Task 9 Flashcards
Collective intelligence
groups can deliver more accurate factual estimates and creative solutions
- greater access to problem solving resources
- enhanced group judgement
- capacity for teams to evolve and thrive
Improving quantitative judgements (Collective intelligence)
accurate estimates can be achieved by aggregating individual judgements by averaging
Wise Crowd Judgement: relies on a large set of independent opinions
PROBLEM: many real world decision making contexts are messy and simple rules of crowd wisdom may not hold
a) correlated info: large groups are subjects to biasing effects of correlated info: social interactions or shared resources inform their opinions (judgement accuracy better when opinions are negatively correlated)
b) Systematic bias: arise from entrenched, value-based positions or anchoring on previous judgements
- what helps: education in logic, changing problem representation, immediate feedback and practice
METHODS to improve accuracy:
- Delphi protocol: experts interact through facilitator who provides feedback about others estimates
- Idea protocol: use guided social interaction to avoid bias
Distributed processing (collective intelligence)
Complex problem solving: coordination of cognitive resources and mental activity
SWARM intelligence: enables dynamic and unified group work and decision making - outperforms experts
Crowdsourcing: practice of obtaining info or input into task by enlisting service of a large number of people, paid or unpaid (still relies on some central control)
Problems with crowdsourcing:
- produce problematic or ineffectual solutions
- may not be egalitarian
- may produce unbalanced views
- difficult to minor free riding and manipulation
- may require active promotion of collective learning and tailored feedback
Small group collaboration (CI)
Small, collaborative consensus-seeking group is the most popular forum for decision making.
PROBLEM: Group thank may compromise decisions by introducing dominance effects and correlated judgments
- biases, social status, different thinking styles and social processes may influence group decision
Measure and predict a groups collective intelligence:
- A single underling factor (C) explainer 43% of group performance variability (predicts group achievement, while individual intelligence doesn’t)
- group performance depends on soft-skills: correctly and appropriately respond to mental states
- intermediate level of diversity in thinking styles is good in small groups: too much diversity = conflict
- qualitative aspect of group communication can improve CI: shared mental model and shared info (hidden profile: bias towards only sharing and discussing already shared info and avoiding private info)
Collective intelligence in conversation:
Use of structured elicitation methods may increase rigor with which decisions are made
PROBLEM: most researchers don’t use a structured approach
- expert decision making requires testing in a wide variety of environmental decision making
- who shall be consulted?
Crowdsourcing initiatives in conservation are improving
- important to coordinate effort and make data freely available, give feedback to contributors
- lack of systematic evaluation: better understanding of mechanisms and supporting success
measure collective intelligence in primary healthcare teams
Healthcare teams must constantly adapt to changing contexts, create solutions and perform a wide array of tasks
Aim: develop framework to study CI in healthcare teams
- CI: prevent repetition of past mistake, work more efficiently etc
Method: 3 -step procedure
- find existing framework (didn’t work in that case as there is none)
- develop initial framework
- test and redefine framework
Results: focus on 2 interrelated frameworks (initially related to educational fields)
2 encompassing dimensions were extracted:
a) collective dimension
b) intelligent dimension
Domain was subdivided into clinical and organizational domain.
Swarm intelligence (in which characteristics do social insects have to be successful?)
Collective behavior that emerges from a group of social insects
- math models to describe their behavior and apply it to businesses
Social insects have to be successful in 3 characteristics:
- flexibility: colony adapt to changing environment
- Robustness: when 1 fail, group can still perform
- Self-organization: activity neither centrally controlled nor locally supervised
- business executives find the third point often hard: through self-organization the group behavior emerges from collective interactions of all individuals
Foraging solution (swarm intelligence)
Ants find shorts path to a food source by laying and following chemical traits
- nest mates attracted to path with highest concentration of pheromone
Basic rule:
- lay pheromone
- follow trait
Example: telecommunication network calls go through many intermediate nodes before coming to target point. Routing of such calls are based on ant foraging:
- software agents leave digital pheromone to reinforce pats through uncontested area
- mechanisms exporting digital pheromone enables program to adjust to changing conditions (too crowded route = abandoned while searching for a new one)
Dividing tasks (swarm intelligence)
the way insects allocate labor hold valuable insights: honeybee colony has special individuals for certain tasks
Basic rule:
- individual performs a task for which it is specialized
- unless it perceives an important need to perform another function
Example: ants carrying food to their nest:
Bucket brigade: Ants pass food down a chain,, but they are not stationery and transfer points aren’t fixed
Applied to companies: in a concern each worker was responsible for one part of order only, so that the next person couldn’t begin until the first person was done
PROBLEM: different people work in different rates and customer demands make it difficult
SOLUTION: bucket brigade was successfully applied to concern
Simpler rules rule
Complex collective work can emerge from individuals following simple rules
Finding simple rules to shape an organization isn’t always easy:
- unpredictable and counterintuitive behavior can arise from very simple rules
- minor change in rule can radically alter group behavior
- predicting group behavior with simulation modeling
Raiding new markets (swarm intelligence)
Swarm intelligence may hold important lessons for business seeking to find and exploit new markets
a) mass recruitment: laying pheromone to attract a large group of ants
- large colonies who can defend their foodsource
b) tandem-recruitment: ant vibrates its antennae to convince one other ant to follow to food side
- small colonies who cannot defend themselves but need to be flexible
c) Group recruitment: ant vibrates its antennae to get a number of nest mates to follow
- medium sized colonies use this in fast changing unpredictable environments - flexible
Implications for markets:
volatile, short-lived markets with competition = medium sized enterprises
For group recruitment to succeed, components should:
- explore new opportunities while exploit existing ones
- enable person with idea to recruit others
- Allow people to be recruited
- Let system self-select the best ideas
- Support winning ideas with resources
Crowdsourcing
works especially well for certain kinds of tasks: fast to complete, low cognitive load, low entry barriers, objective, verifiable, can be broken up to independent sub-tasks
Crowdsourcing subjective tasks: user rating, surveys, opinions etc
PROBLEM: Gaming the system: provide low effort or random answers to reduce work needed to be rewarded
SOLUTIONS IN REAL LIFE:Social norms and sanctions are largely absent, as workers are interchangeable and usually identified only by their worker ID
- no contracts and the reputation doesn’t matter
SOLUTIONS ONLINE
1. change of task design: same amount of work to enter an invalid and believable response as a valid
2. Signal workers that output is monitored and valued
Experiment: complete 3 questions with varifiable quantitative answers and provided 4-6 keywords summarizing the article
- by answering that, workers had already a reasonable judgment of quality of the article
- placed before subjective questions (invalid comments dropped from 49-3%)
Collaborative crowdsourcing
usually crowdsourcing happens independently and there is no ability to coordinate
Can we assume that workers would help each other without financial incentives?
STUDY COLLABORATIVE TRANSLATION: Workers were asked to work interactively on a translation of poem on open-source platform for real-time editing
- improvement of completion of task: when it was already seeded with a couple of sentences that were contributed by ‘others’
- 0,15 cents for contribution
RESULTS: after 48 hours published and crowdsourced translation was complete (and it was good)
- some workers replaced poem with new one after it was done and others translated it for free
Gains in effort, motivation, coordination and quality can be achieved by letting people work together:
- fun and meaning may be an important component
- maybe also possible with more complex tasks
Different applications for swarm intelligence
Honeybee colony splits when it becomes too large: cooperations could use that to determine when to spin off some of their operations
Insects cluster their colonies dead and sort their leaves: banks could use it to analyze their data for interesting commonalities among customers
Difficulty with swarm intelligence
- people have problem to understand how it works
- Group behavior can be a frightening concept
- Insects and people shouldn’t be described with same frameworks BUT in certain environments they are constraint in the same way
