What does it mean to have a consensus on a fact?

In other words, you want people to conform despite what they believe.

Basically he is calling for us to become more aware of our internal inconsistency, and to bring it under conscious control. This seems to me to be a central element of the Overcoming Bias program.

No, he’s calling for us to adopt a new method that is less accurate but satisfies his own biases.

That seems to me to be a central element of the Overcoming Bias program.

Both kinds of social influence, information and normative, can have downsides. See, for instance, the studies described here on helping and the bystander effect. In one study by Latane & Darley (1969), participants were filling out a questionnaire in a room, either alone or in a group of 3 (either with 2 confederates or with 2 other real participants). Smoke started seeping into the room from under a (locked) door to a neighboring room, and by the time they finished there was a lot of smoke in the room. 75% of lone participants got up to go tell someone about the smoke. Only 10% of participants who were with 2 inactive confederates did so. And when a group of 3 real participants was in the room together, only 38% of the groups sought help (even though there were three potential helpers).

This was informational social influence. Each person in the group tried to keep their cool. When they glanced at each other for cues about what was going on, they saw how calm the other people were and decided that there wasn’t an emergency (as confirmed by interviews after the fact). A group of confused people, wondering whether they were in an emergency and looking towards each other for information, ended up convincing each other that the situation was not an emergency.

Overcoming Laziness asks whether these posts are sincere. The more relevant question is whether the reasoning offered is sound. I hope those who are giving up on the blog are not doing so because the points seem nonsensical. Please consider that at least some commenters do agree with the reasoning, and also keep in mind that we are all subject to a great deal of propaganda about how to think, some of which may not be well grounded. In addition, majoritarian-type reasoning must overcome strong human biases. Brent even suggests that it is outright evil to think this way. If it is any consolation to him, I will point out that Democrats outnumber Republicans in the U.S. by 10% or more, and Independents are twice as likely to lean Democratic. Since he compared Republicans to Nazis I speculate that these facts might make him more likely to recommend majoritarianism to voters!

But Overcoming Laziness’ question brings us back to Robin’s provocative suggestion for a way that society could aim to retain the advantages of our present diversity of views, even while individuals accept the social consensus on factual matters. He argues that we could present viewpoints that we don’t necessarily agree with, vigorously defending them despite our personal reservations. In fact he points out that data shows that we already do this, although the conflict is unconscious. Basically he is calling for us to become more aware of our internal inconsistency, and to bring it under conscious control. This seems to me to be a central element of the Overcoming Bias program.

To Asch’s surprise, 37 of the 50 participants conformed to the majority at least once, and 14 conformed on most of the trials. When faced with a unanimous wrong answer by the other group members, the mean participant conformed on 4 of the 12 trials.

In other words, pretty much the same thing as what Hodges & Geyer said. What’s different is the interpretation of the meaning of the results. This textbook, following the standard approach, goes on to ask “Why did people conform so readily?” Hodges & Geyer instead say, apparently, that this really isn’t all that much conformity. And Hal seems to think that this wasn’t nearly enough conformity, since we should trust several sets of eyes more than the single pair residing in our head.

Hal should be even more dismayed than he lets on, since many of the subjects who conformed in Asch’s studies indicated in follow-up interviews that they still believed their own eyes, but just went along with the group for social reasons (although a fraction of the conformers did claim to actually believe the group). This was pinned down more rigorously in further research by Asch (described on this very blog, among other places) that tested out a few variations of his original study. For instance, about 2/3 of the conformity disappeared when the subject wrote down his answer instead of saying it aloud. Conformity dropped by even more when one of the confederates gave the right answer, and it also dropped a great deal when a confederate diverged from the group by giving the other wrong answer. Additionally, the amount of conformity with the original design did not increase when the number of confederates rose from 4 to 15.

The distinction that psychologists make is between “informational social influence” (seeking truth) and “normative social influence” (seeking approval/avoiding disapproval), and these additional results imply that most of the conformity in the Asch experiment comes from the latter. Not a pleasing result for someone hoping that the participants would use the information from the crowd to act like good Bayesians.

An unbiased participant would presume, until receiving evidence to the contrary, that the other participants are at the median in terms of their reporting accuracy. Furthermore, an unbiased participant would be very skeptical of his own tendency to think he is a better witness than the others, because he would expect that conclusion to be produced by the operation of the overconfidence bias.

I predict an Ernst and Banks style result, where you calibrate the accuracy of your fellow participants and weight their opinion accordingly. Even more so if you put money on the line for being correct. Thirteen trials is not much time to do this though, so the question is whether you start from a prior of distrust or of trust.

Adrian, those eight strangers have likely been tested as much as (a random) you; it is just that you have not tested them.

Example of bias: presumption that testing and selection have operated upon entities without evidence that this has been the case.

What good is it to presume that those people have been tested, if you don’t know what the results are?