|
一般方法論 -- 開欄文
|
瀏覽3,769 ｜回應19｜推薦2 |
|
|
|
子曰：「工欲善其事，必先利其器」(《論語•衛靈公10》)；「器」在這裏指的大概是工具。但是，要把一件事「做對」，「工具」之外，還得講究「方法」。「雖不中不遠矣」 (《大學章句10》)，講的應該是「態度」；這個道理用在「方法」上也可以說得通。這是「科學基礎論」中「科學方法」研究的對象。
大概在初中時，家父給了我一本討論「科學方法」的書；書名已經忘了。這是我第一次接觸到這個主題；自然印象深刻。後來成長過程中，我對它一直非常注意。「邏輯學」之外，我還讀過笛卡爾和波普的書。本部落格過去登過不少這方面的評論。
循此處各《開欄文》之意，另立此欄。
本文於 修改第 4 次
|
|
愚蠢、愚蠢人、和愚蠢行為 -- David C. Krakauer
|
|
|
推薦1 |
|
|
|
下文放在「人工智慧縱橫談」一欄也適合。放在這裏是因為：用了「錯誤」的「方法」，或把「方法」用在「錯誤」的場合，都會造成「愚蠢」的結果。
What Makes Humans Stupid
It takes intelligence to get things spectacularly wrong. An essay on our undoing.
David C. Krakauer, 07/08/26
I have never heard a rock described as stupid. And the same would be true of a river, a hurricane, and even a thermostat. Stupidity seems to be a sophisticated form of behavior despite its ignominious associations.
Human beings can land autonomous rovers on Mars, sequence a genome in hours, and engineer nanometer circuits. And yet conspiracy theories, anti-scientific political movements, and institutional hatred proliferate on a scale that might embarrass the meager success record of a medieval alchemist.
One might say that stupidity implies a capacity for getting things right before it can get them spectacularly wrong. Stupidity is not the opposite of intelligence but its evil twin, the dissimulating Cain to a cerebral Abel.
And perhaps surprisingly, the degree of stupidity available to any system scales directly with the intelligence that system possesses—more intelligence begets greater feats of stupidity. It would be a stretch to call a bacterium stupid, and we know that cats and dogs achieve modest feats of it. But human beings, equipped with language, abstraction, technology, institutions, and ideology, can be stupid on a truly civilizational scale. This is not a joke; it is close to a law of nature. A law that might very well be our undoing.
We have thousands of research programs on intelligence, and not all of them are intelligent, including studies of IQ, AI, animal cognition, and collective problem-solving. These take place in celebrated departments and are published in prestigious journals devoted to understanding how minds make hard problems easy.
These days we cannot take a step without crashing into another article on intelligence and AI. Researchers from all fields without any knowledge of intelligence research and its history have become self-declared “thought leaders” in natural and artificial intelligence.
Yet stupidity gets almost no attention at all. It is treated as a mere absence, as if once we subtract intelligence what remains is stupidity. It is likened to a form of psychological darkness experienced after you have switched off the lights of deliberation. But this is a mistaken belief.
Darkness does not do anything pernicious in the way that stupidity does. Stupidity takes an easy problem and, with great effort and misdirected ingenuity, makes it hard. That effort is the key to grasping stupidity, in that you need sophisticated machinery to be genuinely, consequentially stupid.
Here is how I like to think about this from a scientific perspective. If intelligence means making a problem of difficulty X easier, by deploying tools, using mathematics, and adopting strategies that reduce its cost, then stupidity means making a problem of difficulty X harder.
And contrary to expectations, the most reliable way to make an easy problem hard is to bring to bear an impressive apparatus of complicated theories, elaborate beliefs, and sophisticated algorithms that sound tremendously convincing but perform worse than doing nothing.
A person who does not know the answer to a question is merely ignorant. A person who constructs an ingenious hundred-page argument for the wrong answer is stupid. As great writers, artists, and philosophers throughout time have understood, such constructions require intelligence of a high order.
Consider this analogy. When a meteor collides with the Earth, there is no sense in which it is doing anything wrong. It is obeying classical mechanics, and that is all there is to say about it. But when a bird flies into a glass window, something different has happened, something we can with justification call an error.
Life, uniquely among physical systems, has the capacity to be wrong. Stupidity is a very special species of failure and not all errors qualify. People err from ignorance (insufficient data), from noise (a signal is distorted), or from the honest misapplication of a rule to a novel situation and changing context. These are the misfortunes of our lives about which we should be forgiving.
Stupidity begins where error is elaborated, defended, refined, institutionalized, and made the foundation for further action. Stupidity makes everything progressively worse. And it is this elaboration of erroneous belief and behavior that demands, paradoxically, the prior existence of intelligence.
Some of the most impressive feats of stupidity consist in applying a perfectly good theory to the wrong problem.
Some would argue that quantum mechanics is one of the most “intelligent” inventions of physics, one that provides a framework of extraordinary precision for describing the behavior of subatomic particles, despite its truly counter-intuitive requirements.
But there is a small cottage industry of thinkers who have tried to apply quantum mechanics to human consciousness, decision-making, and psychology to explain why people are indecisive and why they eventually make up their minds. There is nothing wrong with the mathematics and the physics is sound. But the application is peculiar.
What was an elegant and parsimonious description of photons and electrons becomes, when imported into psychology, an absurdly over-complicated way of saying that people sometimes change their minds. A phenomenon that a novelist could illuminate in a paragraph has been buried under a formalism designed for a completely different scale of reality.
The theory did not become less intelligent, it was asked to solve a problem it was never designed for, and in the process, it made that problem harder to understand, not easier. This flies in the face of the purpose of science, which is, as the physicist and philosopher Ernst Mach described it, “the completest possible presentment of facts with the least possible expenditure of thought.”
The early cybernetics movement offers a similar cautionary tale. Norbert Wiener and his colleagues at MIT developed a powerful framework for understanding feedback, control, and communication in machines and organisms. The mathematics was original, the engineering applications impressive, and the cybernetic enterprise even provided the groundwork for the development of complexity science.
Then the management consultant Stafford Beer and others tried to apply cybernetic control theory to the management of entire national economies, most famously in Salvador Allende’s Chile, where Project Cybersyn attempted to run the Chilean economy through a network of telex machines. The idea was to model the economy as a dynamical system with inputs and outputs, and to use real-time feedback to optimize production and distribution.
Unfortunately, an economy is not a servomechanism. The feedback loops in a national economy involve millions of adaptive agents with private information, conflicting goals, and a tendency to evolve their preferences. The seduction of applying cybernetic methods to complex systems is described beautifully by my Santa Fe Institute colleague, the writer Francis Spufford, in his novel Red Plenty:
“The world was lifting itself up out of darkness and beginning to shine, and mathematics was how he could help. It was his contribution. It was what he could give, according to his abilities. He was lucky enough to live in the only country on the planet where human beings had seized the power to shape events according to reason.”
A simple behavioral intervention, such as adjusting a price, changing a regulation, or God forbid, simply asking people what they need, could often have achieved in an afternoon what the cybernetic apparatus was struggling to model in weeks.
The intelligence of the cybernetics as a framework for control of simple systems is unquestionable, but its application to the economy inflated the difficulty of the problem it was meant to solve.
The Austrian novelist and essayist, Robert Musil, in a 1937 lecture, “On Stupidity,” described what may be the most important distinction in the whole literature of stupidity.
He separated “honorable stupidity,” which is a simple cognitive limitation, or the inability to grasp a difficult argument, from “intelligent stupidity,” which he considered far more dangerous.
This insight is the central concern of my favorite novel of Musil’s, The Man Without Qualities. Intelligent stupidity marshals all the resources of the intellect in the service of an error. It is not the failure to think; it is thinking flamboyantly and systematically in the wrong direction.
A student who cannot follow a mathematical proof is not stupid in Musil’s sense. A math professor who builds an elegant theoretical edifice to defend a proposition a child could see is false is intelligently stupid.
Musil was not alone in taking stupidity seriously. Novelists have long understood the intelligence-stupidity relationship with a clarity that most scientists choose to ignore. Perhaps because fiction can show the process by which an intelligence can devour itself, a rather unpalatable spectacle to a rationalist.
Jonathan Swift’s Gulliver’s Travels describes the Academy of Lagado, whose researchers deploy a variety of elaborate experimental methodologies. One is extracting sunbeams from cucumbers while another softens marble into pillows. A third colleague breeds naked sheep. Swift’s satire is obviously targeted at a variety of forms of misdirected systematic inquiry.
These are forms of intelligence trapped in frameworks so rigid that they produce outcomes that are worse than doing nothing.
In the sky city of Laputa, brilliant mathematicians and accomplished musicians apply projective geometry in order to build with no right angles and chefs prepare meals based on geometric figures. These are abstractions of exquisite subtlety and power in mathematics that become helpless when directed at the wrong practical problems.
William Gaddis in his The Recognitions presents a society of forgery, misattribution, and counterfeiting in which enormous ingenuity is expended in the service of inauthenticity. The protagonist, Wyatt Gwyon, produces forged Flemish paintings that require more skill and knowledge than original compositions. His forgeries are technically masterful, art-historically impeccable, and completely fraudulent. Each of his many characters talk past one another in dialogues of escalating misrecognition, deploying considerable verbal intelligence to deepen general confusion.
Thomas Pynchon’s Gravity’s Rainbow describes a vast bureaucratic and military apparatus of World War II that functions as a machine for converting rational planning into catastrophe. Pynchon’s cartels and rocket engineers are not unintelligent, quite the opposite, and their competence is the V-2 rocket, that threatens to annihilate the war-ravaged cities of the West. Pynchon describes a world in which institutional intelligence becomes a form of collective stupidity.
A number of philosophers have sought to provide a framework for understanding this perverse phenomenon. Erasmus, in 1511, in The Praise of Folly, suggests that folly is the engine of human accomplishment. Without self-delusion and overconfidence, nothing would ever get attempted. The challenge for Erasmus was not whether a civilization produces stupidity, but whether it will produce the kind that can be survived.
Dietrich Bonhoeffer, writing from a Nazi prison in 1943, suggests that stupidity is not a cognitive defect but an imposed sociological structure. People under the spell of power tend to surrender their capacity for independent judgment and become “stupid” instruments.
And Carlo Cipolla, an Italian economic historian, in the Basic Laws of Human Stupidity published in 1976, defined a stupid person as someone who causes losses to others while deriving no gain, or even harm, for themselves.
Cipolla was daring enough to propose that the proportion of stupid people is constant across all populations, from professors, plumbers, generals, and janitors. I would suggest, in line with the speculations of Swift and Musil, that it only gets worse with complication.
If stupidity is costly to its practitioners why does natural or perhaps even cultural selection not eliminate it from populations of organisms? One obvious possibility is a change in the environment that renders a previous behavior obsolete.
For millions of years, maintaining a fixed angle to a distant celestial light source, the moon and the stars, proved to be a vital navigational heuristic across the animal world. It is efficient, reliable, and requires minimal neural hardware.
Once humans invented artificial illumination, a strategy that had worked for eons became suicidal. A moth spiraling into a candle flame is not failing to navigate, it is using a time-tested rule-of-thumb rendered catastrophically wrong by a change in context. Intelligence and stupidity turn out to be chronometrical. A brilliant heuristic and a fatal one can be the same heuristic, separated by an unexpected shift in the world.
Sea turtles have followed moonlight to the ocean for 100 million years and now they orient toward the artificial illumination of mushrooming beachfront condos. Albatross have evolved to scoop fish from the ocean surface and now unknowingly collect floating plastic and feed it to their chicks.
The fungus Ophiocordyceps infects carpenter ants and coopts their behavioral circuitry. An infected ant climbs to a precise height on a plant stem, clamps its mandibles onto the underside of a leaf, and dies, ideally positioned for the fungus to disperse its spores. The ant’s behavioral program has been hacked. The ant is performing a sophisticated sequence of actions, including climbing, orienting, and gripping in the service of another organism. This is an ant perfectly captured by Bonhoeffer’s idea of competence co-opted by a malicious agent to solve the wrong problem.
Perhaps specialization produces local intelligence that when overextended can generate global stupidity. This would be a change in scale and domain that corresponds to the moth’s change of environment.
The physicist Lord Kelvin marshaled all his physics knowledge to prove that heavier than air flying machines are impossible less than a decade before the Wright brothers flew a heavier than air airplane.
Nikola Tesla rejected quantum mechanics and the theory of relativity and argued that future human civilization would run on the energy of the Earth through “earth resonance.” And Percival Lowell used his telescopes to map “canals” on the surface of Mars that he claimed were alien irrigation ditches.
These are all examples of the overcommitment, and overdevelopment of an idea, far from the territory in which an idea once grew and flourished or from which it was unceremoniously banished.
Artificial intelligence is by design the most powerful cognitive artifact ever created. It is engineered to minimize user effort by performing tasks that humans would otherwise find time consuming or impossible.
The problem of course is that the better the tool gets the less the user needs to think for themselves. And in a vicious spiral, the less the user thinks the more dependent they become on their tools. That is until the tool disappears and the whole system collapses.
If intelligence is a necessary precondition for stupidity, and intelligence and stupidity scale together such that it takes real intelligence to be spectacularly stupid, then super-intelligence will be the opening act to an era of super-stupidity.
AI hallucination might be the first evidence of this dynamic. Large language models produce fluent, confident, detailed text that is, with some regularity, factually wrong. And this is not a simple bug but a structural feature of systems that optimize for appeal and plausibility rather than truth.
And the danger is not that the AI will be wrong, after all, humans are wrong all the time, but knowing this, humans have invented means to detect and correct errors. We call this the scientific method.
The danger is that an AI will be wrong in ways humans can no longer detect because the very capacities that would catch the error have been outsourced to the machine or exceed the capacities of human minds.
We face the prospect of a stupidity so sophisticated that it becomes indistinguishable, to its beneficiaries, from intelligence. This is the parable of Douglas Adams’ The Hitchhiker’s Guide to the Galaxy, where the answer to the ultimate question, the meaning of life, the universe, and everything, is 42.
I would like to make a modest proposal and suggest that we need a science of stupidity as rigorous as our emerging sciences of intelligence.
This will not require billions of dollars of investment.
It would involve inquiries into the mechanisms by which intelligent systems produce stupid outcomes.
It would include studying the evolutionary dynamics that maintain stupidity despite its selective costs.
It would promote the development of design principles that distinguish tools which enhance cognition from tools which replace it.
And it would include surveying the institutional conditions under which collective intelligence degrades into collective stupidity.
Stupidity is not what remains when intelligence is subtracted, it is an active mechanism with its own logic, its own dynamics, and a capacity for unbounded growth parasitic on ingenuity. In a world obsessed with ever more powerful cognitive technologies, understanding stupidity is not merely an academic exercise, it might prove to be the most intelligent thing we do.
Related Reading
Adams, D. The Hitchhiker’s Guide to the Galaxy Pan Books, London (1979).
Bonhoeffer, D. After ten years. In Behtge, E. (Ed.) Letters and Papers from Prison SCM Press, London (1953).
Cipolla, C.M. The Basic Laws of Human Stupidity Il Mulino, Bologna, Italy (1976).
Erasmus, D. The Praise of Folly (1511); translated by Miller. C.H. Yale University Press, New Haven, CT (1979).
Gaddis, W. The Recognitions Harcourt, Brace & Company, New York, NY (1955).
Musil, R. On stupidity. Lecture delivered in Vienna (1937). In Pike, B. & Luft, D.S. (Eds.) Precision and Soul: Essays and Addresses University of Chicago Press, Chicago (1990).
Musil, R. The Man Without Qualities (1930–1943); translated by Wilkins, S. & Pike, B. Alfred A. Knopf, New York, NY (1995).
Pynchon, T. Gravity’s Rainbow Viking Press, New York, NY (1973).
Spufford, F. Red Plenty Faber and Faber, London (2010).
Swift, J. Gulliver’s Travels Benjamin Motte, London (1726).
Lead image: jdoms / Adobe Stock
David C. Krakauer is the president and William H Miller Professor of Complex Systems at the Santa Fe Institute.
本文於 修改第 2 次
|
|
統計架構：古典式、頻率式、和貝斯式 -- Leandre Sabourin
|
|
|
推薦1 |
|
|
|
統計學雖然算是一門「專業領域」，但在人人都對「人工智能」朗朗上口的當下，多點常識總有用得上的時刻。對數學、數字、或公式敏感的朋友，不妨跟上一篇「笑話」同時看來調劑、調劑。
Differencing the Three Statistical Framework : Classical, Frequentist and Bayesian
Leandre Sabourin, 01/02/26
Bayesian statistics is an interesting branch of statistics, that is less used, but is gaining in popularity in many different aspects. Although less common because of its mathematical complexity, it is now gaining a lot of momentum accross many different fields, such as data science, medicine, artificial intelligence, etc.
Some software (like JASP) offer the possibility to perform traditional statistical test like the ANOVA into the frequentist or bayesian approach. In this article, we are going to set the basis on what is distinguishing this approach, compared to traditional statistical framework. This will set the tone for future articles where I will take a closer look at Bayes’ theorem.
0. Statistical Framework
In statistics, there are three main framework used to define the probability of an event to happen. Probability is a numerical measure of how likely an event is to happen. There are three principal framework by which you can do statistics. Those framework are (1) the classical approach, (2) the frequentist approach and (3) the Bayesian approach.
1. The Classical Approach
In the classical framework, outcomes that are equally likely have equal probabilities. For example, if you roll a six-sided dice, your probability or rolling a four is 1 in 6. Similarly, your probability of rolling a 2, 3 or 5 is the same.
The classical approach expresses probability as the ratio between the number of favorable outcomes and the number of equally possible outcomes in the sample space. The formula for this framework can be seen as :
P(A) = Number of favorable outcomes/Total number of possible outcomes
Classical approach
2. The Frequentist Approach
The frequentist approach define the probability of an event to happen based of the frequency of this outcome in a sequence of event defined a priori. The frequentist approach is the one mostly used in research nowaday because most of the statistical test rely on this approach.
In essence, the role of researchers and statistician is to recruit a sample of participant that is big enough that they will be able to (1) find a significant effect and (2) infer the results from this sample to the rest of this population. The formula for this framework can be seen as :
P(A) = Number of times event A occurs/Total number of trials
Frequentist approach
There are two subjective components to this approach, the first one being the choice of the statistical test and the it’s components (i.e : post-hoc tests, parameters, etc). To ensure a good analysis, researchers have to justify their choice and have a methodological approach to their experiment. Otherwise, they could attempt to ‘hack’ their test by optimizing the parameters that will allows them to find an effect, without there being an actual one.
The second one is about the number of participants. There isn’t a rule for the amount of participant you have to recruit in your study. Although researchers can do a power analysis to see at what number of participant they are most likely to find a significant effect (more detail about this here), this analysis is still based on a subjective number of participant defined by earlier research in the scientific literature.
3. The Bayesian Approach
The third framework and the one we will be most interested today is the Bayesian approach. In this framework, we are incorporating a prior knowledge and beliefs about an event to happen. Instead of having to rely of a sample size like the frequentist approach, the subjective components of this framework relies on our belief of an event A to happen given an event B. This means that different people can have different probabilities for the same event based on their knowledge and experiences.
One of the strengths of the Bayesian approach is its ability to update beliefs over time. With new evidences coming over time, the initial belief (prior) we had can be revised into a new belief (posterior) For example, if we are rolling a six-sided dice, we would hold the assumption that we have 1/6 chance of getting any of the numbers from the dice. However, if we do 100 attempts, we realize that we are getting the number three 90 times out of these attempts. This new evidence will dramatically change our belief about the fairness of this dice.
Conditional probability
The Bayesian approaches uses principles from conditional probability to predict that an event A will happen given an event B. It updates our knowledge about one variable when partial information about another is known
P(A〡B) = P(A and B)/P(B)
P(A ∣ B): probability of A given B
* P(A and B): probability that A and B happen together
* P(B) : probability of B happening
*Note : the sign ‘∩’ (原文所用符號；此處以 “and” 代替) “means ‘ intersection (“and”), where ∪ represents union (“or”)’. So in that case, P(A ∩ B) refers to the probability of A and B to happen together (i.e : the probability that someone is a math major and a male in a sample).
The main formula used in the Bayesian framework to express that conditional probability is through the Bayes’ theorem, which allows a way to update our beliefs about the probability of an event based on new evidence. In simple terms, it helps us calculate the likelihood of an event happening after we have some additional information.
P(A〡B) = P(B〡A)P(A)/P(B)
This formula expresses how the prior probability P(A) is transformed into a posterior probability P(A∣B) once the evidence B has been taken into account.
Summary
In this article, I presented the three statistical framework used nowadays as well as how probability is being represented. This can be summarized in the following table :
三種統計架構比較表 (請至原網頁查閱)
As we can see, they all have a radically different way to see the probability of an event to happen. understanding the difference between the three will set the tone for futur articles on the Bayesian Statistics. Consider subscribing if you learned something new today, and want to receive notification when I publish a new article!
Written by Leandre Sabourin
Psychology M.Sc. student writing about statistics and psychedelics science
本文於 修改第 1 次
|
|
有關「統計學」的笑話10則 -- Md Johirul Islam
|
|
|
推薦1 |
|
|
|
下文雖然以「笑話」命題，其主旨則在於：
* 釐清「統計」和「統計數字」的概念及意義；
* 說明應用「統計」的正確方式與場合；
* 幫助一般人了解「用數字說謊」的欺騙行為。
故置於此欄。
10 Statistics Jokes That Are Funny… Until You Understand Them
Md Johirul Islam, 06/16/26
Statistics is one of the few subjects where a sentence can sound harmless, a chart can look convincing, and an average can quietly ruin your life.
That's also what makes statistics jokes so good.
They're not just random punchlines — they play with uncertainty, data, averages, probability, false conclusions, and the weird ways humans misunderstand numbers.
If you've ever laughed at the phrase "correlation doesn't imply causation", this post is for you.
Let's break down 10 statistics jokes that are funny… until you realize they're also uncomfortably true.
1. "A statistician drowned crossing a river that was, on average, 3 feet deep."
Why it's funny
This is one of the most famous statistics jokes because it exposes a classic mistake:
An average does not tell the whole story.
Even if the average depth of the river is 3 feet, some parts could be:
* 1 foot deep
* 2 feet deep
* 10 feet deep
And if the person steps into the deep section, the average suddenly becomes useless.
The joke is funny because it shows how relying on averages blindly can be dangerous.
2. "Correlation does not imply causation… but it sure gets published a lot."
Why it's funny
In statistics, correlation means two things move together.
For example:
* ice cream sales go up
* drowning incidents also go up
That doesn't mean ice cream causes drowning.
A third factor — like summer weather — may explain both.
The joke is funny because people, media, and sometimes even researchers love seeing a correlation and jumping straight to a cause.
It's basically a joke about how humans misuse statistics with great confidence.
3. "My data scientist friend says I'm below average. I told him that's mean."
Why it's funny
This joke works because of the double meaning of mean.
In statistics:
* mean = average
In normal language:
* mean = unkind
So if someone says you're below average, calling that "mean" works in both senses.
Simple, clean, nerdy wordplay — exactly the kind statisticians love.
4. "If your head is in the oven and your feet are in the freezer, then on average you're comfortable."
Why it's funny
This is another joke about the danger of averages.
If you average two extreme conditions:
* very hot
* very cold
you might get something moderate.
But no real person experiences an average in that situation — they experience both extremes at once.
The joke is funny because it shows how an average can hide reality instead of revealing it.
5. "The plural of anecdote is not data."
Why it's funny
This line is funny because it sounds like a grammar correction, but it's really a statistics lesson.
An anecdote is a single story:
* "My uncle smoked every day and lived to 95."
* "My friend never studied and still got an A."
But isolated examples are not enough to draw reliable conclusions.
Statistics depends on:
* larger samples
* patterns
* evidence
* uncertainty
The joke is really a warning: one dramatic example is not the same as data.
6. "Never trust statistics you didn't fake yourself."
Why it's funny
This is a twist on the famous cynical line:
“There are lies, damned lies, and statistics.”
It jokes about how statistics can be manipulated by:
* cherry-picking data
* choosing misleading graphs
* using biased samples
* reporting only favorable results
Of course, the joke is intentionally absurd — you obviously shouldn’t trust fake statistics either.
It's funny because it exaggerates a real fear: numbers can look objective even when the story behind them isn't.
7. "Statistically speaking, the average person has one breast and one testicle." (男人數 + 女人數)/2
Why it's funny
This joke is shocking, but mathematically it makes sense if you average anatomy across a whole population.
If you combine everyone together and compute certain averages, you can end up with statements that are technically true in aggregate but ridiculous when applied to an individual person.
* That's the core joke:
* the statistic may be numerically valid
* but it becomes absurd when interpreted literally
It shows that population averages are not personal descriptions.
8. "I'm great at statistics. I can make numbers say anything I want."
Why it's funny
This joke points to a real concern: statistics can be abused.
You can manipulate interpretation by:
* truncating the y-axis on a chart
* selecting convenient time ranges
* ignoring outliers
* choosing a bad sample
* reporting percentages without context
The numbers themselves don't lie — but people can absolutely use them to mislead.
The joke lands because everyone who's seen a suspicious graph knows it's not entirely a joke.
9. "There's a 50% chance this joke is funny."
Why it's funny
This joke plays with the way probabilities are casually thrown around.
People often say "50% chance" when they really mean:
* maybe yes
* maybe no
* I have no idea
But in statistics and probability, 50% has a precise meaning.
So the joke is funny because it pretends to give a mathematically rigorous estimate of humor — which is obviously not how jokes work.
It's funny because it applies statistical precision to something completely subjective.
10. "Our model has 99% accuracy."
"Great. What's the class distribution?" (class distribution：「類別分布」)
"…99% of the data is the same class."
Why it's funny
This is a classic machine learning/statistics joke.
Imagine a dataset where:
* 99% of emails are not spam
* 1% are spam
A lazy model could simply predict “not spam” for everything and still get 99% accuracy.
That sounds impressive, but the model is actually useless.
The joke highlights a huge lesson in statistics and ML:
A metric can look great while hiding a terrible model.
The humor comes from knowing that accuracy without context can be meaningless.
Written by Md Johirul Islam
Software Engineer at Amazon
本文於 修改第 2 次
|
|
「意識」：我的了解-Siegbert
|
|
|
推薦1 |
|
|
|
「我的了解」一詞，出於我對下文作者的基本尊重。也許，稱為「天馬行空之冥想」，或「東拉西扯之胡謅」，比較貼切。置於此欄，以其為「思考方法」之負面教材也。
What is consciousness?
Siegbert, 06/05/26
Religious and spiritual people often think of consciousness as the soul.
Scientists and critical thinkers usually assume that consciousness is a property of the brain, something physical, biological, and not yet fully understood.
I want to argue for a third possibility:
Consciousness may be a property of mathematical principles and quantum mechanics.
At the most basic level, information can be distinguished into 0s and 1s. But between 0 and 1, there is not only a hard binary choice. There is also a space of potential states, uncertainties, probabilities, and transitions.
In this perspective, consciousness does not emerge from fixed information alone. It emerges from the space between fixed informational states.
A bit may be described as either 0 or 1, but the process of becoming 0 or 1 may contain all possible intermediate information states. This uncertain field of infinite potential choices is what we experience as free will: the openness before a decision becomes inevitable.
Before we choose, infinite possibilities seem available. After we choose, reality collapses into one path.
This is why I describe free will as a kind of informational superposition: the ambiguity of decision-making before circumstance forces a final selection.
The second feature of consciousness is its lack of perfect replicability.
We are not able to reproduce our thoughts in exactly the same way five minutes later. Even if we repeat the same opinion to someone, we do not use the exact same words, rhythm, emotion, tone, or sequence. Something always changes. This is important.
A machine can copy a file perfectly. But a conscious being does not simply repeat. A conscious being reinterprets.
This second feature resembles the uncertainty principle. Every thought process contains an element of uncertainty. This uncertainty makes our thinking flexible, diverse, surreal, redundant, inefficient, and creative.
In mathematical language, we are not simply computations of 0s and 1s. We are not just binary machines. We also contain an error parameter:
e
This e is not merely a mistake. It is the creative instability inside consciousness.
So the mathematical description of conscious information might look like this:
[0+e] [0−1+e] [1+e]
These are not just numbers. They are informational states disturbed by uncertainty.
Now, what happens when many e’s accumulate, repeat, interfere, get stored, and become activated again?
The answer is simple: We forget.
Memory is not a perfect archive. It is an unstable reconstruction.
How much do you remember about what you ate last Monday morning? What was the name of the neighbor who moved out of the residual building five years ago? What kind of present did someone give you on your last birthday?
You may remember some of these things. But most likely, many details are already gone, blurred, or rewritten.
Forgetfulness is not a failure of consciousness. It may be one of its central properties to focus on what is really important.
A perfectly deterministic system would store and retrieve everything exactly. But consciousness does not work like that. It compresses, distorts, selects, forgets, and recreates. Maybe this evolution made our brains the fittest for survival.
Another almost mythical property of consciousness is awareness itself.
We are not only processing information. We are aware that we are processing information.
We experience our thoughts, our bodies, our environment, and other people. We do not merely act; we observe ourselves acting.
In mathematical language, this is known as self-reference.
Self-reference can be expressed through recursive functions, iterations, and loops. A system becomes self-referential when it can refer back to its own informational states.
I describe this self-reference as an index function:
i
applied to conscious information states:
[0+e]i(0+e) [0−1+e]i(0−1+e) [1+e]i(1+e)
This means that the system does not only contain information. It also indexes, observes, and references its own information.
In simpler words:
A conscious system is not only a calculator. It is a calculator that watches itself calculating.
This self-reference may be one of the deepest roots of awareness. The brain does not merely compute external reality. It constantly checks, updates, and interprets its own internal states.
That may be why consciousness feels so strange from the inside.
We are, in a sense, inefficient and slow-witted quantum computers. We burn enormous amounts of energy not only to process the world, but to self-reference our own actions, memories, emotions, and decisions at every moment in time.
The principles of superposition, uncertainty, and self-reference may together give the brain its most mysterious abilities:
the ability to experience free will,
the ability to forget,
and the ability to remain aware of itself.
Consciousness is not a simple substance, not merely a soul, and not merely brain matter.
* It is a dynamic mathematical process.
* It is information disturbed by uncertainty.
* It is memory shaped by error.
* It is computation observing itself.
In a certain sense, we may be immortal algorithms. But we are also so unique in our experience, so unstable in our inner states, and so dependent on our exact path through reality, that we can never be perfectly replicated.
Perhaps this is what makes consciousness so difficult to define.
It is not just what we are. It is the process by which we keep becoming ourselves.
What do you think consciousness is?
Leave a comment below.
See you!
Written by Siegbert
spiritualist and philosopher
本文於 修改第 2 次
|
|
真相/真理：三種不同的觀念 -- Ronald Bailey
|
|
|
推薦1 |
|
|
|
不同的人對「真相」或「真理」有不同的「定義」。至於把「真相」或「真理」拿來當工具騙選票、混飯吃、或混砲打的人，可能就在你身邊。
索引：
confirmation bias：偏聽偏信；依照自己的成見、偏見、利益、或立場等因素，「選擇性」的處理相關資訊或證據。
correspondence Theory of Truth：「真理對應論」；在傳統「認識論」中，另外兩個關於「真理」的理論為：「真理相容論」和「真理貓鼠論」；在近代「認識論」中，其它關於「真理」的理論有：「真理建構論」、「真理共識論」、「真理誤指論」、「真理自道論」、「真理言行論」、「真理多相論」等等；另請參考下文作者提供此術語的「超連結」。此處的中文譯名與通行者有出入。
Theories of truth：《關於「真理」的各種理論》；此為《維基百科》條目，對以上各種關於「真理」的理論有簡單說明。
The Surprising Divide Over What Counts as True
A new study finds that what people think about facts, authenticity, or coherent beliefs explains why they disagree about what is true.
Ronald Bailey, 05/15/26
Maria and Peter are students and meet up for a late dinner. Peter asks Maria whether Tom is at the party that they intend to go to after dinner. Maria answers that Tom is at the party. After all, Tom had told her that he would be at the party. When they arrive at the party, it turns out that Tom had changed his plans, and is not at the party.
This was the scenario posed to research participants in a new study by a team of European researchers. They were then asked: Was Maria's answer true or false?
It's pretty clear that Maria's answer is false, at least from my point of view. In other words, I am fully embracing the correspondence theory of truth. However, the study, published in Cognition, shockingly found that only just over 50 percent of participants would agree with me. Apparently, many other people tend to identify truth with how well a statement fits within a person's coherent set of beliefs or whether a person's beliefs are authentic, that is, they are sincere and honest.
To probe how ordinary people think about what is true, the researchers first created conceptual maps of 200 participants asking how similar they think truth is to other related concepts. For example, correspondence related to "reality" and "fact"; coherence to "justification" and "reason"; and authenticity to "honesty" and "transparency." While many participants endorsed notions relating to all three conceptions of truth, in a "winner-takes-all" summary of the judgements, 55 percent aligned most strongly with correspondence.
三種不同的真相/真理觀念統計圖：
54.69% 的人接受真相/真理的「對應性」觀念(correspondence)
10.42% 的人接受真相/真理的「一致性」觀念(coherence)
34.90% 的人接受真相/真理的「真誠性」觀念(authenticity)
In other words, just a bare majority believes that truth is defined by factual reality.
The researchers then wanted to see if these concepts of the truth remained stable in individuals over time. So three months later, they managed to contact 128 of the original participants and ask them to consider what is the truth in the above Maria vignette. In this case, the choice was binary: Was Maria's statement true or false? As the researchers explained, "A 'true' response reflects an authenticity- or coherence-based understanding, as it emphasizes Maria's sincerity or justification at the time of speaking, while a 'false' response reflects a correspondence view, judging truth based on factual alignment with reality." I have no trouble accepting that Maria could try to justify her sincere and honest belief that Tom was at the party, but the plain truth is that he wasn't there.
In the later survey, it turns out that an individual's concept of truth does modestly predict how he or she evaluates the truth of Maria's statement. The researchers report, "Overall 68 (53.13%) participants responded that Maria's answer was false (agreeing with correspondence theory) and 60 (46.89%) that her answer was true (agreeing with an authenticity or coherence notion of truth)." Again, a bare majority endorsed factual reality as the standard for determining what is true.
In an article describing their findings over at Psyche, the researchers outline how different conceptions of the truth can cause conflict:
Imagine someone makes a statement about climate change. The discussion unfolds predictably: one side posts links to data (correspondence), the other side cares less about data and replies with accusations of bad faith (authenticity), or they argue that the statement is untrue because it doesn't fit everything else they already believe to be true (coherence). In such disagreements, giving more of the evidence that convinces you could risk making the conflict worse, not better.
We have all been there, haven't we? Even for those who endorse the correspondence theory of truth must still grapple with the pervasive problem of confirmation bias.
As I reported a while back, research by the Yale law professor Dan Kahan finds that as scientific literacy goes up, so too does partisan polarization on the issue of climate change. In other words, the more science people know, the more they are able to seek out and find information justifying their beliefs.
Nevertheless, the European researchers suggest hopefully that understanding the differences in the conceptions of what is true may help us more fruitfully navigate political and policy disagreements.
Start your day with Reason. Get a daily brief of the most important stories and trends every weekday morning when you subscribe to Reason Roundup. (請至原網頁訂閱)
本文於 修改第 3 次
|
|
「『非』現實論」 -- Jack Preston King
|
|
|
推薦1 |
|
|
|
下文的標題是個「邏輯謬誤」中的「『類型』錯誤」；作者論述「非」的是：「人們對現實的『認知』」；並不是「現實『本身』」。這個「邏輯謬誤」的歷史悠久，可以上溯到釋迦牟尼；如「色即是空」一類說法。
The Case Against Reality
Jack Preston King, 04/17/26
We do not perceive the world as-it-is
Donald Hoffman is a Professor of Cognitive Sciences at the University of California, Irvine, and an award-winning researcher into perception, evolution, and consciousness. Know this going in. Hoffman is a professional scientist, and The Case Against Reality: Why Evolution Hid the Truth from Our Eyes is in every way a serious science book.
It also turns everything we think we know about reality on its head.
The basic premise is simple. At the heart of the “evolutionary algorithm” is fitness. Biological evolution does not work by accurately revealing “objective reality.” It works by identifying and emphasizing environmental fitness payoffs and hazards. The senses of every creature provide critical environmental information regarding what is likely to help them live long enough to reproduce VS what may kill, injure or sicken them. And fitness payoffs/hazards vary according to the creature; they are not intrinsic to the thing encountered. The same evolution that shaped flies to experience excrement as a tasty food source shaped humans to turn away in disgust.
We do not perceive the world as-it-is. We perceive the world as a cognitive interface drawing our attention to fitness payoffs and warning us away from hazards (still a payoff, since we avoided death today). Aspects of reality that serve neither function get edited out of awareness by natural selection, and so are unknown to and potentially unknowable by us. Even with advanced scientific thinking and instrumentation what we can know about reality is limited by our biologically-evolved senses, following evolution’s blind survival algorithm.
We have no idea what’s really “out there” in the world. We can only know how things in our environment relate to human survival. Things that impact our survival, we see twisted in the light of our need (we see their fitness payoff, not their intrinsic nature). Things that don’t impact our survival, we don’t see at all.
In Hoffman’s view — backed by science! — the chance human perception accurately describes objective reality is statistically zero.
From The Case Against Reality:
Darwin’s idea of natural selection entails the FBT [Fitness Beats Truth] Theorum, which in turn entail that the lexicon of our perceptions — including space, time, shape, hue, saturation, brightness, texture, taste, smell and motion — cannot describe reality as it is when no one looks. It’s not simply that this or that perception is wrong. It’s that none of our perceptions, being couched in this language, could possibly be right.
… That revolutionized view leaves in its wake an evolutionary biology that is itself transformed. Still recognizable… are the landmarks of universal Darwinism: variation, selection, and heredity. But gone from objective reality are physical objects in spacetime…
… The FBT Theorem asserts that if reality outside the observer has any structure beyond probability, then natural selection will shape perception to ignore it.
… The key insight of the theorem is simple: the probability that fitness payoffs reflect any structure in the world plummets to zero as the complexity of the world and perception soars… the laws of probability dictate that Truth has less chance than your lottery ticket.
I can’t wait to try this one out on Twitter/X.
THEM: There is no evidence for the existence of God!
ME: There is no evidence for the existence of anything.
THEM: But… But Science! Science describes verified objective reality!
ME: So you deny evolution, then?
LOL.
I listened to the Brilliance Audio audiobook of The Case Against Reality: Why Evolution Hid the Truth from Our Eyes, narrated by Timothy Andrés Pabon. Pabon is terrific, with a clear voice, solid pacing, and lots of inflection. He always sounds interested in the text, which kept me always interested in the text. Well done.
Quotes in this essay are used under fair use copyright guidelines.
This post contains affiliate links. I may earn a small commission if you make a purchase through these links, at no extra cost to you. Rest assured, I only link to books that I personally recommend.
Thank you for reading!
Written by Jack Preston King
I write for people who intuit that there is more to reality than meets the eye. JackPrestonKing.com Follow me on Bluesky @jackprestonking.bsky.social
Not a member? Read this story free on jackprestonking.com!
本文於 修改第 1 次
|
|
社會科學實驗難以重複問題 -- David Randall
|
|
|
推薦1 |
|
|
|
We Know Social Science Is Shoddy. It's Time to Actually Fix It
David Randall, 04/15/26
“Half of social-science studies fail replication test in years-long project”: that’s the headline from reports on the latest mass replication project sponsored by the Center for Open Science (COS). COS has been sponsoring several mass replication projects in the last decade, each of which organizes dozens or hundreds of researchers to replicate recent research in different disciplines. As famed reproducibility researcher John Ioannidis notes, “the results are ‘not surprising’, because they are in line with those from smaller, earlier studies.
A combination of flawed statistical procedures, slipshod research techniques, skewed publication incentives, and politicized and disciplinary groupthink has led to the mass production of irreproducible research. The key takeaway is that you cannot trust recent research to produce a true result. Entire bodies of science may be untrustworthy, as they are built on mountains of individually unreliable results.
Most commenters emphasize how this latest project points out the unreliability of social science research. Skeptic that I am of quantitative social science—my presumption is that much of it is piffle piled on piffle—I actually think that the report should make one optimistic about the capacity of the social sciences. As much as half of their results reproduce! Social sciences are roughly as capable of producing reproducible results as basic life sciences. This actually is a great vote of confidence in the social sciences, which has aimed to provide a quantitative, scientific basis to the study of human behavior.
The social sciences can achieve true results, but not nearly as easily as its practitioners have hoped, and too casually assumed. Any social science results (as any scientific result) should be achieved multiple times, by different researchers, before it receives even preliminary recognition as a plausibly true result. The rate of social science research, if it is to be at all reliable, must be much slower. The social sciences can produce truths, but not if they continue business as usual.
The question is how to change business as usual. At this point, there may be diminishing returns to these massive replication projects sponsored by COS. They provide useful publicity to remind scholars and the public about the existence and the seriousness of the irreproducibility crisis, but at this point most scholars in the affected fields are aware of the problem. They may choose to resist reform, but that is no longer from ignorance. What we need now are ways to build upon these mass-replication projects. We must figure out ways to change the standard operating procedures of the social sciences (and sciences) to create standardly reproducible and reproduced research.
The National Association of Scholars (NAS) has made a series of suggestions for how precisely to reform scientific procedures, and in particular the government’s financial incentives for scientific and social scientific research. But broadly, we should set social science research on an even keel by the following reproducibility reforms:
* Require pre-registered research hypotheses and establish born-open data for all research data.
* Separate data collection from data analysis and establish standard research procedures assigning the two functions to different, independent researchers.
* Establish standard research procedures for different, independent researchers to conduct analysis for every research hypothesis.
* Set high definitions of required statistical power for any quantitative social science research, and make clear that all low-power studies—“qualitative,” “evidence-based,” and other euphemisms for abandoning statistical rigor—have no social-scientific value, and no professional value in any discipline worthy of public support.
Of course these changes would require massive changes in academic culture. Social science researchers will need to accept massive numbers of negative results as normal, and to distribute publications, tenure, and other professional rewards for researchers who get negative result after negative result. Social science departments will need to determine how to distribute job offers and promotion for individual members of massive research teams, where it is more difficult to determine which individuals deserve the most credit. Maintaining researcher independence among cooperating teams will be a practical problem. Independent peer review will become much more difficult if most or all researchers in a particular subdiscipline necessarily will be taking part in a particular cooperative research project.
Perhaps most importantly, reformed social science research will become much more expensive. It costs far more to assemble data with the number of research subjects necessary to produce data with high statistical power. Social science will become far more difficult to fund—and government and private foundations well may determine that it isn’t worthwhile to invest money in social science that costs several times as much for each research result.
All these are great challenges. But if we do nothing, social science will continue to produce results where their veracity is essentially a coin flip. The Center for Open Science has made clear the terrible state of the status quo in the social sciences. Social science scholars and public representatives now should move to make the comprehensive changes in the procedures, cultures, and funding that shape American social science research. Social science research can determine truth. We should start at once to make sure that it does.
David Randall is the Director of Research at the National Association of Scholars.
本文於 修改第 1 次
|
|
關於「受理論制約」的說法 – Paul Austin Murphy
|
|
|
推薦1 |
|
|
|
請參考：
Feyerabend, Paul
Hacking, Ian
Wittgenstein, Ludwig
下文相當鬆散；與其說是一篇「文章」，不如說是作者把他的「讀書心得」或「讀書札記」拼湊在一起。不過，「受理論制約」這個議題值得討論；故轉載於此。
Observations of Feyerabend’s Theory-Laden Underpants
The idea of theory-ladenness became popular in the 1960s, and has remained popular in various sections of academia ever since. It’s often said to have begun with the work of Kuhn, Hanson and Feyerabend in the late 1950s. This idea impacts on science, and is said — by some — to lead to “anti-science”, “relativism” and “the attack on objectivity”. Whatever the case is, scientific theory is obviously distinguished from observation. So are all observations theory-laden? Just some? Or is it more a question of degree? The following essay relies on the work of the philosopher of science Ian Hacking, whose position is nuanced. Hacking sees the truth in much of what Kuhn, Hanson and Feyerabend argued. Yet at the same time he stated that “[t]here have been important observations in the history of science, which have included no theoretical assumptions at all”. Hacking also argued that the overuse of the term “theory-loaded” effectively made it “trifling”.
Paul Austin Murphy, 03/25/26
Philosophical positions on the distinction between observation and theory range from the old “naïve” view (i.e., that scientific observations must be pure) to rejecting the distinction altogether. Many philosophers take a position somewhere down the middle, but even that middle has various grey areas.
The philosopher Ian Hacking took a nuanced position that’s partially against some distinctions between theory and observation, but one which questions the complete rejection of the distinction too.
What is original to Hacking is his stress on experiment, which he believed had been largely ignored by philosophers. More relevantly, we had the distinction between observation and theory, but now we have Hacking’s idea that “[e]xperiment supersedes raw observation” too. It may be a surprise to some that Hacking simply inverts this particular binary opposition. Perhaps there’s little point in saying that observation supersedes experiment or that experiment supersedes observation.
N.R. Hanson and Paul Feyerabend
The fixation on theory-ladenness at least partly began with the philosopher N.R. Hanson. In his 1959 book Patterns of Discovery, he came up with the term “theory-loaded”, which in subsequent years became a bit of a cliché. Hanson argued that every sentence and term is theory-loaded.
Many of Hanson’s examples are convincing. However, is what is drawn from them legitimate too?
The American philosopher of science Dudley Shapere adds to Hanson. In his case, however, it’s the nature of scientific devices which concerns him. Hacking states that Shapere
“ makes the further point that physicists regularly talk about observing and even seeing using devices in which neither the eye nor any other sense organ could play any essential role at all”.
It can be said that there’s no serious problem with scientists using everyday terms in their own non-everyday work. Why shouldn’t they use the words “observing” and “seeing”? What’s more, it can be doubted that many scientists will be troubled with the quibbles philosophers have with their using these words.
Feyerabend went further than Hanson and Kuhn.
In his 1977 book Against Method, he argued that the observation-theory distinction is bogus. In other words, all scientific observations are always theory-laden.
Some readers may now be wondering what exactly Feyerabend meant by “theory”. (This is true about many other uses of that word too.) Hacking picked up on Feyerabend’s use of the word when he wrote the following:
“Unfortunately the Feyerabend of my quotation used the word ‘theory’ to denote all sorts of inchoate, implicit, or imputed beliefs.”
The word “theory” is often thrown around like confetti. Thus, on Hacking’s reading of Feyerabend’s position, one doesn’t need to express one’s theory explicitly, or even know that one has a theory in the first place. In addition, it’s often not the case that a person has a theory “behind” his words or statements: it’s that other people believe that he has. So a theory can be vague, unexpressed and projected onto others.
Despite all that, Feyerabend still believed that theoretical assumptions underlie
“the material which the scientist has at his disposal, his most sublime theories and his most sophisticated techniques included, is structured in exactly the same way”.
Now an everyday cliché can be used:
If everything is classed as a theory, and if theories can be found in every statement or term, then there are no theories at all. The word simply ceases to have any point.
Hacking agrees:
“Of course if you want to call every belief, proto-belief, and belief that could be invented, a theory, do so. But then the claim about theory-loaded is trifling.”
Hacking wrote the following too:
“Of course we have all sorts of expectations, prejudices, opinions, working hypotheses and habits when we say anything. Some are contextual implications. Some can be imputed to the speaker by a sensitive student of the human mind.”
Did Feyerabend really include expectations, prejudices, opinions and habits under the catchall term theory? Indeed, can’t we have expectations, prejudices, opinions and habits and it still not be the case that what we say (or everything we say) is theory-laden?
Hacking: No Theories At All
Oddly enough, although the theories-are-everywhere idea can easily be criticised, Hacking’s own position seems odd too, at least at first. He continues:
“There have been important observations in the history of science, which have included no theoretical assumptions at all.”
Following on the what was said a moment ago, it can be provisionally accepted that these important observations included no theoretical assumptions at all. Yet, at the very same time, the people who made them were (well) over-laden with expectations, prejudices, opinions and habits…
So what?
Hacking’s position may even strike a scientific realist as being extreme. Just for one. Why were such scientists making their important observations in the first place if they were completely free from theoretical assumptions? Of course, we’ll need to see examples here. Hacking, being an historian of science [see here] as well as a philosopher of science, cites plenty.
To ram the point home. Hacker argued that “[t]here are plenty of pre-theoretical observation statements, but they seldom occur in the annals of science”. Just to remind readers. Hacker took a fairly strict position on what a scientific theory is, whereas someone like Feyerabend was very loose with the term.
Logical Positivists and Quine on Observation
Feyerabend was largely reacting against logical positivism. [See here.] Hacking puts the positivist position at its most extreme when he tells his readers that the
“positivist, we recall, is against causes, against explanations, against theoretical entities and against metaphysics”.
More clearly, “The real is restricted to the observable.”
All this depends on what “the real” means. If it means observable, then that statement is true by definition. Of course, the core of the planet Earth can’t be observed, and neither can distant planets and quarks. What about numbers? Positivists had various answers to some of these examples, but not to all.
To run through Hacking’s list. Being against causes is a Humean position. Obviously, causes can’t be observed. Constant conjunctions can be observed, but not the nature of the cause itself. To be honest, I can only guess at what “against explanations” means. Is the argument that if you rely exclusively on observation, then you don’t need explanation too? The case against metaphysics is obvious from a positivist and observation-based point of view.
Now take Hacking’s criticisms of W.V.O. Quine’s position, which stresses not observations, but “observation sentences”. Quine, as quoted by Hacking, states that we should “drop the talk of observation and talk instead of observation sentences, the sentences that are said to report observations”. Hacking had a problem with Quine’s distinction (observations vs observation sentences) within another distinction (observation vs theory).
Hacking stated that Quine was “quite deliberately writing against the doctrine that all observations are theory-loaded”. Quine articulated this position in his 1974 book The Roots of Reference. That was long after Kuhn, Hansen and Feyerabend had first articulated their own controversial theories. As many readers will know, the critics of Kuhn and Feyerabend focussed on their “relativism”, rather than their stress on theory-ladenness. Of course, these two issues have been intimately tied together.
The first thing that can be said here is that Quine’s observation sentences simply seem to be proxies for… well, observations. And Hacking’s general point against Quine is that his theory of observation sentences is just as naïve as some talk about (mere) observations. So Hacking clarifies Quine’s position, which is essentially communal in nature.
Quine believed that “observations are what witnesses will agree about, on the spot”. That’s a non-scientific (or non-academic) three words to use: “on the spot”. Still, Quine provided details. He argued that
“a sentence is observational insofar as its truth value, on any occasion, would be agreed to by just about any member of the speech community witnessing the occasion”.
What’s more, “we can recognise membership in the speech community by mere fluency of dialogue”.
So we have these everyday utterances from Quine: “on the spot”, “just about any member of the speech community”, etc. But let’s remember here that Quine was a pragmatist of sorts. [See here.] So why shouldn’t he have been imprecise on these matters and simply focussed on his own self-referential observation sentences about how communities use words and sentences.
Hacking too has a problem with the above. He stated that “[i]t is hard to imagine a more wrong-headed approach to observation in natural science”. His main argument, at least at this point, was against Quine’s stress on the community. He cited the case of Caroline Herschel, the wife of William Herschel:
“No one in Caroline Herschel’s speech community would in general agree or disagree with her about a newly spotted comet, on the basis of one night’s observation. Only she, and to a lesser extent William, had the requisite skill.”
This seems to go against any notion of communal truth. Alternatively put, it shows that truth can be arrived at independently of any community.
Various communal ideas about truth and knowledge can be dated back to Wittgenstein [see here], as well as before, and Quine was much influenced by him. [See here.] Hacking’s own stress was different. The last two words, “requisite skill”, are important here. In the community of the late 18th century there wouldn’t have been many — if any — people with same skills as Caroline Herschel. But did she or didn’t she discover eight new comets? Yes she did. [See here.] Yet she didn’t need — or rely on — any community to do so, except in less direct and rather obvious senses. In other words, Herschel needed the traditions of science, the devices of science, to share a natural language with various communities, etc. However, none of this is directly connected to Herschel discovering comets, and the knowledge she gained by doing so.
Written by Paul Austin Murphy
MY PHILOSOPHY: https://paulaustinmurphypam.blogspot.com/
My Flickr Account: https://www.flickr.com/photos/193304911@N06/
本文於 修改第 1 次
|
|
「社會科學方法論」概論 ---- 呂曉波/江軍
|
|
|
推薦1 |
|
|
|
「社會科學方法論」概論
呂曉波主講/江軍整理，2018，6月
導論：方法論上的焦慮
在簡短感謝張佑宗主任的介紹後，呂曉波教授隨即點明了政治學界現存的「方法論焦慮」：「現在大家都存在一種焦慮的狀態…我是不是一定要做實驗？都不做實驗，我們是否就沒有辦法發表在頂級的期刊？」呂曉波教授認為，一方面，問題意識才是關鍵，而方法是為問題服務。另一方面，不見得唯有用因果推論的方法才能得出好的因果推論，而用其他方法得出的結論則無效。唯因所有的方法都是基於一系列假設，當假設都不成立的時候，得出的結論自然無效。換言之，因果推論的方法不是穩健結論的保證。另一種焦慮則是大數據、網路爬蟲等新方法不斷出現，學習的速度跟不上方法的演進。但呂曉波教授強調，這依然取決於你的方法對應到什麼問題。研究方法就如同學語言，要不斷積累和深入，並在學習的同時，思考方法的應用、背後的假設、與打破假設的可能。職是之故，今日演講也將順著因果推論方法的框架、假設、應用與限制等四個面向展開。
因果推論方法出現之前：如何解決實證問題？
在因果推論的方法出現之前，大家是用怎麼樣的方法來解決實證問題？呂曉波教授指出，社會科學家在沒有因果推論方法之前，仍在探討因果關係，亦即當某個條件發生時，會造成怎樣的後果。這些研究往往會碰到三個實證上的問題：
第一是理論的概念和測量方式不一致(measurement error)，例如「民主」和「非民主」編碼為二元變項0和1的轉錄標準往往不甚清楚。
第二是遺漏變數(omitted variable bias)，意思是有一些變數和自變數與依變數相關，但沒有被納入分析。如此一來，推論結果自然會出現偏差。
第三則是內生性問題(endogeneity)，亦即自變數與依變數互為因果。在因果推論方法出現之前，解決上述實證問題的方式，通常是做敏感性分析(sensitivity analysis)─如果要克服測量的不一致，那就選擇另一個變項重新測量一遍、把概念再衡量一遍、抑或是再跑一次模型。簡言之，就是用不同指標來衡量、抑或是用不同模型與數據來源測試。若該分析結果能支持研究論述，那理論就是正確的。
呂曉波教授認為，敏感性分析在選擇實驗組時即有偏誤可能，而模型的挑選與處理也不見得是客觀的。
什麼是因果推論？
因果推論的基本概念為反事實分析(counterfactual analysis)，也就是一件事發生和沒發生，在結果上的差異。因果推論方法有三個基本假設：首先，我們無法看到在個人層面上的因果效應，只能看到群體平均效果(average treatment effect)，這表示在群體層面上，實驗組和對照組作為兩個群體，在許多特徵上一致，唯一差別是一組接受處理(treatment)，而另外一組沒有接受。第二個假設則是一致性(consistency)，亦即同樣一組人接受相同處理，其結果會相同。最後，在做實驗時，我們是藉由抽取樣本並給予處理來證成因果關係，但我們會推論：對整個母體而言，處理的效用是和樣本一樣的。也就是說，「給一小部分人的處理」和「給全部人的處理」，兩者的效果相同。
接下來，呂曉波教授進一步分析因果推論方法的兩類實證問題。第一類是干擾變項(confounder)：當研究者想探討D對Y的影響，但有一個變數L同時對D和Y有真正影響。此時，若不考慮L，而僅收集D和Y的資料，兩者間會有統計上的顯著相關，但這卻不是真正的因果關係。例如當天氣熱時，我們衣服穿得少、冰吃得多，但不代表穿衣服和吃冰間有因果關係，唯因這是天氣熱造成的結果。第二類則是案例選擇(selection)問題：當我們推論Y是造成L的條件時，卻只選擇具備Y條件且出現L結果的案例作為佐證。例如對所謂發展型國家(developmental state)研究的批評之一，就是研究者只注意成功案例，而不注意那些未發展成功的案例，所以這個理論無法推廣。呂曉波教授特別提醒大家：研究者必須注意研究對象的選擇是否會導向特定研究結果。
因果推論的方法：主要研究方法與假設
一旦研究者試圖做出因果推論，就須考慮兩個策略：第一，因果證成策略(identification strategy)，意指有沒有一些假設讓你能宣稱你的研究具有因果關係。第二，統計估計策略(estimation strategy)則關乎統計估計方法的使用。如果一項研究無法滿足方法背後的假設，無論該研究使用的方法多花俏，其結果都不具因果效力，故研究的關鍵就在於研究設計如何滿足方法背後的因果證成策略。
那麼，因果證成策略有那些呢？呂曉波教授介紹了五種常見的因果證成方式：
1. 實驗法(Experiment)：研究者能人為的隨機操縱實驗組和對照組，並假設在平均效果上，隨機分配的實驗組和對照組，兩組間沒有特性差異。此種方法的侷限是外部效度(external validity)問題。另一方面，宥於環境和經費，實驗往往難以複製。
2. 自然實驗法(Natural Experiment)：現實情況的改變自然隨機分出實驗組和對照組。和實驗法相同，自然實驗法也假設在平均效果上，實驗組和對照組在特性上沒有差異。由於自然實驗法不是人為操縱，而是環境改變的結果，故有內部效度(internal validity)的問題─研究者必須要提出很多證據來證明隨機分配是成立的。
3. 工具變數(Instrumental Variable, IV)：若今天我們想探討D和Y之間的關係，但有一個干擾變數U會同時影響X和Y，以致於我們無法觀察D和Y間的關係。此時如有一個外生變數Z，可以透過影響一部分的D而對Y產生影響，那我們就可以藉由兩階段最小平方(two-stage least square, 2SLS)的迴歸分析來驗證D和Y間的關係，此時我們稱外生變數Z為工具變數。這個方法的關鍵假設是Z不能對Y有單獨的影響，限制在於很難找到一個符合關鍵假設的工具變數。
4. 斷點回歸(Regression Discontinuity, RD)：有些群體觀察值在特定的時間和地域範圍內會產生隨機差異。這個方法的假設是觀察值在短時間內有大幅度的變化，而這個變化的處理是隨機的。限制是在斷點邊界上確能找到很強的因果效應，但推廣到整個母體時結果卻不一定如此。故斷點回歸僅能滿足狹域的平均政策效果(local average treatment effect, LATE)。
5. 差異中的差異法(Differences in Differences)：此方法是在考慮處理前後的時間差異下，去檢視一個外生衝擊對不同群體的差異效果，關鍵假設則是平行趨勢(parallel trend assumption)，也就是在不加入處理的情況下，實驗和對照組的差異趨勢會是平行的。
結論與問答
在演講尾聲，呂曉波教授做出了四點結論。首先，現在的研究光提出因果效應還不夠，還必須提出對因果機制的檢驗。第二，方法是有其侷限的，若太注重因果推論，反而無法深入思考問題。第三，因果推論往往具有外部效度問題。最後，過度強調方法的結果，造成今日的研究多是檢驗現存理論，而非嘗試發展新的理論。呂曉波教授認為，理論上沒有突破，代表著政治科學無法繼續進步。
在提問時段，聽眾詢問該如何區分方法論、研究方法和研究工具？若一味追求工具而沒從方法打基礎，是否本末倒置？針對此問題，呂曉波教授指出，要有紮實的方法論基礎才能把握研究工具的假設和侷限。他的建議是，尋找對自己研究問題相對應的研究方法來鑽研，但也不要只看一種方法，因為「當你只有一支槌子，所有的東西對你來說都只是釘子」。
另一位提問人則詢問教授，什麼是正確的方法？後進學者又該如何挑戰現有理論，說服前輩學者？呂曉波教授認為，挑戰本身的動機是好的，但很多後進學者意欲挑戰舊說，卻無法確實掌握理論和實證，且故事說得不好。簡言之，若要挑戰，理論功課與實證必須做好，文章的布局(framing)也要把握，如此的論述才會有力。
後記：
上文是我在飆網時歐然看到的一篇文章；原載於《台大政治系系友聯誼會電子報》。原文並無標題，此處的「標題」是我根據「編者」權限加上的。轉載該文源起和講者介紹如下：
2018「胡定吾海外新銳學者講座」
整理：江軍(臺大政治學系研究所)
本系於2018年6月首次舉辦「胡定吾海外新銳學者講座」，邀請到德州大學奧斯汀分校政府系的呂曉波助理教授，由本系張佑宗擔任主持人，並於6/7、6/8、6/13、6/14、6/15進行5場與研究方法有關的演講。
講者簡介：呂曉波(Xiaobo Lü)教授
呂曉波，美國密西根大學碩士、美國耶魯大學博士，2012年榮獲美國政治學會Mancur Olson政治經濟學最佳博士論文獎，現為美國德州大學奧斯汀分校政府系助理教授。
呂曉波教授的研究專長包含發展的分配政治、中國政治、以及比較與國際政治經濟學。其中，社會支出與稅賦的政治及其相應的政治後果，為呂教授最感興趣的研究課題。
呂曉波教授的研究成果已見諸於美國各大頂尖期刊，包括American Political Science Review、American Journal of Political Science、Comparative Political Studies、以及Quarterly Journal of Political Science等。
本文於 修改第 1 次
|
|
《量子力學與現實》讀後
|
|
|
推薦1 |
|
|
|
0. 前言
我不是哲學系科班出身；對量子力學的了解也止於物理系研究所第二年的程度。對阿洛唷和阿倫哈兩位教授的大作，自然沒有能力從純學術的角度評論(請參見本欄上一篇)。
以下從「方法論」角度，以對「科學基礎論」有興趣的人這個身份略表淺見；前者是我把這兩位教授大作置於此欄的原因。
1. 破彼論
1.1 論述策略
兩位教授在該文第3節第1段引用庫恩博士1973的論文，並甚為推崇。如果我對庫恩理論的了解還算得上六、七不離十，他本人並不接受兩位教授所描述的「科學現實論」；請參見以下第2、3兩節。
接受兩位教授所描述「科學現實論」的自然/社會科學家應該寥寥無幾。他們兩位大作的論述策略非常近於「稻草人」。
1.2 論述方式
兩位教授在第3節最後兩段使用「經驗論」和「實用論」兩個名詞來揶揄他們想像中的「科學現實論」者。兩位教授這篇大作的論述方式可稱之為「文字遊戲」；請參見以下第2、3兩節。
2. 立自義
1) 這兩位教授忽略了一個重要的「科學基礎論」原則或「假設」：
所有的「科學理論」都可能被「新發現」修正/推翻；或被「新理論」取代。
我雖然不盡同意庫恩的理論，以我對「典範移轉論」的了解，他顯然接受上述「原則」/「假設」。它也正是兩位教授看不上的「經驗論」觀點。
2) 換句話說：用「真」、「假」來形容一個「科學理論」是18世紀的思考邏輯；19世紀以降，形容或判斷「科學理論」的標準是：
它能不能幫助我們解決日常生活上的大、小問題；或者說，它能不能幫助我們解決自然界帶給我們的困難與災難。
這個立場正是兩位教授看不上的「實用論」觀點。
3. 結語
1) 沒有一位頭腦清楚的自然/社會科學家會排斥上述「經驗論」或「實用論」的「觀點」。
2) 自然科學能不能100%的「描述」外在「現實」是個「假議題」；或者說，它只是個提供某些哲學家寫論文(混飯吃？)的「議題」。對一個像我這樣也自稱「現實論」的人來說，一個「自然科學理論」能「描述」近於80%以上的「現實」；一個「社會科學理論」能「描述」近於30%以上的「現實」；我就偷著樂了。
本文於 修改第 1 次
|
|
|
本城市首頁
看回應文章
等級：8
