Today, exchange-traded funds (ETFs) that rely solely on artificial intelligence (AI) to make investment decisions are readily available. Quant hedge funds have been prevalent in the industry for years. And the widely-held view—as reflected extensively in our Capital Markets Vision 2022—is that AI and machine learning will continue to have a big say in the future of investment decision-making for asset managers.

But while the potential of AI in asset management is substantial, it may not be limitless. In my view, it’s important to take a prudent perspective of how the mechanics of AI can impact its use in investment decision-making, and its limitations in relation to transparency, could restrict its use in certain product types.

There are two questions I’d like to consider in looking at this.

First, when, and how, will large, traditional asset managers turn over to machines the full investment decision-making process of mutual funds, ETFs and institutional products? And second, is this something they can either afford to do—or afford not to?

The answers to both questions will have significant implications for the future of asset management. In some industries, and within asset management alternative investments, AI is already used today not only to analyze but also to act. It’s unclear whether AI in the mutual fund industry will ever reach that point—so the ceiling to its application may be lower than we think in that regard.

First off, what’s the problem?

Different forms of artificial intelligence, such as machine/deep learning and natural language processing, are being used across the industry and gaining in terms of application and effectiveness every day. The goal is to give machines vague instruction or successful examples and large amounts of data, and then receive investment recommendations in return. With the ongoing search for alpha continuing to prove difficult despite turning over every rock in sight, AI should be a boon for investment management. As said before, it’s already being used to make investment decisions in pockets of the industry and supplement the work being done by human portfolio managers. Given all this, what’s not to like?

The issue lies in the mechanics. The ability to rely on machines to identify and execute investment opportunities could be somewhat limited moving forward, especially for products like ’40 ACT mutual funds. By definition, there is a “black box” aspect to machine learning, in that there is a lack of transparency into why the technology has identified a certain trend or opportunity. Over time this complexity only increases.

So, how can you fully trust AI?

Typically, the answer is simple: results. Testing the outcomes of AI outputs is the best way to gain confidence that it’s heading in the right direction. However, we all know that, in investing, “past performance is not indicative of future results.” This puts asset managers in a difficult situation. Relying fully on machines to make decisions can—and indeed should—be uncomfortable for asset managers, since it is difficult to be a sound fiduciary without fully understanding the models.

This issue is magnified with ’40 ACT mutual funds and their heavily regulated nature. Hedge funds, with their ownership structure and potentially more sophisticated investors, can feel better about the strategy if their investors have full transparency into their methods. Mutual funds, though, tailor to the smaller common investor and are subject to intense regulatory scrutiny. A mutual fund that relies on machines to make investment decisions undergoing a period of poor performance could become a regulatory or public relations nightmare or fuel an accelerating political bandwagon.

The real win for AI in the asset management industry may be more in the back and middle office operations than the front office. Across reconciliations, transaction management and risk management, for example, we know asset managers have begun experimenting with AI to optimize their operating models. In these areas, there would be less emphasis on transparency and more on results. The risk and cost reductions that could come along with AI could instead decrease cost and risk and aid in the investment decision-making process, rather than replace it.

Where does this leave us?

The fact is, the potential for AI is huge, but AI is not infallible. It’s simply a different way of getting to a desired outcome. When it comes to using AI successfully in the mutual fund front office, without trust or transparency it may never get the benefit of the doubt. In the future, AI may have the ability to provide greater transparency than today: explainable AI (XAI) is being discussed, but with varying levels of skepticism (as should anything that requires DARPA involvement to create). Will something like this be required in investment management? Will a well-defined process for utilizing these tools fill the gap? Is this a process, regulatory, or marketing and communicating issue? Or will the AI gains in the industry be relegated to alternative investments rather than mutual funds?

On these questions, the jury is still out. It will be interesting to see the industry’s final verdict—though that may take years to emerge.