Marcus read about confirmation bias a year after that conversation in his kitchen, and it turned out to be a tool he could use on everybody except himself. He named it in his brother-in-law inside of a sentence. He named it in a colleague on a Tuesday and mentioned it to his wife that night, and she said mm, and kept loading the dishwasher.
He had never once caught it in himself while it was happening. He had caught it twice afterward, weeks later, looking back.
Marcus is invented, and he is the same man as in Part I. Keeping him invented means the argument runs on a way of reasoning rather than on anybody's drinking.
That gap has a name, the bias blind spot, and it is close to universal. This post covers three things. Reading about a bias does not much help you see it in yourself. Being smart does not help either. And the training that does work does not work by improving introspection.
What introspection actually gets you
Nisbett and Wilson made the case in 1977 and their argument is more careful than its reputation. The pop version is that you cannot know your own mind. What they wrote is narrower and harder to dismiss.
Their claim is that when people explain why they concluded something, they are not reading off a process. They are producing a plausible theory about themselves and reporting that. Sometimes the theory is right, and when it is, the accuracy is a coincidence rather than evidence that introspection worked.
They put a fence around it that hardly anybody quotes. They acknowledge the studies "do not suffice to show that people could never be accurate about the processes involved," and they restrict the claim to reports "about the role of critical stimuli in response to questions asked a few minutes or seconds after the stimuli have been processed."
They also concede that a person knows things about himself an observer does not. What he does not have is a window onto the machinery.
Which lands somewhere specific. Marcus knows his own history, his plans, and what he is feeling. Whether the reason Marcus gave his wife was his actual reason is a different kind of question, and paying closer attention will not answer it.
The part that should bother anybody who reads about this
In 2002, Emily Pronin, Daniel Lin and Lee Ross ran a study that has aged well and is not comfortable.
They gave ninety-one Stanford students six scales and asked them to rate themselves against the average student there. Seventy-nine of the ninety-one, which is 87 percent, put themselves on the better side.
Then the researchers described the better-than-average effect to those seventy-nine, explicitly, by name, and asked whether their own ratings had just been affected by it.
Nineteen of them said their responses had been biased. That is 24 percent. Another 63 percent said their ratings had been accurate and objective. The remaining 13 percent concluded that they had been too modest.
That is the bias blind spot, and what it shows is not that people are vain. Being handed the description of a bias, minutes after committing it, moved nineteen people out of seventy-nine. The rest took the description and pointed it at somebody else, and one in eight came away thinking they had undersold themselves.
Two things that number cannot do. Everybody in the study got the description, so there is no comparison group and no way to say the description caused the nineteen. And endorsing a bias when a researcher names it and asks you directly is an easier task than noticing it in yourself unprompted on a Tuesday.
There is a detail in that paper worth keeping, because it stops the finding from being a blanket claim about self-flattery. The asymmetry showed up for biases people cannot readily introspect and disappeared for shortcomings they could see plainly.
Being smart does not help
The obvious hope is that this is a problem for other people.
West, Meserve and Stanovich tested that directly in 2012 across the classic biases and reported that none of the blind spots "were attenuated by measures of cognitive sophistication such as cognitive ability or thinking dispositions related to bias." Then the sentence that does the damage: "If anything, a larger bias blind spot was associated with higher cognitive ability."
And one more, which decouples the two things people assume travel together. Being free of the blind spot does not help a person avoid the actual biases. Seeing yourself clearly and judging well are separate capacities, and having one does not deliver the other.
The measurement problem
The Pronin result treats a self-rating as an error, and some of those ratings were not errors. Roughly half of any group is above the median on a given trait. A Stanford student who put himself on the better side and then told a researcher that the rating was accurate may simply have been right. The design cannot separate him from the student flattering himself, and it counts both the same way.
The objection reaches further than one study. The West and Stanovich finding has the same structure. If the blind spot is the gap between what you say about your bias and how biased you actually are, a person who is genuinely less biased can post a larger gap without failing at self-knowledge. The training results are measured on the same instruments the training targets, so what has been shown to move is a score. Whether it moves anything on a Tuesday is the question I keep asking of other people’s studies and have not asked of these.
Something does work, and it is not introspection
Morewedge and colleagues ran two experiments where people got a single training session, either a computer game or an instructional video, aimed at particular biases. Confirmation bias was one of the three in the first experiment.
The reductions were large and they lasted. Across the two experiments the games cut bias by between about a third and a half right away, and by between about a quarter and a third at follow-up, which was eight weeks in one experiment and twelve in the other. The videos did less and still did something.
A 2025 study got a reduction from about forty minutes of scripted lecture and group case work, run with national risk analysts and compared against master's students.
The line worth keeping from that one is not the reduction. It is that the blind spot turned up in every group and had no relationship to anybody's actual confirmation bias score, before the training or after it. The analysts, who do this professionally, were no better attuned to their own bias than the students.
What the working interventions have in common is that none of them trains a person to inspect his own reasoning while it runs. They install a procedure. Look for the case that would knock the idea over. Ask what else would produce this. The procedure is portable and it can be run out loud, in front of somebody, which is the opposite of introspection.
None of this has been tested on anyone in recovery, as far as I can find.
Marcus catching his own confirmation bias twice, weeks after the fact, is close to what the good version looks like. What would improve on it is a procedure he can run out loud with somebody else, not faster self-detection.