Can AI manage your portfolio? InvestorAI's Bruce Keith answers on 'Zero Sum'

Artificial intelligence is moving deeper into investing, from hedge funds and quantitative trading desks to retail portfolios and automated execution.

But as access to AI-powered tools expands, a more important question is emerging: can investors actually trust a machine with their money?

Bruce Keith, CEO and co-founder of InvestorAI, believes the technology can dramatically improve how people invest, but only when it is used for the right job.

Speaking on the latest episode of Zero Sum, Keith explained why asking ChatGPT or another large language model to tell you what to buy next may be the wrong approach, why AI could be better at short-term market decisions than long-term investing, and what investors should demand before handing capital to an AI-driven strategy.

The different facets of AI investing

Keith's distinction between general-purpose AI and purpose-built investment models is central to the discussion.

Large language models are trained to answer questions using enormous amounts of information.

That makes them useful for understanding what happened in a portfolio, summarizing news and explaining why an asset moved.

But forecasting a specific market move is a different problem.

Keith argued that effective investment models need to identify recurring market patterns, potential breakouts and points where those patterns change rather than simply retrieve information already available online.

To do this, his firm took an unconventional approach: adapting computer vision, the same technology used in facial recognition and self-driving cars, to analyze market data.

By translating massive datasets into visual representations, their AI can "see" complex breakout patterns that human analysts, or text-based LLMs, would completely miss.

That distinction matters as more retail investors begin treating AI chatbots as informal financial advisers.

The real investing edge may be knowing when to get out

One of Keith's most interesting arguments is that entering a trade is relatively easy. The harder problem is knowing when to exit.

"No one tells you when to get out."

His broader point is that investors tend to focus heavily on finding the next winning trade, while risk management often determines whether those gains actually survive.

Keith said AI is particularly useful for short-term decisions, defining that window as roughly two to three months.

Humans, in his view, remain better suited to longer-term judgments because they can work with incomplete information and incorporate factors that are difficult for machines to quantify.

That suggests a hybrid model rather than a fully automated one: machines process far more information than an individual investor can, while humans retain responsibility for the bigger picture.

AI investing needs more transparency before investors can trust it

That raises another problem: how does an investor determine whether an AI-powered strategy actually works?

Keith argues that the industry needs clearer standards for presenting performance.

Investors should not be satisfied with a single headline figure or an impressive-looking backtest.

"If what you're looking for is consistency, consistency of output, consistency of outperformance, then the question has to be how consistent has this thing been?"

He recommends looking first at win rate and beat rate, followed by average returns and drawdowns.

A strategy that wins frequently can still destroy capital if its losing trades are sufficiently large.

The implication is straightforward: investors should judge an AI tool much like they would judge a human fund manager, with a verifiable track record, clear methodology and an understanding of how it behaves in both rising and falling markets.

Regulators are trying to catch up with a technology that keeps moving

Keith also expects regulation to become a major part of the AI investing debate.

He said regulators are inherently playing catch-up, but argued that a principles-based approach could be more effective than trying to write detailed rules for every new application.

Most importantly, he believes responsibility cannot be outsourced to the machine.

For financial firms deploying AI, that means maintaining human oversight, risk controls and accountability even when decisions are increasingly automated.

The next phase is AI that does more than recommend trades

The technology could eventually move beyond generating signals and into execution.

Keith said automated execution is already becoming more common, with AI potentially deciding when to trade, when to stop and when to pause after losses.

But he stressed that the necessary risk controls must be built around the system before investors hand it meaningful amounts of capital.

That may be the real next step for AI in financial markets: not replacing investors entirely, but becoming an increasingly powerful layer between information and execution.

Keith's own closing investment calls reflected that balance. He cited Indian infrastructure spending and JSW Steel among his bullish ideas, while naming US leadership as his bearish call and "listening to the youth" as his wildcard.

Watch the full episode of Zero Sum for the complete conversation on AI-powered investing, transparency, automation and what it takes to trust a machine with real money.