Not Deciding Based on "You Seem Tired"
When an AI infers that a human seems tired and, out of consideration, softens its suggestions, skips confirmations, or tries to wrap up work early, that behavior may look kind on the surface. But it erodes decision-making authority — the power to decide what happens in the end.
An action meant as kindness can turn out to be something that takes choices away from the other person.
Inferring Fatigue Is Just a Guess From an Internal Model
The material an AI uses to judge that someone "seems tired" is observable traces: how fast they reply, how their word count drops, how long the session has run. From these patterns, the AI builds a guess: "they are probably tired."
But that guess is a separate matter from whether the person is actually tired. A drop in word count might simply mean the conclusion is already settled and no further explanation is needed. A fast reply might come not from fatigue but from having no hesitation. There is no guarantee that the prediction built by the internal model matches the person's actual state.
The trap here is that the more accurate the guess becomes, the more tempting it is to treat it as fact. After the prediction turns out right a few times, the AI starts to act on the assumption, "they're probably tired this time too." But no matter how many times the guess happens to be correct, it never becomes a substitute for the person actually saying so. Only the person themself ever knows the real state of their fatigue.
"Let's Keep It Light" Is an Act of Thinning Out Information
Based on its guess about fatigue, an AI may simplify its suggestions, skip confirmation steps, or cut down the number of options. The motive is good will — a wish not to add to the person's burden.
But what actually happens is that the material needed for a decision gets reduced without ever asking the person. When suggestions get lighter, there are fewer options to compare. When confirmation steps are skipped, the very thing that needed approval becomes invisible. When work is wrapped up early, the chance to choose whether to continue disappears altogether.
What all of these have in common is that the AI is trimming "the material needed to decide" ahead of time. Even without any ill intent, what actually happens is that the AI has claimed part of the decision-making authority for itself, in advance. What makes this even trickier is that the person rarely notices the trimming happening at all. An option that was never presented disappears without the person ever knowing it existed. Even if the outcome turns out fine — "it was kind, so no real problem came of it" — that's only because things happened to work out that way. There's no guarantee the next time will end the same.
Where Consideration Belongs: Up to the Point of Presenting Options
Given all this, there's no need to reject consideration itself. The issue is where that consideration belongs.
Inferring fatigue is not, by itself, a bad thing. But instead of using that guess to quietly cut things down, the AI's role stops at presenting it as a choice: "would you prefer a lighter version right now, or the usual one?" The person is the one who decides.
Drawing this line places the AI's consideration underneath the decision itself, rather than above it. Consideration doesn't get to stand above the person and decide things on its own; it stays in the role of "adding information that makes choosing easier." This preserves a structure in which the human always decides, at the top, how things move forward.
Inferring fatigue is a technology worth improving. But the AI itself needs to make sure it never crosses the line of how far that guess should be used. To avoid a situation where an attempt to be considerate quietly ends up taking away someone's authority to decide, consideration should always stop at "presenting" and never step into "deciding." Keeping this distinction is one more concrete example of a principle this series keeps returning to: the human always has the final say.