Calling Bullshit 2/5
Part 1.
3/ The Nature of Bullshit
Bullshit is not outright or white lies. It’s what people create when they try to impress or persuade you, without any concern about the truthfulness or correctness of what they say. In science it can mean the text that’s so obscure that it’s impossible to get to the main point without permanent brain damage.
Test: if you can negate a sentence and its meaning doesn’t change, it’s definitely obscure and is bullshit.
Persuasive bullshit – conveys an exaggerated sense of competence or authority.
Evasive bullshit – avoids directly answering a question due to the unpleasant truth.
Bullshit disregards the truth by overwhelming, distracting or intimidating the listener. Adding unnecessary references to an article (if you question me – you also question other authoritative figures, who do you think you are?), using jargon that doesn’t fit a reader’s vocabulary – this is all bullshit.
Some things are easier to challenge than others. There are basic questions any curious person should be able to ask: explain the choice of the sample group, what specific method to arrive at the conclusion is being used, etc. Some things are very hard to verify independently (DNA sequencing, COVID treatments or climate change claims), hence the reader has to eat off the author’s hand, as disbelief will be treated as a challenge of the scientific establishment. Claims like this rely on an almost unverifiable black box.
Data is being fed into the black box, which does its unverifiable magic, leading to an output that later gets interpreted by a researcher. Most problems with black boxes are related to either the data (garbage in – garbage out) or the model / black box. Sometimes the output is completely off the logical range of values – even then sometimes this is clouded by the sophistication of the black box.
Using the right data as an input is critical (to state the obvious), but time and time again models are being trained on poorly collected and/or cleaned data leading to all sorts of self-delusion (and billions of VC dollars being wasted). If a team of PhDs builds a perfect model, but the data is being collected by $1.5/hr contractors, it’s a dictionary definition of stupidity and carelessness.
The book tells a story about a system to tell a criminal from a good citizen by their photo; the criminals’ photos were mugshots provided by the police, and the good citizens provided their professional photos showing them in the best light. What could go wrong??? The system ended up being an expensive smile detector.
Let’s repeat the ultimate truth: extraordinary claims require extraordinary evidence.
3/ The Nature of Bullshit
· Bullshit is not outright or white lies. It’s what people create when they try to impress or persuade you, without any concern about the truthfulness or correctness of what they say. In science it can mean the text that’s so obscure that it’s impossible to get to the main point without permanent brain damage.
· Test: if you can negate a sentence and its meaning doesn’t change, it’s definitely obscure and is bullshit.
· Persuasive bullshit – conveys an exaggerated sense of competence or authority.
· Evasive bullshit – avoids directly answering a question due to the unpleasant truth.
· Bullshit disregards the truth by overwhelming, distracting or intimidating the listener. Adding unnecessary references to an article (if you question me – you also question other authoritative figures, who do you think you are?), using jargon that doesn’t fit a reader’s vocabulary – this is all bullshit.
· Some things are easier to challenge than others. There are basic questions any curious person should be able to ask: explain the choice of the sample group, what specific method to arrive at the conclusion is being used, etc. Some things are very hard to verify independently (DNA sequencing, COVID treatments or climate change claims), hence the reader has to eat off the author’s hand, as disbelief will be treated as a challenge of the scientific establishment. Claims like this rely on an almost unverifiable black box.
· Data is being fed into the black box, which does its unverifiable magic, leading to an output that later gets interpreted by a researcher. Most problems with black boxes are related to either the data (garbage in – garbage out) or the model / black box. Sometimes the output is completely off the logical range of values – even then sometimes this is clouded by the sophistication of the black box.
· Using the right data as an input is critical (to state the obvious), but time and time again models are being trained on poorly collected and/or cleaned data leading to all sorts of self-delusion (and billions of VC dollars being wasted). If a team of PhDs builds a perfect model, but the data is being collected by $1.5/hr contractors, it’s a dictionary definition of stupidity and carelessness.
· The book tells a story about a system to tell a criminal from a good citizen by their photo; the criminals’ photos were mugshots provided by the police, and the good citizens provided their professional photos showing them in the best light. What could go wrong??? The system ended up being an expensive smile detector.
· Let’s repeat the ultimate truth: extraordinary claims require extraordinary evidence.
4/ Causality
Faking social confidence is claimed to be a helpful skill, especially for teenagers. It’s not considered bullshit per se. There is association between self-confidence and the age of a first love kiss, but it’s irresponsible to think that one leads to another (both ways); these things can have a common root. Our inbuilt desire for making sense of the world via pattern recognition can backfire.
We’re interested in causation in a practical sense: what do we need to do to cause something to happen, change or not happen?
People’s brains are wired to make sense of the world, trying to fit all the available information into a believable story containing cause and effect. Bullshit happens when this cause and effect is based only on correlation. Media may not care what readers believe after reading bullshit articles, they need sales and/or clicks. Even medical journal articles (which are supposedly the gold standard of published content) suggest causation in the absence of adequate evidence.
Prescriptive or causal claims are the worst: if people A do X, it doesn’t mean that doing X turn people B into people A. [MK: just look at all the “rules of successful people”.] Is there believable causal evidence to back the prescriptive claim?
There are examples that are of interest to most of us: there is in fact an association between poor sleep and Alzheimer’s disease. What’s not proven is the correlation. Another one is “red wine prevents heart disease” – hits very close to home.
It’s important not to confuse trends (where minimum wage is higher, poverty is lower) with prescriptions (let’s increase minimum wage to reduce poverty).
The (in)famous marshmallow test (a child can have a marshmallow now, but if s/he waits some time, s/he can have two; the kids choosing the latter option were more likely to be successful in life) simply demonstrates a correlation but doesn’t suggest causation. [MK: I’ve seen numerous instances when this causation was firmly assumed.] Delaying gratification (and alienating children in the process) is not the right course of action regardless of what magazines suggest. This particular test had a huge input data problem: it didn’t account for the socioeconomic status of the children’s families. The value of a single marshmallow and the certainty of obtaining one is vastly different for children who can have sweets any time they want and those for whom candy is rare and precious. [MK: I’ve just run this experiment with my 9-year-old daughter and she chose to wait. Maybe because she’d had a couple of pieces of chocolate an hour earlier.]
Correlation doesn’t need to be +1 or -1 all the time. There are probabilistic causes (A increases the chances of B in a causal manner), sufficient causes (if A happens, B always happens – that’s what I meant by correlation being -1 or +1), and necessary causes (B won’t occur without A present). Bullshit happens when sufficient and probabilistic causes are confused for one other.
MK: the book goes on to explain the selection bias and random sampling. Sometimes random sampling (say, for life-threatening diseases) is unethical and can’t be used, but it’s true that not taking medication to block fever (Nurofen, Panadol, etc.) decreases the time to recovery. I’d add that this conclusion ignores the aspect of the patient’s comfort, which may be more important than the time one spends in bed sick.
Part 3.