Slides for my talk for the Web Observatory Launch at IIIT-Bangalore on 17th Feb.
On a hypothesis distinguishing between two kinds of "flow" on the web -- the flow of abstractions and the flow of expressions.
1. Abstraction and Expression on the Web
Srinath Srinivasa
Web Sciences Lab
IIIT Bangalore
http://cds.iiitb.ac.in/wsl
Web Observatory Launch Workshop, IIIT Bangalore, Bengaluru, India,
17 Feb 2015
3. The Free Speech Conundrum
The holy grail of democratic societies – freedom of
speech (and expression) – is suddenly at the center of
a new found controversy
At the core of this debate is a call to distinguish
between “free speech” and “bad speech”
4. Free Speech and Bad Speech
The line is not always clear:
Disagreeing with popular opinion (free speech)
Supporting/opposing a political party (free speech)
Racial slur (bad speech)
Inciting mob violence publicly (bad speech)
Scholarly writing criticizing government or specific religions (free
speech considered bad speech in some places)
Artistic depiction that offends religious sentiments (hotly debated)
5. Characterizing Speech
Claim:
The free speech versus bad speech debate presents a false
dilemma, which can never be completely resolved
Need:
Semantic characterization of speech and conversations and
creating awareness and tool-support for online
conversations based on this characterization
6. Abstraction and Expression
Articulation of our objective
understanding of something
Communicates an idea
ExpressionExpression
Articulation of our
subjective feeling about
something
Communicates an emotion
Abstraction
8. Abstraction
● Semantic construct used to
build our worldview
● Processing is resource
intensive (“System 2” in
Prospect Theory [KT79]
terminology)
● Subject to resistance in
assimilation due to factors like
bounded rationality and
conformance pressures
Images source: Wikipedia
10. Conformance and Diffusion of Ideas
Information diffusion is faster in sparsely connected parts of a network,
rather than densely connected (entrenched) parts due to conformance effects.
Node d in the above figure does not switch to the new idea because of
conformance pressures from nodes e, f and g
Image Source: [Sri 06]
11. Models for Diffusion of Ideas
Typically based on an element of “criticality” that
balances between ability to communicate a new idea
and pressure to conform to existing ideas
Example models [EK 10]
Percolating clusters
Ising model
Cluster density based diffusion
12. Expression
● Semantic construct
encapsulating our emotional
state for communication
● Affects receiver's emotional
state by means of emotional
contagion
● Emotional contagion also
spreads through the web (Ex:
Facebook Experiment [KGH 14])
● Characteristically different from
spread of ideas, which have a
natural resistance due to
conformance effects
Images source: Wikipedia
13. Modelling Spread of Emotions
Models based on spread of epidemics, useful in modeling
spread of emotions
Emotional state entails physiological change in humans
People hence “recover” from emotional states after a while --
akin to the spread of epidemics
Example epidemic models [EK 10]
– SIR (Susceptible-Infected-Recovered)
– SIS (Susceptible-Infected-Susceptible)
14. Mental Model: Linking Abstractions and
Expressions
Axiomatic framework within which we perform reasoning.
Encapsulates underlying assumptions, ground truths and inference rules
Active mental model
Reasoning and deduction carried out within the framework
of the currently active mental model
A sudden change in active mental model usually elicits an emotional
reaction (laughter, terror, etc.)
15. Characterizing Online Communication
Communication (both online and offline) comprise of both abstractive and expressive
elements
The spread of a meme varies greatly if it is spreading due to its abstractive content or due
to its expressive content
Ex: Spread of rumours, primarily due to the emotions they elicit, rather than the content
they embody
Online communication especially prone to this A/E interplay due to online disinhibition
[Sul 04] factors like anonymity, solipsistic introjection, asynchrony, etc.
Specifically, online communication lacks coherence in mental models between
conversationalists.
19. Characterizing Online Communication
Online communication complicated by following factors:
– Lack of coherence between mental models (due to anonymity,
asynchrony, solipsistic introjection, etc.)
– Interplay between abstractive and expressive content in
conversation
Emotions spread faster than ideas
Spread of emotions greatly complicates the spread of
ideas
20. Coherence
Abstraction and Expression can affect group behaviour in
different ways
A given abstraction or expression can gain “coherence” over a
group of people (most people in the group think the same way /
most people in the group feel the same way)
Coherence in abstraction and expression can explain some
failures of crowdsourcing efforts
23. Classification of Groups
Crowds
Group of people having shared attention but no shared abstraction or shared
expression
Rich in insights due to diverse opinions
No major emotional contagion
Members act as individuals
Pose high cognitive load on members
Unstable
Wise Crowds
Share some common abstraction in the form of “ground rules” to facilitate
management of diverse opinions without degenerating
24. Classification of Groups
Herds
Group sharing a common abstraction
“Herd mentality” pertains to every member of the group
thinking in the same way
High in persuasive power
Low on collective insight
Manipulable by external forces if the characteristics of the
herd are known
25. Classification of Groups
Mobs
Groups sharing a common emotional state
Common emotional state could be either positive emotion
(jubilant football fans) or negative emotion (lynch mobs)
Need not have common abstraction (members of an angry
mob may each be venting personal frustrations through the
mob)
Highly unpredictable behaviour
26. Classification of Groups
Gangs
Groups sharing both a common abstraction and common emotion
All members of the group think and feel the same way about
something
Passionate and highly persuasive
Common emotion could be positive (The researcher “gang of
four” on design patterns) or negative (bandits and other organized
criminals)
Powerful and highly impactful collective actions
27. Free speech revisited
What appears as the online free speech conundrum is actually a
complex phenomenon caused by abstraction, expression, dissonance
across mental models and group coherence of abstractions and
expressions
The issue is not (just) a question of what is or should be legal
provisions
The issue is of understanding the cognitive and emotional aspects of
human communication
28. Free speech revisited
The web is affecting who we are as a person – at a very
fundamental level, offering both opportunities and
challenges
Our understanding of web-scale abstraction and expression
dynamics too premature to advocate any form of regulatory
solutions
Specific solution proposals beyond the scope of this talk..
29. May you be born in interesting times...
-- an ancient Chinese curse
Thank You!
30. References
[EK 10] David Easley, Jon Kleinberg. Networks, Crowds and Markets: Reasoning
about a Highly Connected World. Cambridge University Press, 2010.
[KA 79] Daniel Kahneman and Amos Tversky. "Prospect theory: An analysis of
decision under risk." Econometrica: Journal of the Econometric Society (1979):
263-291.
[KGH 14] Kramer, Adam DI, Jamie E. Guillory, and Jeffrey T. Hancock.
"Experimental evidence of massive-scale emotional contagion through social
networks." Proceedings of the National Academy of Sciences 111.24 (2014):
8788-8790.
[Sul 04] Suler, John. "The online disinhibition effect." Cyberpsychology &
behavior 7.3 (2004): 321-326.