> consider this paper that found that neurotypical people give more money to charity when people are watching, but autistic people don’t. The authors write: "[I]n healthy individuals, improving one's social reputation acts as an instrumental reinforcer because better social reputation is rewarding. We think that there are at least two possible explanations for this deficit in ASD individuals. [...]" That is, autistic people being less hypocritical is characterized as a deficit in social reasoning.
When I read this, I thought the paper was correct, because their reasoning is rational (i.e. a rational person should be willing to pay more if they get a greater benefit in the form of a better reputation). However, the details of the method seem to confirm your criticism. The "person watching" in the experiment was "an unfamiliar person" supposedly brought in to compensate for faked computer problems. The expected reputational benefit in this case is likely to be minuscule or zero, since the observed action is (what most would consider) a supererogatory donation & the observer is a stranger whom the experimental subject is likely to never meet again; thus treating the reputational benefit as negligible is the rational decision, & it would probably be more accurate to interpret the study as finding that neurotypical people, but not autistic people, irrationally overvalued the observer's opinion of them.
1) It might be irrational in that particular artificial experiment, but it's likely this would generalise to the real world where they would behave similarly, and therefore receive reputational damage
2) "More hypocritical" is an extremely uncharitable way to interpret the results given that the neurotypicals gave more to charity on average than the austistics. Unobserved, they gave the same amount as the ASD people did both unobserved and observed (not statistically different.)
I'm so deeply scalar consequentialist that "moral obligation" doesn't even truly make sense to me, but perhaps you can explain to me: why would it matter to the measured difference between autistics and NTs whether the easily avoidable helping of another is "supererogatory" or "dutiful"? Say there's an unmonitored pay box for a campsite. Would we expect roughly the same influence of a personally unfamiliar observer on NT rate of paying for the service?
I forget when it was and don't have a link handy, though Eliezer Yudkowsky once cheekily tweeted that one has a duty to maximize expected positive consequences. A kernel of insight in the joke is the fact that these various ethical frameworks are partially just contrivances people use to coherently systematize their moral intuitions in the context of society. While it may be puzzling, an intuition that you're obliged to fulfill consequentialist maxims may catalyze for you a more intuitive understanding of deontological mindsets.
My take: Doing something supererogatory will make the observer feel positively about you. Not doing something supererogatory will make them feel neutral.
Doing something obligatory will make the observer feel neutral towards you. Not doing something obligatory will make them feel negative towards you.
People feel differently about negative and positive things: negative things feel “worse” than positive things that should have the same magnitude.
> Mental illness often isn’t much evidence one way or other about whether someone was victimized. True, many mentally ill people misinterpret “you’re busy” as “you hate me personally and want me to die,” or similar distortions that affect their ability to accurately report on interpersonal situations. But the distortions work both ways. You get mentally ill people going “Eve was busy, so she hates me and wants me to die”, but you also get mentally ill people going “Eve tells me I’m worthless all the time, but I am completely worthless, so she’s just giving me helpful criticism that will help me improve as a person.”
It sounds like you're saying "mentally ill people have a high false positive rate, but they have a high false negative rate too", so it kinda balances out?
But if the context is "a mentally ill person has accused someone of victimizing them and we would like to know if that's true or not", these don't balance out. The opposite, actually: a high false negative rate decreases the probability that any given positive measurement is a true positive.
Consider: someone is tossing a coin and telling you the result. Suppose 10% of the time, if they see tails, they'll say heads. Now when they say heads, you think there's some chance it was tails. (p(heads | they-say-heads) ≈ 90.9%.) But also, 5% of the time if they see heads, they'll say it was tails. Now *when they say heads*, you think the chance it was actually tails is *higher*, becauses the collection "coin flips where they say heads" has the same number of false positives but fewer true positives. (Now p(heads | they-say-heads) ≈ 90.4%.)
I agree that behavior is more powerful in predicting people's actions than diagnosis. But why not consider both? Let's say Heidi and Ivan both are generally 45 minutes late, but we know Ivan has ADHD and Heidi doesn't. Shouldn't we judge Ivan as more likely to be late in the future than Heidi?
Another difficulty when making judgment based on diagnosis is that we don't typically know who has a diagnosis! Very many people with diagnoses aren't particularly open about it, and some (probably a smaller percentage) of people who claim to have a diagnosis may be lying.
"Shouldn't we judge Ivan as more likely to be late in the future than Heidi?"
I am surprised that you treat this as the obvious default. You stipulate that Heidi and Ivan "both are generally 45 minutes late". As far as I can tell that literally means that, independent of what slice of time we measure (the past? the past and future?), the two are (mean? presumably?) 45 minutes late. If that's so, no other facts about either of them should change our expectation of their lateness—not Ivan's ADHD, not Heidi's unreliable childcare, not Ivan's unreasonable boss, not Heidi's proclivity for spontaneous human combustion.
Moving in a more colloquial direction, one might imagine that ADHD is a more stable cause of lateness than other causes (so, after the first lateness from each, where Ivan attributes his lateness to ADHD and Heidi attributes hers to, e.g., a train derailment), we might imagine that Ivan's lateness is likelier to continue. But I'm not persuaded that this is the case. People with mental illnesses can enter treatment, regress to the mean, or develop more effective coping/management skills. Other causes for lateness may be statistically much less tractable! If Ivan has ADHD that's responsive to medication, and Heidi only commutes by a bus that runs once an hour and experiences 40% service gaps, I think it's probably easier for Ivan to become reliable than it is for Heidi. And so on.
> I am surprised that you treat this as the obvious default.
I’m not that surprised. Let’s say A and B are both usually 45 minutes late, but B is a lower-status person than A. Should you judge B as more likely to be late in the future than A? Yeah, you probably should.
I mean, epistemically, they’re both just as likely to keep being late, of course, but chances are it’s in your best interest both to stay on A’s good side, and to further degrade B, bonding with the people who already don’t respect the latter very much—especially if A is one of them.
Therefore, you should publicly express self-righteous contempt for B’s lateness, while ignoring A’s or making excuses for it. And, since, for most people, the easiest way to fake is to not actually fake, your optimal approach is probably to convince yourself that B is indeed more likely to be late in the future than A.
I would argue no to your hypothetical. Ivan compared to Heidi has one more factor that introduces more variance to being late. There is medication that could help treat Ivan's ADHD. If you do not know whether Ivan has been medicated in the past, this allows for two options. He was not medicated before, but is medicated now, and could arrive earlier than is usual. He was medicated before, but forgot a pill, and could arrive later than usual.
You could research base rates to see which of these two possibilities are more likely, but in general, I would predict Ivan to have more variance in arrival time than Heidi, but when they are high starting at 45 minutes late, greater variance makes me think Ivan has a higher chance of arriving on time.
I wrote all this, but honestly, using all this math is a little silly, and you are probably better off just treating your friends the same, for social reasons anyways.
A major problem common to discourse about neurotypicality vs. neurodivergence is that very different types of psychological dispositions are conflated with each other. This has several bad consequences:
- Neurodevelopmental conditions like ASD or ADHD aren't themselves mental illnesses, but are then commonly perceived as such.
- Common mood disorders and other mental illnesses that just as often as not don't co-exist in one person are confused with each other.
- A somewhat positice correlation between neurodevelopmental conditions, and mental illness, distracts from the fact just as often as not someone will have one without the other.
- The fact that any such conditions can exist as a matter of degree as opposed to themselves can mislead many into deluding themselves and each other into assuming they're 100% mentally healthy, which may not be true for something like one third of the population at any given time.
- Some of those who are mentally ill are better at hiding it than others, meaning they can weaponize that status against other mentally ill people better, as well as being able to rationalize their own anti-social behaviour when nobody notices the patterns well enough to check the other person.
“Argument screens off authority” is basically the same as “argument screens off status”, and, when the argument is about you, also “behaviour screens off status”, which is epistemically true, but probably not what people care about most of the time. After all, officially labelling you as mentally ill is an excellent way to lower your status for most ordinary interactions, and the whole point of low status is to disadvantage you for the benefit of higher-status people.
Thank you for elaborating your position. I think I agree with your general framework, to the extent that we can observe past behavior it does seem a more relevant predictor of future behavior than diagnosis. I don't think I agree with your application of it to the Nonlinear situation. From my perspective, as a person who has never met Alice or Chloe, yet who wants to evaluate their credibility as they are the key witnesses, what am I to do? I don't have direct observations of their behavior to go on. I have some of Nonlinear's reports of their behavior, for example the many allegations against previous employers, and that to my mind is much stronger evidence. But the history of institutionalization certainly helps support and contextualize the inference I would make from that long pattern of making allegations against previous employers. So it does seem relevant. In a legal context there is a rule (federal rule of evidence 403) which says that even relevant evidence can be excluded if its prejudicial effect substantially outweighs its probative value. Maybe that's the case here? I guess it depends on how you think the audience (EAs in this case) will respond to a mental health diagnosis. I think this community is pretty good about that. So I still think including it was fair game.
> consider this paper that found that neurotypical people give more money to charity when people are watching, but autistic people don’t. The authors write: "[I]n healthy individuals, improving one's social reputation acts as an instrumental reinforcer because better social reputation is rewarding. We think that there are at least two possible explanations for this deficit in ASD individuals. [...]" That is, autistic people being less hypocritical is characterized as a deficit in social reasoning.
When I read this, I thought the paper was correct, because their reasoning is rational (i.e. a rational person should be willing to pay more if they get a greater benefit in the form of a better reputation). However, the details of the method seem to confirm your criticism. The "person watching" in the experiment was "an unfamiliar person" supposedly brought in to compensate for faked computer problems. The expected reputational benefit in this case is likely to be minuscule or zero, since the observed action is (what most would consider) a supererogatory donation & the observer is a stranger whom the experimental subject is likely to never meet again; thus treating the reputational benefit as negligible is the rational decision, & it would probably be more accurate to interpret the study as finding that neurotypical people, but not autistic people, irrationally overvalued the observer's opinion of them.
1) It might be irrational in that particular artificial experiment, but it's likely this would generalise to the real world where they would behave similarly, and therefore receive reputational damage
2) "More hypocritical" is an extremely uncharitable way to interpret the results given that the neurotypicals gave more to charity on average than the austistics. Unobserved, they gave the same amount as the ASD people did both unobserved and observed (not statistically different.)
I'm so deeply scalar consequentialist that "moral obligation" doesn't even truly make sense to me, but perhaps you can explain to me: why would it matter to the measured difference between autistics and NTs whether the easily avoidable helping of another is "supererogatory" or "dutiful"? Say there's an unmonitored pay box for a campsite. Would we expect roughly the same influence of a personally unfamiliar observer on NT rate of paying for the service?
I forget when it was and don't have a link handy, though Eliezer Yudkowsky once cheekily tweeted that one has a duty to maximize expected positive consequences. A kernel of insight in the joke is the fact that these various ethical frameworks are partially just contrivances people use to coherently systematize their moral intuitions in the context of society. While it may be puzzling, an intuition that you're obliged to fulfill consequentialist maxims may catalyze for you a more intuitive understanding of deontological mindsets.
My take: Doing something supererogatory will make the observer feel positively about you. Not doing something supererogatory will make them feel neutral.
Doing something obligatory will make the observer feel neutral towards you. Not doing something obligatory will make them feel negative towards you.
People feel differently about negative and positive things: negative things feel “worse” than positive things that should have the same magnitude.
On this specific point:
> Mental illness often isn’t much evidence one way or other about whether someone was victimized. True, many mentally ill people misinterpret “you’re busy” as “you hate me personally and want me to die,” or similar distortions that affect their ability to accurately report on interpersonal situations. But the distortions work both ways. You get mentally ill people going “Eve was busy, so she hates me and wants me to die”, but you also get mentally ill people going “Eve tells me I’m worthless all the time, but I am completely worthless, so she’s just giving me helpful criticism that will help me improve as a person.”
It sounds like you're saying "mentally ill people have a high false positive rate, but they have a high false negative rate too", so it kinda balances out?
But if the context is "a mentally ill person has accused someone of victimizing them and we would like to know if that's true or not", these don't balance out. The opposite, actually: a high false negative rate decreases the probability that any given positive measurement is a true positive.
Consider: someone is tossing a coin and telling you the result. Suppose 10% of the time, if they see tails, they'll say heads. Now when they say heads, you think there's some chance it was tails. (p(heads | they-say-heads) ≈ 90.9%.) But also, 5% of the time if they see heads, they'll say it was tails. Now *when they say heads*, you think the chance it was actually tails is *higher*, becauses the collection "coin flips where they say heads" has the same number of false positives but fewer true positives. (Now p(heads | they-say-heads) ≈ 90.4%.)
I agree that behavior is more powerful in predicting people's actions than diagnosis. But why not consider both? Let's say Heidi and Ivan both are generally 45 minutes late, but we know Ivan has ADHD and Heidi doesn't. Shouldn't we judge Ivan as more likely to be late in the future than Heidi?
Another difficulty when making judgment based on diagnosis is that we don't typically know who has a diagnosis! Very many people with diagnoses aren't particularly open about it, and some (probably a smaller percentage) of people who claim to have a diagnosis may be lying.
"Shouldn't we judge Ivan as more likely to be late in the future than Heidi?"
I am surprised that you treat this as the obvious default. You stipulate that Heidi and Ivan "both are generally 45 minutes late". As far as I can tell that literally means that, independent of what slice of time we measure (the past? the past and future?), the two are (mean? presumably?) 45 minutes late. If that's so, no other facts about either of them should change our expectation of their lateness—not Ivan's ADHD, not Heidi's unreliable childcare, not Ivan's unreasonable boss, not Heidi's proclivity for spontaneous human combustion.
Moving in a more colloquial direction, one might imagine that ADHD is a more stable cause of lateness than other causes (so, after the first lateness from each, where Ivan attributes his lateness to ADHD and Heidi attributes hers to, e.g., a train derailment), we might imagine that Ivan's lateness is likelier to continue. But I'm not persuaded that this is the case. People with mental illnesses can enter treatment, regress to the mean, or develop more effective coping/management skills. Other causes for lateness may be statistically much less tractable! If Ivan has ADHD that's responsive to medication, and Heidi only commutes by a bus that runs once an hour and experiences 40% service gaps, I think it's probably easier for Ivan to become reliable than it is for Heidi. And so on.
> I am surprised that you treat this as the obvious default.
I’m not that surprised. Let’s say A and B are both usually 45 minutes late, but B is a lower-status person than A. Should you judge B as more likely to be late in the future than A? Yeah, you probably should.
I mean, epistemically, they’re both just as likely to keep being late, of course, but chances are it’s in your best interest both to stay on A’s good side, and to further degrade B, bonding with the people who already don’t respect the latter very much—especially if A is one of them.
Therefore, you should publicly express self-righteous contempt for B’s lateness, while ignoring A’s or making excuses for it. And, since, for most people, the easiest way to fake is to not actually fake, your optimal approach is probably to convince yourself that B is indeed more likely to be late in the future than A.
I would argue no to your hypothetical. Ivan compared to Heidi has one more factor that introduces more variance to being late. There is medication that could help treat Ivan's ADHD. If you do not know whether Ivan has been medicated in the past, this allows for two options. He was not medicated before, but is medicated now, and could arrive earlier than is usual. He was medicated before, but forgot a pill, and could arrive later than usual.
You could research base rates to see which of these two possibilities are more likely, but in general, I would predict Ivan to have more variance in arrival time than Heidi, but when they are high starting at 45 minutes late, greater variance makes me think Ivan has a higher chance of arriving on time.
I wrote all this, but honestly, using all this math is a little silly, and you are probably better off just treating your friends the same, for social reasons anyways.
A major problem common to discourse about neurotypicality vs. neurodivergence is that very different types of psychological dispositions are conflated with each other. This has several bad consequences:
- Neurodevelopmental conditions like ASD or ADHD aren't themselves mental illnesses, but are then commonly perceived as such.
- Common mood disorders and other mental illnesses that just as often as not don't co-exist in one person are confused with each other.
- A somewhat positice correlation between neurodevelopmental conditions, and mental illness, distracts from the fact just as often as not someone will have one without the other.
- The fact that any such conditions can exist as a matter of degree as opposed to themselves can mislead many into deluding themselves and each other into assuming they're 100% mentally healthy, which may not be true for something like one third of the population at any given time.
- Some of those who are mentally ill are better at hiding it than others, meaning they can weaponize that status against other mentally ill people better, as well as being able to rationalize their own anti-social behaviour when nobody notices the patterns well enough to check the other person.
“Argument screens off authority” is basically the same as “argument screens off status”, and, when the argument is about you, also “behaviour screens off status”, which is epistemically true, but probably not what people care about most of the time. After all, officially labelling you as mentally ill is an excellent way to lower your status for most ordinary interactions, and the whole point of low status is to disadvantage you for the benefit of higher-status people.
Thank you for elaborating your position. I think I agree with your general framework, to the extent that we can observe past behavior it does seem a more relevant predictor of future behavior than diagnosis. I don't think I agree with your application of it to the Nonlinear situation. From my perspective, as a person who has never met Alice or Chloe, yet who wants to evaluate their credibility as they are the key witnesses, what am I to do? I don't have direct observations of their behavior to go on. I have some of Nonlinear's reports of their behavior, for example the many allegations against previous employers, and that to my mind is much stronger evidence. But the history of institutionalization certainly helps support and contextualize the inference I would make from that long pattern of making allegations against previous employers. So it does seem relevant. In a legal context there is a rule (federal rule of evidence 403) which says that even relevant evidence can be excluded if its prejudicial effect substantially outweighs its probative value. Maybe that's the case here? I guess it depends on how you think the audience (EAs in this case) will respond to a mental health diagnosis. I think this community is pretty good about that. So I still think including it was fair game.