Swarm Intelligence Flashcards
What does this lecture look at?
when and how groups make better decisions than individuals (and also when the reverse is true…)
Where do the swarm intelligence experiments come from?
the labs of Stephen Pratt (Arizona State University) and Takao Sasaki (University of Georgia)
What is swarm intelligence?
A group of individuals processes information together and makes “sophisticated” decisions
What questions are asked in this lecture in regards to swarm intelligence
- What are the underlying mechanisms?
- In what sense are these decisions sophisticated?
What is a previous lecture’s example of continuous choice swarm intelligence
Fairgoers guessing the weight of an ox
What is a previous lecture’s example of multiple choice swarm intelligence
Game Show:
Can ask one “expert” or the audience
Single “expert” = 65% correct
Audience = 91% correct
What is another previous lecture’s example of continuous choice swarm intelligence
Ants mark paths walked with pheromone
Ant colonies can solve complex spatial optimization problems
What are four human oriented examples of swarm intelligence
- User-driven content
- Open access interactive designs
- Prediction markets
- Reviews and reputations
What is another terms for the many wrongs principle
Central limit theorem
What is the basis of the many wrongs principle
Continuous choice
Group decision = mean
As group size increases, accuracy increases
Describe Condorcet’s Jury Theorem
Binary/Multiple choice (e.g. true or false)
Group decision = majority
As group size increases, accuracy increases
What species does this lecture look at?
Temnothorax rugatulus
Describe what happens when temnothorax rugatulus nests are destroyed
Colony needs to find a new site to relocate to (“house-hunting)”
What is the purpose of temnothorax rugatulus paint markings
Help them identify and track each individual ant in the colony
Describe temnothorax rugatulus nest preference
Consistent nest preferences for nests with smaller entrances
Describe what happens after temnothorax rugatulus nest is destroyed
triggers an emigration process:
~1/3 of colony members go out scouting for potential new nest sites. If they find one they like, they return home and start recruiting others to it.
What are the two kinds of recruitment in this species?
Tandem run
Transport
Describe this species recruitment process
Recruitment is conditional on nest quality
-> more ants are recruited to the better nest option.
Ants first recruit by tandem running, but then switch to transport once “quorum” number of ants is reached in new nest.
Quorum is reached faster in better nest; this speeds up recruitment further; colony settles in better nest (i.e., a collective decision has been made).
What is the ants algorithm?
Positive feedback (how good a nest is influences scouts’ recruitment efforts)
Quorum-sensing (size of quorum can be adjusted depending on urgency of emigration)
Can individual ants make a choice for preferred colony?
yes
How do individual ants’ choices for preferred colony compare to general group preferences?
They have the same preference
What are the four useful features of this species
- Small colony size
- Colonies have consistent preferences and make consensus decisions
- Observable recruitment behavior
- Isolated individuals can also make decisions (and have the same preferences as colonies)
What is the benefit of small colony size
Easy to identify & track each ant
What is the benefit of consistent preferences/consensus decisions
Allows us to measure the accuracy of their decisions
What is the benefit of observable recruitment behavior
Allows us to record how information is transmitted among group members
What is the benefit of isolated individuals making decisions/having same preferences
Allows us to directly compare the performance of colonies and individuals
What does the first case study of this species focus on
Emergence of group rationality from irrational individuals
Describe the Decoy Effect
- An example of irrational decisionmaking (failure to choose optimal option)
- Relative (rather than absolute) evaluation of available options can lead to irrationality
What do classical optimization models of animal behavior assume (and what is the reality)
Individuals are “fitness maximisers” and hence should make rational choices
Reality: Individuals of multiple species have been shown to make irrational choices
Describe the first case study
Ants are giving decoy A (smaller area, lighter) and decoy B (darker area, bigger)
What were the expected results of CS1
if individual ants are rational, they should choose Decoy A and Decoy B equally
If individuals are not rational, ants in the A group should choose Decoy A more than Decoy B and ants in the B group should choose Decoy B over Decoy A
What were the results of CS 1
Individual ants are irrational, but colonies are rational
What were the conclusions of CS 1
Rational decisions emerge from irrational decision makers because lone individuals make relative evaluations, but in colonies each individual only makes single-evaluations
What did Case Study 2 focus on?
Do colonies have a larger cognitive capacity than individuals?
What is the overarching theme of CS 2
More choices is not always better… leads to cognitive overload
What are groups able to do?
Choose best option without individuals making direct comparisons
How do groups choose best option without individuals making direct comparisons
Individuals visiting the better nest are more likely to recruit others than individuals visiting the worse nest
Describe CS 2
Ants given either two target nests or eight target nests
What are the expected results of CS 2?
If there is cognitive overload, they will make more good colonies in the 2 option condition and will make an equal number of good/bad colonies in the 8 option condition
If there is no cognitive overload, they will make more good colonies in both conditions
What were the results of CS 2
Individual ants experienced cognitive overload, colonies do not
What else was found in CS 2
Individual ants NOT in colonies visited more nests than individual ants in colonies
What were the conclusions of CS 2
Colonies have a larger cognitive capacity than individuals because colonies distribute cognitive tasks among individuals
What did CS 3 look at
When and how do individuals outperform groups?
Describe the psychophysics of brightness discrimination
Alpha = detection point
Higher alpha = worse detection
Lambda = maximum accuracy
Higher lambda = higher maximum accuracy
Describe CS 3
There was a darker (constant) nest, a middle lit (comparison) nest, and a brightly lit (home) nest
What were the results of CS 3
Performance of individual ants improves as difference increases
For small differences, performance of colonies exceeds that of individuals
For large differences, individuals outperform colonies
What were the conclusions of CS 3
Colonies can more precisely discriminate options than individuals
Individuals perform better than colonies when the choice is easy
