StockCharts Insider: The Dark Side of AI Stockpicking

Before We Dive In…

We’re in the early stages of an era defined by artificial intelligence (AI). And while people on the extreme polar ends might characterize AI as either panacea or doom, what we do know, on a more practical level, is that AI is an extraordinarily effective productivity tool. But when it comes to your financial investments, blind trust is dangerous—especially when the system operates like a black box. Too many unknowns. What data is it relying on? What context might it be missing? What if it’s hallucinating facts and presenting them with confidence? 

If those machine-driven decisions go wrong, you’ll end up eating the losses. So, would you trust your portfolio to a machine? You’d be surprised to learn that some people do. 

There are plenty of ads for AI-generated stock picks. Machine precision, they say. Fast or easy money. Algorithmic prophecy for the masses. What you don’t often see is push-back. Despite this, AI is still a useful tool. But the risks are often not laid out. Before you hand more trust to the algorithm, here are ten risks AI stock pickers probably won’t tell you.

1 - The Black Box Problem (Opacity Risk)

An AI system may come up with a decent stock pick. Still, it often won’t explain:

• Why that stock was selected
• Which variables and criteria matter most and why
• How the quality of that pick evolved over time

What it boils down to is this: you can’t validate what you can’t see. 

If the AI system you subscribed to won’t show its logic, you’ll need to supplement this by doing your own analysis in addition to the pick that it gave you. Otherwise, you’re just trusting a system whose process is opaque.

2 - Regime Shift Risk

Historical data is all we have to look at when making investment decisions. It goes without saying that AI models are trained on the same data. But can it detect transitions? Maybe yes, maybe no. But is the criteria transparent enough for you to evaluate?

Human knowledge can take in all sorts of seemingly disparate information, from headlines to technical and fundamental developments. For instance, an AI system may not take into account sudden interest rate shocks, geopolitical conflicts, sudden earnings misses, or supply chain disruptions.

The point here is that AI risks extrapolating the past into a market environment that no longer exists.

It takes human knowledge to identify that mismatch. Some AI system managers may or may not catch transitions. But if it isn’t transparent, the same can be said about the risks you’re taking, unless you do your own homework.

3 - Overfitting & Signal Fragility

This is a classic trading system problem that existed well before AI:

• Following patterns too precisely
• Fitting a trading system too closely to past data to make it “profitable”
• Too much focus on one technical or fundamental aspect at the expense of other factors

The risk here is that what worked in the past may not be relevant in tomorrow’s market.

The data it based its performance on may now be irrelevant. In the end, you can end up with a system that outperforms in backtests but fails in a live market. 

4 - Crowding Risk

If multiple AI stockpicking systems train on similar datasets or optimize on similar factors, then they may converge on the same pick. 

But if the systems aren’t trained to detect emerging conditions other larger investors or can intuit, then those picks, once they unwind, will unwind together.

5 - Garbage In, Garbage Out

All AI systems depend on data inputs. That can include financial statements, market feeds, and third-party data. Do you know where an AI system's data is coming from?

It’s an important question because if any of those datasets happen to be incomplete, outdated, erroneous, or even biased, then the output can reflect any or all of those.

6 - Curated Reality vs. Market Reality

If an AI model is trained on survivorship-biased datasets and idealized market conditions, then it may be missing an important aspect of market reality.

Markets are messy. Some companies that have survived and thrived may have experienced periods of near ruin (take SoftBank as an example). Some companies may have reached high valuations through a massive surge in investment capital, even when it was highly uncertain at the time whether those companies were going to justify their valuations in the future.

Leading stocks tend to look clear in retrospect. But the then-present duration will often have painted a very different picture.

7 - High Costs and Hidden Friction

This risk applies to AI-based short-term trading systems. Frequent trading comes with higher fees, transaction costs and slippage. Short-term traders already deal with this.

The thing to watch out for is that even decent trading signals can underperform once you factor in costs. And these costs aren’t typically accounted for in a system’s performance report.

8 - Incomplete Execution Awareness

Suppose an AI tool was correct in spotting what to buy and when to buy it. You then proceed to buy the stock and either blow up your account on an immediate pullback, or barely make any profit once the stock reaches its target.

Do you see what’s missing here? The system didn’t tell you anything about position sizing, risk management, or even trade management (adding to positions, or reducing positions, or trailing your stop losses).

The main point: execution, not just selection, often determines your outcome. Many advertised AI systems don’t cover this, as such decisions have to be made on an individual basis.

9 - The “Machine Knows Better” Trap

This is all about behavioral drift. You’re doing well with an AI system. Perhaps you start to get overconfident. Your analytical skills begin to atrophy and your decision-making gets more passive. You start slacking on your independent analysis. 

Overall, you were a better stockpicker before AI than after you adopted the system. And then the machine messes up. You see the mistake, and it’s an error that the older you would never have made.

10 - The Accountability Gap

If an AI-generated signal results in a monumental loss, who’s responsible? Is it the model, the platform provider, or the user?

When you did most of the work yourself, losses were part of a feedback loop. You learned from it. You improved (hopefully). But with no accountability, there’s no workable feedback loop. This detachment can be a real disservice, depending on how remote you are from the stockpicking process.

Enter the Dark Side—AI as a Marketing Weapon

This is where we move beyond inherent risks toward manufactured and possibly deceptive risks.

In addition to AI’s limitations, some platforms may overstate a system’s performance, often presenting simulated results as real results. It gets worse. Some may even fabricate track records or use hard-to-verify metrics. Supposing that’s the case, how would you know?

You can use AI to enhance analysis. In fact, we featured some of these functionalities in our ChatBot article. But AI can also be used to create highly-convincing yet deceptive marketing tactics. That’s just the world we now live in. So, watch out.

Regulators, namely the SEC and CFTC, have begun flagging such practices. We don’t need to cover all of this, but do look it up. I think you get my main point. If regulators are already asking hard questions, investors should be asking even harder questions, simply because their money is on the line.

The Insider Take: Where AI Actually Fits

If you’re going to subscribe to an AI-driven stockpicker, use it as a scanning engine, not a decision-maker.

It can help you generate ideas, find patterns in the market, and narrow your selection, albeit, at the expense of other stocks that may actually present (sometimes hidden) opportunities.

Don’t use AI to make a final judgment before pulling the trigger. Don’t pull the trigger without considering your risk allocation (e.g. position sizing, diversification, etc.). And watch out for trade execution costs and errors, as we covered earlier in this piece.

If you need additional info on any of these picks, use StockCharts’ tools to help you analyze your stock picks. If you’re unsure, use StockCharts’ own AI Chat Assistant which gives you plenty of information but not stock picks (it knows its limits and acknowledges them honestly).

A few more words of Insider Wisdom:

• If you can’t explain the pick and how it came about, don’t trade it
• Treat AI stock picks as hypotheses, not conclusions
• Always confirm by looking at price behavior on the charts
• Be cautious when coming across widely promoted “AI picks”
• Simpler systems you understand will often outperform complex ones you don’t (no black boxes)

And That’s a Wrap

AI stockpicking platforms may be popping up as the latest tool in every investor’s toolkit. But it doesn’t replace manually looking at structure, context, and risk management. In fact, it may demand more of all three. I’d argue that in markets, the biggest risk isn’t being wrong (we all make bad calls every now and then). It’s being overconfident and wrong on a huge scale. That’s where it may cost you more than just capital to recoup.