A lot of the application of artificial intelligence in the wealth management industry has centered on the marketing and operations departments. But AI is also reshaping the investment process.

There are many applications for AI in investment management, but buying and selling securities should still be left to humans, argued panelists at the RIA Edge Los Angeles conference this week.

“In investment management, it’s important that there is always a human who has approved or decided that something should be purchased,” said Seth Merrill, former chief investment officer at Crewe Advisors. “If you delegate that, you can delegate that algorithmically to AI. But there has to be a moment at which you say, ‘I have risk management over these investment decisions, and if AI is going to help with the implementation or execution of those, I still retain ultimate responsibility for having said ‘Buy this today,’ or ‘sell this today.’”

AI can be a powerful tool to help determine whether a decision should be made, Merrill said.

“AI can trade off hundreds or thousands of factors and can help you winnow your initial impression about what the most important factors are down to a list that you can work with before making that decision,” he added.

Andy Stout, chief investment officer at Allworth Financial, said he would not use the technology to ask whether to buy or sell certain stocks because it’s designed to please the user.

Related:RIA Edge Podcast: Building a National RIA Brand Without Losing Identity with Kay Lynn Mayhue

“It’s always going to come back and say buy,” he said. “You can use it to build the data, build out the research to support your views one way or another and provide you with that, but some sort of end buy/sell signal, that’s a bad idea, at least right now.”

Allworth has developed a mock rebalancing tool, using AI on the front end to write the code. An advisor can put a model portfolio in the tool, as well as a tax budget or unrealized gains, and it’ll show them how to achieve their goals. If an advisor likes the rebalancing, they push a button, and it is sent directly to the RIA’s trading team, Stout said.

“It’s not AI making the decisions, but it’s AI guiding the decisions and the human still making that final input,” he said.

Merrill said AI grapples with non-stationary time series, meaning that the statistical properties of these series change over time.   

“Everyone wants to comb through the last 50 years of economic history in the U.S. and develop signals and patterns that the machines can run on our behalf. But we can wake up tomorrow and choose with a different set of institutions and incentives, different investments than we did yesterday,” he said.

Related:RIA Edge Podcast: Concurrent’s Nate Lenz on True Independence at Scale

There is an opportunity, however, to use AI in portfolio construction, he argued, because it’s a selection of weights in asset allocation or security selection.

“For me, there is an opportunity with AI in portfolio construction because portfolio construction sits at the nexus of the psychology of the individual you’re working with, the translation of that personality, psychology and experience into risk models, and the implementation of those risk models into an asset allocation program,” he said. “That’s a place—understanding individuals and their psychology, behavior and motivations—where AI can make a real difference.”