Notes on the Black Swan by Nassim N. Taleb:Pg.36. Non-scalable and Scalable situations. ie: Mediocristan/guassian vs. Extremistan/Mandelbrot
Comparing situations in which the average value is largely unaffected by the addition of a new individual value (physically contrained sets like height or weight, or systems where something is constructed from scratch in each transaction) with situations that are unconstrained - such as personal wealth, or booksales - where there tend to be either absolute losers with zero sales or insignificant property, and true giants like Harry Potter or Bill gates - where the addition of a single extra individual (a black swan) can totally upset the previous average value.
Pg 40: The Turkey lesson about corroborative past - 1000 days of feeding, and on the 1001st day, a twisted neck.
Pg 54: The error of confirmation: "No evidence of X" is not the same as "Evidence of no X"
Pg 59: Switching things round:
"There are no black swans." -->
"All swans are white" ==>
"Anything not white cannot be a swan"
According to (faulty) confirmative logic, that which confirms the second - the sighting of a red mini for example, also confirms the former, that all swans are white.
Pg 68: Cost benefit + information:
*Information is costly to obtain, costly to store and costly to manipulate and retrieve.
To reduce these costs we use patterns and rules to compress/reduce the amount of information needed to re-create the object/event/uncompressed information
However - in so summarizing - we also add more 'order' to the world as we percieve it than there really is - and hence are prone to underestimating the degree of randominity extant in the real world.
Pg. 71 - this error of derived rules - the narrative falacy (drawing of 'story'-like conections between things perhaps not truly connected) and the ludic falacy (using a simulation or 'game'like abstract representation of a real-life situation to draw conclusions, predictions and rules concerning that real life situation - and placing too much faith in them) also effects memory - creating a palimpsest - as you remember not the actual event first recorded, but your memory of remembering of that event, and then the memory of the memory of remembering that event. Like a photocopy of a photocopy, sometimes the details get fuzzy, and sometimes parts of that attendent fuzz get interpreted as something else entirely.
Pg. 76: Narrative fallacy:
"Joey seemed happily married. He killed his wife."
"Joey seemed happily married. He killed his wife to get her inheritence."
The second intuitively seems more likely, though it is in fact not.
Pg. 80: The power of emotional connection - "One death is a tragedy - a million is a statistic." Stalin. That which cannot be easily and vividly imagined, effects us but slightly.
Pg 100: The problem of silent evidence, and the self-sampling assumption. ie: of course all gamblers have beginner's luck - otherwise they wouldn't have become gamblers. Those that prayed but drowned anyway, did not survive to tell the story of how their praying saved their lives.
Pg.107: That which does not kill you makes you weaker. The idea that people are somehow hardened through sub-lethal physical hardship and adversity - which 'weeds out' the weak from the strong - is fallicious because it neglects to state that the surviving 'strong' are also weakened by their trial. The progeny of those so selected perhaps, but again how much strength is derived from nurture, and how much from nurture. How much genetic, how much epigenetic..?
Pg 112: Visable effect. "Look at what I did for you." is more easily marketed than "Look at what I avoided for you." ie: A catastrophe visibly combatted is much more media-friendly than a catastrophe so well prevented that even its threat never broke the surface tension of public perception. (eg. the politician who has saved the most lives is actually the man who exposed the safety records of car manufacturers and made seatbelts maditory, but who even knows his name..?)
Pg. 139: People's evaluations of [the certainty of] their own knowledge is frequently exaggerated. Usually by a factor of twenty. A 2% estimate of error in some area of knowledge is in reality around 40 - 45%. A better education usually increases the error margin, rather than closes it. ie: a graduate is only slightly more knowledgable than X, but is many times more confident in that knowledge.
Pg. 146: Experts and and empty suits: Techné = craft. [know-how] Epistémé = knowledge. [know-what] Imagine you will have an operation - do you want 'know-how' in the form of a surgeon..? Or 'Know-what' in the form of a popular science writer..?
Expert professions: livestock judges, astronomers, test pilots, chess masters, physicists, accountants, pure mathematicians, insurance analysts (as long as dealing with constrained "mediocristan" subjects). The subject matter does not 'move' - chess is chess today, and chess tomorrow, the mechanics of flying does not radically change from day to day, today's cow is tomorrow's cow. ie: the black swan liability is low.
Non-Experts 'professions': stockbrokers, pyschiatrists, psychologists, court judges, councilors, intelligence analysts, financial advisors. High variability linked with overconfident evaluations and high margins of error... Empty suits and empty words, however convincing they may be.
Pg. 151 & 153: Accuracy of prediction. Study reveals that a PHD makes no difference in accuracy. However - a big reputation makes people worse at prediction than those who have none. Famous people also make bad predictors.
Pg. 159: Points of reference: A subconscious tendency to use (random) numbers as seed crystals around which to build belief. A study showed that people shown a (random) number were influenced by it in later tasks involved in estimation - a high number produced higher estimations than a lower number for example.
People find it almost impossible to evaluate something in isolation without some [arbitrary] point to work from.
cf: James No2's lack of absolute originality - everything is related to past matters. What goes before sets the stage for what comes after.
Pg. 171/2/3: Limitations on forecasting essentially historical events - Soft historical studies/social science - are too based in the narrative fallacy to be dependable as anything but fiction. Popper.
"Fundamentally unpredictable" to know the future is to make the future into the present. To correctly predict the future you must also predict accurately the technology of the future. If you can do that, then simply by idea difusion, you will probably be able to create the technology of the future today.
The law of iterated expectations - "If I expect to expect something at a later date, then I already expect that something in the present."
Predictions of the future are really merely re-runs of today and yesterday, perhaps with a few extra bells and whistles. ie: predict tomorrow with reasonable certainty, but not next week.
Pg. 178: problems with prediction from incomplete information. Billiard balls.
Two balls colliding - easy. Three, possible with precise knowledge of initial states... Nine - you need to take into account the gravitational effects of anyone near the table... Fifty-six impacts - you need to exactly know the states of every single elementary particle in the universe.
Billiard balls are easy - they have no personal needs, drivers or desires. predicting collisions between people however...
... And yet that is what social scientists and economists believe they can [reliably/accurately] do. To give themselves a leg to stand on they assume that everyone acts "rationally" (acting toward their best interests in all futures) and "coherrently" (adherring faithfully to habits and predilictions).
The chief fallacy of this is two-fold - it assumes everyone within a given sample has equal knowledge and the ability to put it into practice, and that all habits/predilictions are strongly held.
Given that people frequently make irrational and inconsistant choices - the general theories of economic optimization fail. There can be no "general theory" from which to make [long-range] predictions concerning any topic in which people are a variable - the branching error margins for their decisions/actions become exponential.
Pg. 195: Prediction errors and motivation [aka anticipated utility/affective forecasting]. It is thought a certain degree of prediction bias is innate and serves as a cue toward [beneficial/aversive] action. We overestimate, sometimes drastically, the impact of future events, both good and bad, on our lives. That new car becomes less a vehicle than a key to heaven; that impending redundancy becomes not so much a hiccup but an all consuming tidal wave.
Without such a fanfare and hullabaloo however - would we ever get up off the sofa..?
Pg. 196: Forward is easy, backward is hard. An ice cube is melting. Predicting the shape of the puddle it will make is difficult, but not entirely impossible, if you measure sufficiently well the starting conditions. But now, given a random puddle of water, try to reconstruct exactly what made it...
Helenus, of the iliad (son of priam and hecuba captured and made to tell the acheans how to capture troy) was a seer who predicted backward - seeing the past from its barest remnants.
Pg. 198: The death of determinism. Is everything determined..? Is randomness truly random or deterministic chaos...?
Random has no predictable properties, not even a predictable unpredictability. Chaos however, conceals a predictable order beneath a superficial surface of fake 'random'. It has predictable elements - but they are [very very] hard to know.
It doesn't matter. For any practical purpose, both are the same. Randomness is incomplete information. Opacity. And induction the fallacy of believing facet of life X. is non-random from insufficient observation.
Pg. 205: Barbell strategy in investment - hyper-conservatism and hyper-aggression. Because risk assessments are always error-prone the best way to invest is in the same way as you would gamble so as to invite good fortune and minimize losses. ie, bet big on sure things, but also bet a little at huge odds just on the off-chance. "Medium sized" bets are the worst - they are lauded as 'safe' but for relatively little return in reality risk huge losses should you invest heavily.
ie: two horses: Mr. Racecourse-Rocket and Mr. No-Hoper. MRR always always places in every race he ever enters. Lousy odds, but a big enough bet each way will always pay for your dinner, if maybe not dessert. MNH however wins a race maybe once in a blue moon, (if there's always an eclipse at the same time) at truly stupendous odds (he's the google of racehorses) - put a fiver on him every week, okay, you'll bleed out slow, but if he ever comes in, you can retire.
ie: invest at both ends of the spectrum, at something like 85% / 15% split - but never in the middle.
(This btw. is also why casinos limit single bets to sizes that won't ever bankrupt them - because the combined losses of the many always more than cover the lucky high-rolling few on any given day).
Pg. 216: Sherwin Rosen - the economics of superstars. The differential in earnings between X and Y in extremistan professions - TV celebrity, sports and films etc is increasing. But this does not mean that X is hugely better than Y at their profession, only that these professions instigate a 'tournament' effect - the winner only has to win by even the merest of margins to scoop the whole pot.
Of course, they can also win by pure luck. And once the initial event has occurred, social coherence/adherence effects re-inforce the outcome as people gravitate toward the winner - heaping upon him or her the adulation and material prizes which may indeed make that person win yet again.
ie: two books in the same catagory - horror say - one's a Stephen King, one's a Mr. Nobody. All other things being equal enough people would instinctively buy the former over the latter to make sure that Mr. Nobody stayed Mr. Nobody, however good Mr. Nobody might really be.
Pg. 218/219 Merton's ideas of "preferential attachment" (herding in choices / preferences) and cumulative advantage (the first million is the hardest - success attracts greater successes etc.) leads to Sociologist Pierre Bourdieu noting the link between the globalization of culture and the increased concentration of success. ie - the tournament effect above doesn't really depend on the actual number of players, it still tends to funnel success toward an ever smaller proportion of the population - given that that population is sufficiently well linked together both in comms hardware/software and through shared cultural reference points. MTV say, or concepts of female beauty - Whatever the credo, a punter is still a punter.
In general - the big get bigger, and the small either get smaller or stay the same. ie: We feel - like female birds at a lek - more comfortable with choices that already have been agreed with. (Which also explains the sudden attractiveness you gain from having a girlfriend :) ).
Zipf's law. Linguistics - the idea that the more you use a word, the less effortful it will be to use again. ie: we tend to fish out words from our private lexus with an ease and regularity dependent upon the frequency of their past use. (on the part of the individual and within his/her linguistic enviroment - leading to local dialects appearing quickly if isolated sufficiently)
eg - in English there are roughly 60,000 main words, of which only a few hundred constitute the bulk of written English and even fewer in casual conversation.
Pg. 227: Michael Marmot - Whitehall studies - shows, even adjusting for disease, a higher social status positively effects longevity. eg: an Oscar is worth an extra five years of life to an actor.
People who live in societies with relatively flat social gradients live longer en toto than the populaces of steeply-graded societies. Being a loser kills.
Pg. 233. The Guassian lie: The precipitous fall off slopes of the bell curve - as you move away from the 'norm' - produce probabilities so unfeasibly low that black swans seem completely out of the picture in any sane world. Trouble is, it is massively sensitive to errors in the measurement of sigma (standard deviation) to the point where a very small miscalculation can result in gross (a factor of a trillion for example at 20-21 sigma) underestimation of risks - if a guassian curve is applied to a scaleable situation - such as the stock-market - and used as a reliable predictor of risk... everyone, sooner or later, is totally fucked.
A Mediocristan (guassian) plot has an inconstant (exponential for example) decline as you move away from the centre in either direction, and an extremistan (Mandelbrotian) plot - follows a constant decrease as you move away. Anticipating risks with much greater levels of probability.
The key is to correctly apply the right curve to the right situation, rather than a one-size fits all approach.
Pg. 244: How do you tell where you are..? If you are dealing with yes/no questions, in which magnitudes do not matter or are non-existant (dead/not dead - pregnant/not pregnant - have a something/Don't have a something) then you are in mediocristan. If the magnitude does matter (how much money - how many hits) then you're in extremistan.
Pg. 250: Guassian Assumptions:
One: coin flips have no memory. They are isolate in time and space. You never 'get better' at flipping coins, a head today does not influence the likelihood of another tomorrow.
See - "preferential attachment" and "cumulative advantage" both mean real-world events do not obey Gaussian assumption. Suddenly forty heads in a row becomes more likely than a completely random mix.
Two: Random walks without wild leaps. A Guassian world view has steps of equal (value one) length. The real world has indeterminite step lengths (usually value one) and contains the possibility of a wild leap in times of acute stress.
Pg. 290: Popper. [paraphrase] "Philosophy collapses into empty sophistry without being based around problems that lie outside the strict philosophical domain."
2 comments:
I love how you transmute ideas and one way assumptions thinking so often leads down.
You are a much more tempered and logical when read, than, Heir F.N. yet despite your 'calculated' comment, you are very much playful with ideas, not seeing both sides of the story, but 'the rubix cube of all stories'.
Interesting as ever. I'll be reading. I'll have too, with such 'tirgid' thinking, you can't just rush through these thoughts, like an advertisement.
It's a good book actually, I've decided to go through most of my library and keep a store of the salient points on the blog for future reference - to save on paper cuts and eyestrain.
:)
Post a Comment