious and the technology is already embedded in a number of different arenas. It's not really something that investors and managers can avoid or escape. So while skepticism is definitely healthy, I think everyone is going to have to come to terms about where this technology is currently part of the process and where it might be in the future, so that there isn't surprise later on when regulatory questions get asked."

AI here to stay, but investing still needs human guardrails