Why AI Can’t Decide What We Should Do

A new experiment called FutureX is poking at a nerve that most AI benchmarks avoid. Usually, we test models on "historical" data—stuff that’s already happened. FutureX, however, asks humans and LLMs to predict live, unfolding events in real-time. It’s a test of foresight under genuine uncertainty.

The early data shows something fascinating: Humans still win most categories. But AI is winning on diligence. An AI doesn't get bored, and it doesn't skip the "boring" variables.

But here’s the thing, diligence isn't the same as wisdom. We’re getting better at predicting what might happen, but we’re still remarkably mediocre at deciding what to do about it.

Prediction is Cheap; Judgment is Expensive

The MIT economists who study this make a distinction that most tech enthusiasts miss. Prediction is just using data to fill in missing information. Judgment is knowing the "payoff": understanding what an outcome is actually worth to you.

Think of it like a corporate acquisition. An AI can crunch the numbers and tell you there’s an 82% chance of a 10% revenue lift. It can simulate every market fluctuation. But the AI cannot tell you if that 10% lift is worth the cultural burnout of your engineering team or the risk of your board losing faith in your vision.

Prediction is the map; judgment is deciding where you actually want to go.

The "Prethought" Problem

As prediction becomes a commodity (something you buy by the gigabyte), the real bottleneck shifts. It's what researchers call "prethought." Before you even look at a forecast, you have to do the heavy lifting:

• What are our actual constraints?

• Which risks are we willing to swallow?

• What does "success" look like beyond just a number on a spreadsheet?

AI makes the forecast easy, but it makes the decision harder because it gives us more possibilities to weigh. If you treat AI as a "decision-making machine," you’re essentially building a Ferrari with a brick on the gas pedal and no one in the driver’s seat.

Why the "Human in the Loop" is a Lie

We often talk about "keeping humans in the loop" as a safety measure. That’s the wrong way to look at it. Humans aren't there to double-check the AI's math; we’re there to own the consequences.

An AI doesn't get fired if a merger fails. It doesn't feel the weight of a ruined reputation. If you can’t own the downside, you haven’t actually made a decision; you’ve just followed an algorithm.

The strategic win over the next decade isn't going to the company with the best model. It’s going to the leaders who have the stomach and the clarity to apply judgment when the predictions are everywhere. We are entering the era of the judgment bottleneck. Prediction is now a commodity, but the weight of an irreversible decision cannot be outsourced.

AI can give you the forecast. Only you can own the consequences.