| |
|
|
Detailed notes to Wisdom of Crowds |
|
Written by Wei-Jing Zhu
|
My running notes, scaffold, to the entire book and audio book.
Collective wisdom:
Intro:
average of the group is better than the best person in the group
ex. weight of a cow
In stock market, voting, etc: stop chasing the expert. Listen to the crowd.
Con: madness of crowds. juries, democracy, ....
Collective intelligence:
sunken submarine location: off onlyh by 220 yds.
Part 1.
ex. Who wants to be a Millionaire - corwd always vote right.
Group estimate of weights,
room temperature,
jellybeans estimate.
agents solving a maze, using group knowledge.
Space shuutle accident - the Market picks the culprit out of the four candidates in 30 minutes. O-ring due to Thyaco.
Diverse independent judgement: errors will cancel out. But of
course, you need to have some nformation in everyone. e.g.
Independent opinions.
Average of running time = mediocrity.
Yet decision making: we are programmed to be collectively smart.
Decision about the future: bookie in sport betting: the crowd predicts highly accurate the outcome.
Google's pagerank as collective voting
IEM poll: predict election market
HSE: Prediction of Hollywood Academy award
Prediction of automobile features
Chap 2. The difference difference makes
Auto industry: many prototypes, by Olds and many others. Diversity get trimmed by the public.
Model: Beehive - exploratory, the list of options are not defined yet, unlike weight guessing.
Diversity is critical. Intelligence alone is not enough.
Grouping only smart people will bring only like-minded people
together. Then the group will know less. Adding newcomers
who know a different set of skills will improve the group mind.
Experts: they spend too much time exploiting than exploring.
Disc 2 Section 3
Note: group wisdom = search problem. diversity = beam size.
Intellignet = top ranks. but still, adding width will improve the
outcome
3. Experts: spectacularly narrow
forecasting future - not trainable. group would make better decision than the most skilled decision maker
Trading - the market is far more superior than even expert stock brokers.
weatherman and bridge player are the only expertise that are better than a group.
Expertise beyond a minimal level is sufficient for group intelligence.
Seer sucker: seers don't exist, but suckers will pay for seeking them.
We try to seek expert, assuming that intelligence reside only in individuals.
Fool by randomness: statistically, in 1000 wars, a random guy will become an undefeated general.
4. Group think
Pay of Pig, Pearl Harbor: homogeneous group - try to be cohesive, insulated from outside opinions, illusion of invunerability
Pressure to conform - easier to change his opinion than changing the group.
Expt: which line is the same. person change his answer. Go
along with the crowd. Yet with one, or a small percentage, of
support, he will become non-conform.
Chap 3.
Army ants march of death. Just follow the ant from the front.
systematic bias: make people dependent on each other.
smart group: diverse and independent.
OK to be biased, but as long as independent, bias will cancel each other out.
People's self interest, autonomous, but we are social beings. Our environment shape our thought.
Social proof: if lots of people do it, there must be something.
This is different from conformity.
ex. Gazing up the sky. Bigger crowd proves, and draw more adherents.
If too many people follow along, the group loses its ability.
ex. NFL decision for field goal vs first down.
Statistics show that conventional wisdom is not crowd wisdom.
Of course, finding the proper crowd wisdom is difficult.
Statistician use past history to find the optimized strategies.
This is extracted from "crowd information", but definitely much harder
to extract than the conventional wisdom.
NFL coach should have been more aggresive.
Easier to follow convention. No one can blame him for being conservative.
No one gets fired for buying IBM
Herding. Even in stock market, brokers herd to keep their jobs.
Better to fail conventionally than to succeewd unconventionally.
Cascade
plank road a roaring success, over building them.
Cascading: not just conformity, peer pressure, or mindless trend following.
Plank roads or Britney CD SEEMS better.
People ignore their own opinions, and look only to the social
proof. ex. Restaurant goers, but the decision due to sequencing -
whoever first decide, speak, or voice their oinion, will influence the
subsequent opinions.
ex. screw - the success of standardized screw was not about better technology,
social norms driven by cascades, but no guaranttee of better quality. Several initializers start the cascade.
Most recently, Internet bubble. Lots of telecom investment. 1000% growth. build lots of traffic.
4. Immitation
Immitation works pretty well in American society.
e.g. weather prediction by looking at others.
parking on street: which side.
monkeys on island imitate
imitation is a great method to spread a good idea fast.
works only when people are willingness to stop when it doesnt make sense anymore.
Need leaders - people who are willing to listen to themselves.
Over-confident people are able to break bad cascade, and lead others by their gut feeling.
This encourage others to listen to themselves.
Iowa farmers adopt hybrid seed corn. Most wait for others to establish the norm.
Fads. No one want to eat in empty restaurant.
Information cascade.
Problem: people decide in sequence.
Should make sure decisions are made simultanesously to gain group wisdom.
ex. Student guess from 2 urns color marbles.
Reward the individual, vs reward them for the group effort.
In the latter, people pay attention to their private information, thus
reduce the influence from each other. By focusing on their
private information, the group as a whole perform much better.
Chap 4
Art of Decentralization
CIA - defragmentation
argues against decentralization, self-managed terams,
Science focus on self-organnized models, e.g. ant colonies
Internet, P2P napster,
Self interested independent people working in decentralized way:
why the CIA not working?
What do we mean by decentralization?
specialization of labor,
diversity
tacit knowledge - context, local information not easily transfered
closer a person is to the problem, the better his ability to solve it.
weakness: no guarantee that valuable information get disseminated.
Linux: centralized aggregation method. decentralized programmers: enough eye balls, all bugs are shallow
lots of duplication, then winnow down
Aggregation mechanism: crucial to the success of crowd
ex. army tactics communicated across ranks, ensuring all squads have the best tactic.
problem with CIA: kind of aggregation that is the problem.
propose Future Map, a market for Intelligence community
self interest will aggregate valuable information to the top.
Chap 5. Coordination
City life. Pedestrians move fast as a dance.
How much milk to store? Coordination problem of how much to go on the store's shelves, distribution problem.
Ex. Whether to go to a bar: too many people, avoid it.
Model whether to go to bar:
agents with different strategies. end up approximating the optimal capacity of the bar.
3. Best place and time to meet in NYC.
Shelling focal point.
Convention, culture, traffic patterns
no coercion needed
power of convention;
ex 1. ask people to give up seats, hardest is courage to ask
norm of first come first serve is so powerful
ex 2. jump lines
vehement reaction
responsibility to react goes to the person behind
fear for disruption allows for jump line
allowing for line-cutting shows the resilience of a queue
multilane in supermarket - people misjudge speed in another line
trade places in line for money? violate egalitarian nature
5. company profit motive; should be void of effects of history and convention , but no.
pricing
movie tickets
matinee are not egalitarian
why not different ticket prices? ex. mediocre movies are great DVD hits
Flocks coordinate gracefully, will markets ?
Expt: Class as market: works even with real people, rather than abstract theories, using limited private information
6. Societ exists: taxes, tipping, TV, trust
Italian soccer: people lose trust in the system
public outrage over NYSE CEO large salary
Ultimatum game: people care not just selfish gain, but justice and fairness
Evolution of cooperation - Axelrod
Best strategy for repeated dealings: nice, forgiving, retaliatory
Tipping as a counter example to the theory of "future shadow".
Quakers as trustworthy businessmen lead to their success
Problem with trust: as people trust, set up for being exploited
Nielson rating: collective action trap
people don't want to rock the boat. group becomes dumb
Tax fraud:
expt: people chipping into the cummunity pot to make everyone better off.
This is a simplified form of prisoner dilemma, without the punishment for defection.
people are one of three groups: 25% selfish and free-ride, minority of altuists, and the rest are conditional consenters
3 types of trust:
trust your neighbors will do the reasonable thing
trust that the govt will spend tax wisely
trust that the govt will punish the freeriders
Part II
7. Traffic: failure to coordinate
after charging for busiest hour road tax, alleviate traffic. Why not apply to highways and other traffic problems?
Traffic pattern:
sparse: freeflow
dense: no more gaps, stop and go pattern
cars leaving theh front are slower than the ones coming in from behind, thus a traffic jam that moves backwards
This is example that, unlike the graceful flocks, traffic shows ugly face of the crowd.
Problem is diversity, which works well in decision making (search for
solutions, essentially exploratory and brainstorming), but makes
coordination harder.
Advocate road sensors, autopilot for car flow on intelligent highway.
8. Science collaboration, competition, reputation
speed in solving SARS: collaboration and diversity in exploration
science: reputation is the reward
collaboration: more successful, more likely to collaborate.
reputation grows. Rich gets richer. People listen more to
the reputed.
Columbia disaster committee:absence of dissent, lack of diversity
Polarization occurs, when people constantly compare themselves to everyone else
p186
Polarization
Group tend to polarize to conformity
ex. jury - simply to get a concesus
deferrence to people in higher rank, or talkativeness
no point to have a group just to be consultant. need to give them actual final authority.
expt: students guess ball color from urn, to emulate market condition for interest rate up/down
chap 10: company
Zara clothing: coordinate fashion design, intergrate market
intelligence and customer feedback directly to headquarters, 3 weeks
from design to store
this reduce inventory, sales, thus lower cost, reduce price, ...
outsource: have its suppliers be part of the chain.
Why not outsource all?
Theory of transaction cost: search for personel, negotiate contracts,
build team by HR, ensure trust to and from the group. Takes lots
of work to make a group work
counter-example: Hollywood film making
American corporation: lip service to many management fads, but ultimately a hierarchical org.
Managers are rewarded not by merit, but by meeting self-made expectations. This creates room for abuse.
ex. workers tweek the system, rather than maximize output
Propose decentralization for the corporation: merits
when people create their own environemnt, they are more engaged
coordination: people will find better ways
In-fighting within GM, no sense of the overall company.
People continue to look to CEO, as if he alone matters to the company
Decision markets would enhance the decision making process of a
company. A good CEO should recognize his own limitations, and
listen to the crowd.
Chap 11.
Markets: beauty contest, bowling alleys, and stock prices
Many societies think that short sellers are evil, benefit from other's misfortune. Limiting them causes prices to be off.
Over-price and over-confidence leads to mis-apropriation of resources
Predicting future stock price:
Black: off by less than 100T in 5 years is a good prediction.
LTCM debacle: invest in illiquid markets
Market driven by fear and greed. Rise slowly, drop fast.
Stability of the market depends upon diversity of independent
opinions. - when you sell, someone else thinks it is time to buy.
Herding, polarization, illiquid market, and correlated behavior causes avalanche.
Biggest bubble: pst war bowling alley stock: over creation of bowling facilities.
No bubbles in real markets: don't see TV prices going into
bubble. because when you buy a product, you have independent
judgement of its fair price.
Financial markets: totally dependent on what you think other people
think.
Bubble: look less like a bubble the longer it stays
Beauty contest:
pick who you think is beautiful
pick who you think will win: who you think others think beautiful..
pick who you think who others think who others think beautiful, ...
expt. Caltech class trading "bond" - overpriced, thinking: surely I
know the price is high. but some othe idiot will bid more.
CNBC during dotcom bubble: mere fact that CNBC mentions it will cause rise in stock
MIT expt: two groups of students, one given just that the stock moved
up or down, the other given the entire news. The group with less
information trades better.
News report overplays the significance of a particular incident. Dramatize. Causing people to overreact.
Over-weighting this single piece of information is compounded when everyone else gets the same piece of information.
Information may not be a good thing:
CNBC: misuse of news and position of power.
Expt. guess jellybean count, group guess correctly
then guess again, with the prof pointing out the air space , etc and the group overguess by 15%
Lesson: info from position of power should not over emphasized
Seattle woman jump off a bridge:
the more people, the easier for anonymous.
the more yelling, the more peop;le want to yell
Key to market success: enough independent-minded people. e..g.
when there are people who stand apart from the riot, people will be
more likely to see these examples and not riot as well.
Chap 12. Democracy
deliberative democracy, national day of political polls and debates,
given enough information and chance to talk things over, ordinary people are more than competent to make informed choices
self-interest is not a problem for democracy.
why would anyone vote? a sense of duty and contribution to the whole society already
Found that people don't really vote out of self-interest, but a sense of hope as well.
Expressive voting - enable people to voice themselves
In this book, discuss 3 types of problems: cognition, coordination, cooperation
Democracy is not a way of solving a cognition problem, or a method to
reveal public interest; but a way of dealing with the most important
problem in coordination and cooperation: how do we live together, and
benefit each other?
The decision that democracy makes does not reflect the wisdom of crowds. The decision to make them democratically does.
|
|
|
Home 
Articles 
