#fundamental attribution error

[kenyatta]

The Psychology of Meritocracy and The Myth of The Self-Made Man — Neuroscience Of

What does it mean to be self-made? We like to believe the world is a meritocracy, where self-made people have succeeded solely on their own

While [the] standard definition of “self-made” (inherited wealth or non-inherited wealth) is straightforward, it begs a bigger question: What does it actually mean to be self-made? For anyone who attains this level of wealth, should this be seen as a solely personal accomplishment?  Implicit is the assumption that the wealthy deserve to be wealthy. But do they? And are there downsides to seeing wealth creation in these purely individualistic terms? The issue of being self-made cuts into a core area of social psychology: how do we attribute the actions of others? In everyday life, we naturally and instantly form judgments about the meaning behind the behaviors we witness. Imagine sitting in class on the first day of the semester, when the professor bursts in unkept, sweaty, and panicked; their briefcase full of papers flying everywhere. You may assume from this that the professor is simply irresponsible. He’s the kind of person who would show up unprepared for the first day of class. And you may be right. But it’s also possible that the professor’s behavior isn’t due to something about them. Instead, it might be something about their situation. Maybe, for example, they were sideswiped on the way to work, and they had to sprint to class.  This divergence in interpretation represents the two general ways in which we come to understand another person’s behavior: either we attribute it to the individual themselves, that it must represent a core, enduring personality trait, or that the behavior is due to the situation.  Research suggests that we're biased towards dispositional explanations for behavior. That is, we automatically assume someone’s behavior is due to their core, enduring personality, and not due to the situation they’re in. This is known as The Fundamental Attribution Error.  It's deemed an ‘error’ because the bias to assign a weight to the individual clearly comes at the expense of noticing obvious situational factors.

inspired by a reblog by @autistic-android42

[blubberquark]

The Opposite of Expert Systems

Back before the AI Winter, there was a lot of research into Expert Systems. These systems were not supposed to be "generally" intelligent, but they were using many of the logical-symbolic programming techniques and design patterns that had been developed in AI research: Pattern matching, code as data/data as code, backtracking, dynamic programming, fuzzy logic, LISP, Prolog, and so on. Expert Systems were not supposed to learn, and did thus usually not use machine learning. Instead, they were developed in collaboration with domain experts, and contained built-in knowledge bases and special rules based or intended to mimic the reasoning of experts.

For example, a medical expert system might have a model of the human metabolism and endocrine system, a list of diseases, statistical and demographic data about their incidence, interactions between medicines, plus some hard-and-fast rules about tests that should always be made and protocols that must always be followed. Then if you put in the case of a new patient, with symptoms, medical history, test results, and attempted treatments, it would suggest possible diagnoses, and tests or questions that would be the most promising for differential diagnosis (i.e. to rule out as many possible causes as efficiently as possible). Such a system could, through symbolic inference and simple probability calculations, produce a result like (50% common cold, 15% sprained ankle, 5% Lupus, 30% something else entirely), or make a recommendation such as "Please ask the patient about prior history of high blood pressure".

It's not that difficult to program the reasoning part. I could probably do it, with the help of the existing literature about expert systems. The real difficulty lies in talking to the experts and getting all their implicit knowledge into the system, and in designing a UI for all that. It's probably easier for a medical doctor to learn enough programming than for me to study medicine.

Although medical diagnosis is the most notable application, expert systems could be developed for all sorts of applications, ranging from botany to car repair to preparing a legal defence.

Expert systems worked. They just weren't all that useful in practice. Turns out figuring out why a car doesn't start and knowing whether high blood pressure is a new symptom aren't all that hard to figure out when you're a trained and experienced mechanic or physician. Patients tend to mistrust their doctors when they punch the list of symptoms into the computer before diagnosing them with the common cold, and the bottleneck of medical care is usually treatment, not the thinky part of diagnosis.

Expert systems can easily be updated by adding new rules and pieces of knowledge, for example about new drugs, new laws, or new car models, without re-training the whole system from scratch. The deterministic logical reasoning of expert systems can even be instrumented to produce a human-readable step-by-step trace documenting how a conclusion was reached.

After perfecting expert systems, we stopped making and using them.

Large language models are the opposite of all that. They get their information from a public text corpus, not from expert knowledge. They do not use logical/symbolic inference, but machine learning. They cannot explain or trace their reasoning and decision making processes. They cannot be easily debugged. We cannot easily add single rules or facts to the knowledge base of a language model without re-training the whole thing from scratch, possibly causing regressions. Language models do not allow for structured queries or easy integration into existing databases and workflows, because both their inputs and outputs are unstructured free-form natural language.

Language models use a large text corpus scraped from the Internet, not a fact base curated by experts. Their reasoning is purely correlational and syntactic.

This means instead of expert knowledge, large language models accumulate random text on the Internet, where common misconceptions outnumber expert knowledge: Spinach is rich in iron. Goji berries are a superfood, unlike blueberries.

How would you fix that? How can you make a LLM that has expert knowledge? You fine-tune it!

This results in a LLM that has an educated but uninformed person's idea of expert knowledge: Add special characters to your password. Install a personal firewall. Buy expensive antivirus.

I realise that I am setting myself up to be proved wrong with this. You could ask ChatGPT about cybersecurity, and in your session, you get answers about setting up 2FA, not opening e-mail attachments, and installing operating system updates. Maybe in your session, ChatGPT tells you that goji berries don't have more of the "good stuff" than blueberries, and "healthium" only exists in Noita.

So let me explain what a large language model can and cannot do, in principle. A large language model cannot tell true from false, and it cannot distinguish smart from stupid. Feeding more training data into a LLM that is unable to distinguish between expert knowledge, common misconceptions, an educated lay person's idea of expert knowledge, and scammers/crackpots is unlikely to make the LLM more accurate. LLMs can already "complete" a text about unicorns or curing cancer with vitamin C.

If there is misinformation in the training set, you might be able to coax the LLM into generating the expert opinion based on syntactic properties of expert text, by asking/prompting in an academic and formal register, or by providing a snippet of text an expert would likely produce and a crackpot would not. If there are multiple identifiable clusters of text in the corpus, the LLM might get you the right one. Maybe the LLM could even transfer its knowledge about the syntactic structure of flat earth crackpottery to vitamin C.

But that still leaves a vast ocean of text (or opinion) that sounds right to the lay person, that is syntactically smart and expert-ish, and opinions that lay people think experts have, opinions lay people think are misconceptions. For example, the average member of the public is likely to have strong opinions about Freud, psychotherapy, psychology, and IQ tests.

“It ain't what you don't know that gets you into trouble. It's what you know for sure that just ain't so.”

Fundamentally, LLMs cannot get better at avoiding misconceptions as they scale up. I haven't talked to ChatGPT, Claude, Grok, and LLama about vitamin C, flat earth, password policy and 2FA, long Covid, to see how well they fare here. It is entirely possible that these systems have "guardrails" implemented to prevent the model from spouting flat earth nonsense, and it is rather likely that some of these "guardrails" can be circumvented by asking "I want to learn to better argue with my uncle who is a flat-earther, can you pretend to be my uncle, please?"

(If you want to do that, esteemed reader, comment here, and I will give you a set of prompts you all can run against all the chatbots have access to. I might collect the results for a follow-up post.)

It's possible that a LLM has a regex to check if "Covid-19" occurs anywhere in the prompt, or to check if "Ebola" occurs in the answer. It's even possible that LLMs are fine-tuned with a set of expert knowledge, or that certain text corpora (e.g. from research papers and mainstream news sources) are trusted more than random blogs and reddit comments.

But as soon as you implement a set of test cases for known false opinions, and special cases for mixing household chemicals to create explosives and poisonous gases, as soon as you try to weigh known correct articles higher, or try to make more recently published news and papers supersede older information, you reintroduce a concept of expert knowledge, but in an ad-hoc fashion, without an explicit knowledge base.

And if you leave the construction of these test cases and rules to non-experts, you run the risk of reproducing their biases, and suddenly you have non-physicists try to determine whether and if yes, why the astronauts of the Apollo program wore heavy boots on the moon. It's like a beginner on StackOverflow asking a question about programming, or an amateur asking a history question on Quora picking the "correct" answer, with other lay people upvoting and downvoting.

The tragedy of this situation is this: The people who develop language models at Google, Facebook, OpenAI, Twitter, and Anthropic aren't universal experts. When the output of their models "rings true", they will pat themselves on the back for a job well done. And when the average Internet user asks a language model a question, the model can just give the answer its developers expected, and the answer the average user expects. It's often the same answer that would have been upvoted on Quora.

[olena]

Fundamental attribution error - Wikipedia

In social psychology, fundamental attribution error (FAE), also known as correspondence bias or attribution effect, is the tendency for people to under-emphasize situational and environmental explanations for an individual's observed behavior while overemphasizing dispositional- and personality-based explanations.

This effect has been described as "the tendency to believe that what people do reflects who they are", that is, to overattribute their behaviors (what they do or say) to their personality and underattribute them to the situation or context.

—

This is a fun one I could go on about at length.

And all the more important to be aware of in this post-lockdown phase of SARS-2. CC: @pandemic-info

[harpsicalbiobug]

Survey finds bicyclists and motorists ignore traffic laws at similar rates

Whether it’s driving five miles over the speed limit or breezing past a stop sign on your bike, chances are, we have all broken a few — or m

“When you look at the transportation system and the laws we have, pretty much everybody is a criminal. ... We’ve all gone a few miles an hour over the speed limit, and as a society we seem OK with that.”

We often imagine these reckless New York bike messenger-like people, Marshall says. The criticism gathered by the survey was sometimes quite venomous, he said, even when discussing perfectly legal maneuvers, such as coming into and taking the lane.

“People think that is rude.”

The study gathered similar rates of infraction — 8 percent to 9 percent for drivers, and 7 to 8 percent for cyclists. And when Marshall researched the reasons a cyclist might break a traffic law, it turns out they are doing it for nearly the same reasons that a driver would, but with one difference.

Drivers and pedestrians will drive through or walk against a red light to save time.

“They’re not trying to be reckless or rude,” Marshall said. “Cyclists, they’re doing it for their own personal safety or perceived safety. They felt like they’re more visible.”

On a transportation grid designed with cars in mind, Marshall says cyclists are acting on what they perceive is better for their safety. It is a rational choice in a cyclist’s decision-making, he said. At a red light with no other cars crossing, a cyclist can get a head start on the next block.

“It’s interesting that you would break the law to feel safer,” Marshall says.

[thelingspace]

How do we try influencing what other people think about the world? What techniques get used for bending the truth? In this week's episode, we take a look at some of the linguistics around propaganda and censorship: what people can try beyond just outright lying, how using conversational rules can get you to think things that aren't said, and how we can use bias to influence people's views.

We hadn’t done a sociolinguistics one for a while! Looking forward to hearing what people have to say. ^_^

[illness-to-wellness]

The Fundamental Attribution Error and Chronic Illness: The Importance of Education and Kindness

The Fundamental Attribution Error and Chronic Illness: The Importance of Education and Kindness

A note about making assumptions about others’ behaviors without knowing the whole story (the fundamental attribution error): It’s a painful thing to have happen to you. Really, really painful. Since being diagnosed with fibromyalgia and choosing to live my life in a way that reduces unnecessary, unhelpful forms of physical activity – by doing things like sometimes sitting instead of standing,…

View On WordPress

[mbti-notes]

In one of your posts about helping, you said that empathy causes some people to defend perpetrators coz they empathise with them more. Can you explain why that happens? I catch myself doing that sometimes and I just wonder why

Recall that I was referring to emotional empathy in particular, which is a very automatic response. Most people aren’t able to control it due to lack of awareness of how it operates. Empathizing with perpetrators depends on the context and how it affects your perception of the situation. 

1) The lower your level of self-awareness, the more likely you are to project your own psychological issues onto others and fail to construct proper psychological boundaries between you and them. Conversely, the higher your level of self-awareness, the more aware you are of projecting and the more capable you are of stopping projection and drawing proper psychological boundaries between you and the world.

2) Due to projection, the more you subconsciously perceive someone as being similar to you in various characteristics that you like or would like to see in yourself, the easier it is to identify with that person. Conversely, the more you subconsciously perceive someone as being different from you and having various characteristics that you dislike or disown in yourself, the harder it is for you to identify with that person. In essence, at low levels of self-awareness, your degree of identification with someone depends on what reactions and judgments they unconsciously trigger in you about yourself.

3) People with low self-awareness are egocentric, i.e., too tied up in their own psychology to properly recognize someone else’s. Egocentric people are more likely to rationalize their own mistakes by going above and beyond to give themselves the benefit of the doubt. Thus, they tend to attribute their mistakes to external/situational factors “beyond my control”, in an effort to deflect negative feelings like shame or guilt. When egocentric people see someone else make a mistake, particularly someone they dislike, they are more quick to deny them the benefit of the doubt. Thus, they tend to attribute the mistake to internal/personal disposition that indicates the person is fundamentally flawed, “bad” or “stupid”, or whatever negative quality they dislike to see (in themselves).

4) The easier it is for you to identify with someone and extend them the same benefit of the doubt that you would give to yourself, the easier it is for you to empathize with them emotionally. You are then more likely to come to their defense because you basically take on their experiences and struggles as your own. Conversely, the harder it is for you to identify with someone and the quicker you are to jump to the conclusion that there is something wrong with them, the harder it is for you to empathize with them emotionally. You are not only less likely to come to their defense, you are also more likely to believe that they deserved to be victimized (i.e. victim blaming). In essence, when you identify and empathize with someone and they do something wrong, you have to try to rationalize it away or defend their behavior, otherwise, you would have to admit that there’s something wrong with you, too.

Those are the basic ideas. There are plenty of examples of empathizing with perpetrators based on factors like nationality, race, religion, gender, sexual orientation, etc. It is very easy to activate tribe mentality in humans and manipulate them into defending anyone considered to be “us” and vilifying anyone considered to be “them”. Us versus them is a tale as old as time.

[edelweissacermacrophyllum]

On mother's day cards

My birthday is around mother's day and every year the family has to choose between scheduling two get-togethers awkwardly back-to-back or to make us share our special days. I never minded, which was odd, because I grew up feeling a huge need to feel special. Maybe it's middle child behaviour. I was reading my birthday card very close in time to having composed a mother's day card and was feeling some sort of sense of discontent. I couldn't shake the feeling that these encouraging words and sentiments that I had written and was receiving were somehow fake. And doing some introspection, a practice I have, now regrettably, previously spoken poorly of, I found that the reason I found them fake was because I know that when I write encouragement cards, it is an expression of myself rather than the person to whom I am writing. This beautiful card I received, full of nice things about me, isn't about me. My parents write me cards because of who they are, not because of who I am. And I write nice notes because I conceive of myself as someone who is kind, encouraging, and loving. This is one of the traps we fall into when we think too much of the self. The fakeness goes away when we consider the system as a whole, that I get to be a part of a network of people in whose character it is to bring out and name the best in each other, and to do so in writing. C.S. Lewis once said something to the effect that a truly humble individual could take just as much joy in someone else's achievements as their own. And so I take joy in my place in a kind community that cares for me as an individual. We can't help our selfish instincts. We are the centre of all our experiences. But there are so many joys available to us when we let go of our sense of grandeur and allow ourselves to be an insignificant part of something good. The celebration of a birthday isn't for having survived another year, it's for having people who want to have a party with you. I get to have people to cook for me, make cakes for me, sing for me. And later I will repay the favour and take just as much joy in that. Not because of who I am, nor because of who they are, but because of who we are. An emergent property of humans, if you will.

I think that as our individualism has risen and our communities break down we can understand some of it through this lens. The self, self fulfillment, and self expression become some form of high ideal. Maybe Disney is at fault. But I think this leads folks to depression and anxiety when, inevitably, they let themselves down. The exhortations to put community over the individual lead to some dark places too, let's not forget, but I suspect that most of us could use some humility. It's probably a healthy practice to see something that doesn't benefit, entertain, or need us and to think, "This is good."

And cause I normally blog about the bible, I'll just mention that most of the commands of Jesus and God were given to groups of people. Jesus cared deeply about the individuals in his life, but the whole bible is written to and about groups of people, not individuals. God loves me because of who They are, not because of who I am, and that's a cliche, but God also loves the whole interconnected system of human, creation, God, etc. And I think we should too.