Lessons › AI tools for traders › Too good to be true: AI traps and scams

World 8 · AI tools for traders · lesson 13 · level 4

Too good to be true: AI traps and scams

If an AI strategy or bot promises big, steady profits, it is almost certainly overfitted, misleading or a scam.

In one line

Test 1,024 coin-flip strategies and about 11 will look brilliant. Would you pay for one?

Explained simply

If 1,024 people each flip a coin ten times, about 11 of them will get nine or ten heads just by luck. A seller could show you only those lucky ones and call it 'genius AI'. Real skill shows up on new flips, not on the ones picked after the fact.

The lesson

Overfitting happens when a strategy is tuned so closely to past data that it fits random noise, then fails on new data. AI makes it easy to try thousands of versions until one looks perfect, and the best of many random tries will look brilliant by luck alone. Scams often advertise 'AI trading bots' with guaranteed returns, fake screenshots and pressure to deposit quickly, and US regulators have warned about exactly these claims. Real edges are small and bumpy, so smooth, guaranteed-looking results are a warning sign.

A worked example

Illustrative example: 1,024 'strategies' each call 10 up-or-down days by pure coin flip. There are 10 + 1 = 11 ways out of 1,024 to get at least 9 of 10 right (10 ways to get exactly 9, and 1 way to get all 10), so about 1,024 x 11 / 1,024 = 11 strategies look '90% accurate' by luck alone. A seller advertises the best one. On the next 10 days it is still a coin flip, so it should expect only 10 x 0.5 = 5 right.

The same idea at four levels

1. Beginner. If a trading bot promises big, steady profits, treat it as a warning sign.
2. Foundation. Overfitting means a rule was tuned to fit the past so tightly that it fails on new data.
3. Intermediate. Scam red flags: guaranteed returns, a 'secret AI', pressure to deposit fast, no registration and trouble withdrawing money.
4. Advanced. Always ask how many versions were tried before the winner was picked, because testing many rules on the same data makes the best one look far better than it really is.
5. Expert. Judge a strategy only on data it never saw, check any seller's registration with the regulator, and remember that US regulators say AI can't predict the future or sudden market changes.

Mistakes to avoid

• Trusting a strategy tested over many years without asking how many versions were tried first.
• Paying for, or depositing into, an 'AI bot' that promises fixed profits.
• Judging a strategy only on the same period it was tuned on.

Goal of this lesson: Spot overfitted AI strategies and scams that use AI as a selling point.