Should You Let AI Set Your Goals?

The plan is right there on the screen, and it looks good. You typed one sentence about wanting to run a half marathon, and the AI came back in four seconds with a title, a description, and eight neat steps that build week over week. Nothing is wrong with it. That is the strange part. The steps are sensible, the ordering is reasonable, the whole thing reads like something a decent coach would hand you. And yet a small question sits underneath the polish: a plan you did not write, is that a plan you will actually follow?

It is a fair question, and the honest answer depends less on the quality of the plan than on two things that happen inside you afterward. Both are documented. One decides whether you keep trusting the AI once it slips. The other decides whether you feel enough ownership to keep going when the plan gets hard. Get them right and AI is some of the best goal-setting scaffolding you can have. Get them wrong and the tidy eight-step plan becomes the thing you quietly abandon by week three, wondering why.

You Disown What You Watched Slip

Start with trust, because AI loses it in a specific and slightly unfair way. In 2015 Dietvorst and colleagues ran five studies on what they named algorithm aversion. People forecast real outcomes from real data, and a statistical model forecast the same outcomes alongside them. The model was genuinely better. On one task the human forecasters made 15 to 29 percent more error than the model; on another they made 90 to 97 percent more. Then participants chose whose predictions to tie real money to, their own or the model's.

You would expect them to take the winner. Many did not. People who had watched the model work were less willing to rely on it, even in the conditions where they had just seen it beat the human. The mechanism is the uncomfortable part. Dietvorst and colleagues found that we lose confidence in an algorithm faster than in a person after seeing the same mistake. A human forecaster errs and we shrug, they are only human. The model errs and something in us quietly decides it is broken.

There is a nuance here, and it cuts in AI's favor. People are not born hating machines. In the conditions where participants never watched the model work, many were happy to use it. Aversion switched on after the observed mistake, not before. So the danger is not that you will refuse an AI plan on principle. The danger arrives three weeks in, the first time the AI's advice misfires, when you let that single miss discredit everything it suggested.

Map that onto your eight-step plan. The AI will get something wrong eventually. It will propose a step that does not fit your body, or a timeline that ignores your job, or a rest week in the wrong place. A coach who did that would earn a raised eyebrow and a follow-up question. The AI that did it risks the whole plan. Once you catch the machine being wrong, aversion closes the file, and the seven good steps go out with the one bad one.

You Commit To What You Helped Build

The second force runs the other direction, and it is about ownership rather than trust. In 2012 Norton and colleagues documented what they called the IKEA effect. People assembled plain storage boxes, then bid real money on them. Builders were willing to pay more for a box they had put together themselves, $0.78 on average, than non-builders would pay for the identical pre-assembled box, $0.48. Same box. The labor changed what it was worth to them.

The origami version is sharper. People folded amateur paper frogs and cranes, then valued their own lopsided creations at about $0.23 each, roughly what a separate group of outsiders paid for origami folded by experts, $0.27. Outsiders looking at the same amateur work saw close to nothing and paid about $0.05. The makers were not deluded about quality. They simply valued the thing their own hands had made.

One detail matters for goals more than any other. The effect only appeared when the labor finished. Builders who completed the box valued it far more, $1.46, than builders who assembled only half of it, $0.59. Effort that stalls partway does not create the attachment. Two more findings round it out: the premium held even for people who said they had no interest in do-it-yourself projects, and when participants built something and then destroyed it, the premium vanished. The attachment lives in the finished object, not in the sweat.

A caution before this becomes a slogan. Norton and colleagues measured willingness to pay for physical objects, not whether anyone stuck with a goal. So treat it as an analogy for ownership and commitment, not as a study of follow-through. As an analogy it is a clean one. We over-value what we helped build, and a goal you helped shape sits differently in your hands than one that arrived fully formed from a text box.

The Fix Is A Hand On The Wheel

Put the two forces together and it sounds like a stalemate. AI is better scaffolding than you would build alone, yet you disown it the moment it slips, and you commit to it only when you helped build it. The same researchers found the way through, and it is smaller than you would guess.

In 2018 Dietvorst and colleagues went back to the forecasting task and changed one thing: whether people could adjust the model's output before committing to use it. When they had to accept the algorithm's forecasts exactly as given, only 32 percent chose it. When they could tweak those forecasts, even within tight limits, 73 to 76 percent did. A little editing roughly doubled how many people were willing to rely on the better tool.

Two things keep that result honest. First, the editing did not make the forecasts more accurate. Often human tweaks make a good model worse. The gain came entirely from more people being willing to adopt the better tool at all, not from the edits improving anything. Second, people did not want much control. On average they nudged the model's forecasts by only about 8 percentiles, a fraction of what they were allowed. Dietvorst and colleagues found the preference was for some control, and it barely responded to how much control was on offer. A hand on the wheel, not both hands.

That last point is easy to misread. This is not a case for rewriting the AI's plan from scratch, which would defeat the point of the draft. It is a case for keeping a hand on it. The people who adopted the better tool were not the ones who overhauled it. They were the ones who could reach in and change a little, and mostly chose to.

So, Should You Let AI Set Your Goals?

Which answers the question you started with. Yes, let AI set your goals, in the specific sense of letting it draft. Let it turn your one sentence into a title, a description, and a first pass at the steps, the way a good assistant hands you a rough outline instead of a blank page. Then take the wheel. Rewrite the timeline that ignores your job. Cut the step that does not fit. Reorder the rest. You do not have to change much, and the research suggests a light touch is enough. What matters is that the plan passes through your hands before it becomes yours.

This is the same reason a borrowed goal tends to go slack while a self-concordant goal keeps moving. Ownership is what carries a plan through the hard middle, and a plan you never touched is a borrowed plan with a friendlier interface. AI can spare you the empty screen and break a large goal into steps faster than you would alone. The editing is not a chore to rush past. It is the part that makes the plan stick, and it doubles as the moment you decide whether this is really a goal that fits who you are becoming.

None of this means AI is a gimmick to switch off. Used as a vending machine that dispenses finished plans, it earns exactly the abandonment the research predicts. Used as a drafting partner you edit, it removes the two worst parts of starting a goal: the blank page and the guesswork about what the first steps should even be. The tool is the same either way. The difference is whether your hands touch the plan before you commit to it.

This is exactly how Future You uses AI. Describe what you want, and it suggests a goal title, a description, and steps, then hands every one of them to you to edit, reorder, or throw out. The draft saves you the blank screen; the shaping keeps the goal yours, which is the part that gets you to week twelve. You can see how the editable suggestions work on the features page, and the app is free on iOS and Android.