“Real life is and must be full of all kinds of social constraint – the very processes from which “society” arises. Computers help if we use them to create abstract social machines on the Web: processes in which the people do the creative work and the machine does the administration.” Professor Sir Tim Berners-Lee
So, um, what isn’t a social machine?
If almost any combination of human and computing device can be a social machine, how can we start to understand how these work, without being more specific?
How can we make predictions about success factors with such a general description?
Does a social machine have to incorporate a “machine” in the sense that we might think of a computer, or can machine be used in the wider sense, as in some sense of a Turing Machine; a series of computations?
Can social machines actually cope with the “social constraint” – the “processes from which ‘society’ arises”?
Is it possible to use crowdsourcing to “fight crime”, as Luis Von Ahn has suggested?
In order to explore some of these questions, we look at them within the context of Open Crime Data in the U.K..
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
Crime Apps and Social Machines - Crowdsourcing Sensitive Data
1. Crime Apps and Social Machines -
Crowdsourcing Sensitive Data
Maire Byrne Evans
me1g11@ecs.soton.ac.uk
@maireabyrne
Dr. Kieron O'Hara
kmo@ecs.soton.ac.uk
Dr. Thanassis Tiropanis
tt2@ecs.soton.ac.uk
@thanassis_t
Dr. Craig Webber
cw@soton.ac.uk
https://www.youtube.com/watch?v=Pk7yqlTMvp8
2. “Real life is and must be full of all kinds of social constraint – the
very processes from which “society” arises. Computers help if
we use them to create abstract social machines on the Web:
processes in which the people do the creative work and the
machine does the administration.” Sir Tim Berners-Lee
• So, um, what isn’t a social machine?
• If almost any combination of human and computing device can be a social
machine, how can we start to understand how these work, without being
more specific?
• How can we make predictions about success factors with such a general
description?
• Does a social machine have to incorporate a “machine” in the sense that
we might think of a computer, or can machine be used in the wider sense,
as in some sense of a Turing Machine; a series of computations?
• Can social machines actually cope with the “social constraint” – the
“processes from which ‘society’ arises”?
• Is it possible to use crowdsourcing to “fight crime”, as Luis Von Ahn has
suggested?
• In order to explore some of these questions, we look at them within the
context of Open Crime Data in the U.K..
3. • I was looking at crime data from the Home
Office and found some problems with it.
• It seemed that one solution would be to
crowdsource some of the data.
• This made me start investigating social
machines...
• And even start thinking about defining them.
4. Why is trying to define Social
Machines like herding cats?
• Berners-Lee referred to the 'social constraint'
that these things might overcome.
• But while it's possible to build and observe
them, specification and nice, hard, predictive
science, are a little more elusive.
5. The Mystery of the Disappearing
Crime Data
• The UK Home Office produces crime data, or rather,
distributes crime data.
• Part of the UK Government's Transparency Program.
• Public knowledge of crime - comes from crime data.
• Creates desire for action and drives change.
• Thus the government is held accountable via
transparency.
6. • But there are a few issues:
• How do we measure crime?
• We need to know what crime is, before we
can measure it.
• The question of knowledge of crime.
• The question of recording crime.
• The question of collating the crime data.
• The question of producing crime data in a
timely fashion.
7. Which crimes are reported?
• Certain types of crime are reported to the police because of insurance.
• Police may feel that dominant problems in a neighbourhood are car crime
and burglary.
• Sexual assault, domestic violence.
• Stalking can be hard to quantify.
• When does desire for knowledge of a loved one’s movements become
privacy-threatening surveillance?
• Victim realisation.
• Negative consequences for victims if they report these crimes, not only
from their attacker, but psychologically, morally and socially.
• It is hard to quantify and act on these sorts of crime, given normal police
reporting mechanisms which are geared around the notion of crime as
event (digital), not a process (analogue).
8. How are crimes recorded and
represented on Police.uk?
• Each of the 43 police forces has its own reporting procedures and
practices.
• The Information Commissioner’s Office (I.C.O.) is risk averse with
regard to privacy and the current data protection paradigm
• Police data is anonymised and aggregated with little victim
consultation since geolocation is privacy threatening.
• Data often only arrives at Police.uk after a period of 4-7 weeks.
• The data indicates trends, but is not up-to-date or accurate.
• It can't track crimes.
• Descriptive but not predictive of crime.
9.
10. The Dark Figure
• Victim surveys - the British Crime Survey, (B.C.S.) “dark figure” of
unrecorded crime.
• Only 15% of sexual assault victims report to the police
• Of reported crimes, the conviction rate was around 30%.
• 5% of females have been victims of a serious sexual offence since
they were 16, 20% have been a victim of some sexual offence since
they were 16
• 2.5% of females and 0.4% of males said that they had been a victim
of a sexual offence in the previous 12 months
• In fact, according to victim surveys, the official data is ALL WRONG!
• But policy is built around ‘fear of crime’, a subjective measure, and
which does not align with official police data.
• Do we need to find other ways of creating this data?
11. Crowdsourced crime data?
• http://www.ukcrimestats.com/
• http://www.ushahidi.com/
• https://www.crimereports.co.uk/
• http://www.interneteyes.co.uk/
• http://www.blueservo.net/
• http://www.snapscouts.org/
• (Actually not the last one – it’s a Reductio Ad Absurdum)
• Tip lines
• Crimestoppers
12. The Gendankenexperiment
• Crowdsourced data
• Allows victims control over the process of disclosure
• System is analogue, rather than the digital “either-it-is-
a-crime-or-it-isn’t” of Police open data.
• Might have predictive properties and could even be
used to help prevent crime.
• Expands older, verified, government open data.
• Creates a “grey figure” from more up-to-date less
verified and less formalised data.
• Enables trust in the reporting system.
13. Trust
• Trust is a key concept with reporting some crimes.
• It is recognised that technical architectures can shape
realities.
• A new architecture that re-shapes knowledge and
experience of crime?
• Feeds contextualised knowledge about crime with an
understanding of how current recording systems shape our
knowledge of crime.
• And of course, such a social machine changes the dynamic
of the current transparency regime where KPIs and
performance data are produced by those who are being
held to account with the resulting sometimes tragic
consequences.
14. However - Privacy
• How differently might such a machine be used in Europe and Asia?
• Privacy– vastly different as we traverse the globe, which such an
app could easily do.
• Legal treatments of data that would make a huge social impact if
somehow incorrectly deployed.
• If we have certain expectations of privacy in the U.K. we trust that
our data will not be exposed in a way that reveals our identity.
• We must consider not just “the cyber-infrastructure of high-speed
supercomputers and their networked interconnections, but the
even more powerful human interactions enabled by these
underlying systems.”
• Reporting architecture could potentially be horrifically abusive, if
identities were leaked, lost or let slip.
15. Trust, privacy,
legality and ethics
• How such an app stretches existing social understandings and
norms when it’s global.
• Do we create global systems that impose global standards or
systems that are flexible enough to allow for local interpretations?
• Ushahidi not just lifeblood for solving crime.
• Could potentially spill the lifeblood of those using the system.
Mexican Drug War: http://readwrite.com/2012/08/14/the-problem-
with-crowdsourcing-crime-reporting-in-the-mexican-drug-war
• Anonymity does not depend only on encryption
• Criminal organisations, law enforcement, and citizens are not
independent entities.
• Apprehensions may not lead to convictions.
• Boston bombing – point made that crowdsourced intelligence-
gathering might work, but crowdsourced crime-solving doesn’t.
16. Social Machines & incentives
• So there are some problems, and some benefits to such a machine.
• We saw reasons or incentives for not reporting: Self-identification,
self-blame, guilt , shame, fear of the perpetrator, fear of not being
believed, fear of being accused of playing a role in the crime, lack of
trust in the criminal justice system.
• In the case of a crowdsourced crime-reporting system these
problems or incentives for not reporting are overcome.
• The crowdsourced reporting system helps in creating a machine
that sources such sensitive data.
• Focus on what drives people to use the machine?
• What incentives are there?
• A user asks for help in some way.
17. Anonymous Web
• Victim discloses as little or as much of what has happened as they choose
- turns digital reporting to an analogue process.
• But incentive becomes complex here.
• To understand the mental state of a victim of domestic abuse is a complex
process.
• As stated above, one of the problems with reporting on domestic abuse is
recognition on the part of a victim that a crime has taken place.
• “Knowledge of crime” ebbs and flows in the mind of the victim.
• It is this knowledge that maps into knowledge that is to be captured and
represented by the machine.
• When we talk about goals and incentives, we appear to be talking of a
mental state, goal or intention.
• That got me wondering, “How do we map these analogue states of
knowledge of crime from a crime victim into a definition or
characterisation of a social machine?”
• How they fit with the two approaches to characterisation that I looked at?
18. The top-down approach to
specification
• “Computer mediated social interaction” from Robertson and
Giunchiglia: “Programming the Social Computer”
• Social frameworks provided by humans are so pervasive, given the
ubiquity of personal devices and sensors.
• We must change the way we think about computation and
programming. A social computation is one for which...an
“executable specification exists but the successful implementation
of this specification depends upon computer mediated social
interaction between the human actors in its implementation”.
• Considerations of understandings of incentive structures aligned
with the relevant populations allow us to consider knowledge
representation and formal specifications in new ways.
19. Have we captured the “social”?
• Evolved machines are underpinned with often
perverse, unintended human interactions.
• The to-and-fro of a victim unsure whether or
not they are a victim.
• Is formal specification efficient in trying to
isolate predictors for success where the
"social" is involved?
20. The bottom-up, empirical approach
• Examples of what are generally agreed to be social
machines and see what they have in common in terms
of their inputs, outputs and computational processes,
for example.
• Agreed examples of social machines.
• Local moral judgements were creeping into
specifications. The ‘bad’ echo-chamber, the ‘bad’
spammers, the ‘good’ researchers.
• Organisations of person and machine are used
altruistically or selfishly, by "good" or "bad" people and
speak of “goals” and “intentions”.
• Moral vocabulary = seems unscientific.
21. Genetic variation
• Social machines have some elements of non-random
genetic variation advantageous to characteristics that
enhance survival and reproductive success.
• Each user varies in terms of their intentions/ goals as they
use the machine, and build into it.
• By definition, if the machine continues to survive, then the
variation in the minds of its users as they use it or build into
it has led (truistically) to the machine’s survival.
• “Selection does not have a long-term goal. It starts anew
with each generation, selecting those characteristics that
are advantageous within the environment at that particular
time.”
22. Genetic variation,
epistemological
wrangling
• Makes looking for characteristics that specify the social in social machines hard.
• Depends on the ecological circumstances of their users of whose evolving and
mutating intentions we also cannot speak authoritatively.
1. Neuroscience casts doubt on whether we can relate intention to behaviour at all.
2. Victims of domestic violence do not experience crime as a single, digital, fixed-
state event. Their knowledge of their experience of the crime evolves and
mutates.
3. Devices are more ubiquitous and pervasive. Interactions more intuitive, less
goal-driven and less conscious. Makes analysing intentions very hard.
• Only include devices that people interact with deliberately?
• Perhaps mapping intentions as a form of knowledge representation into system
specifications... ..An act of epistemological wrangling.
23. Turing Machine?
• Is the social machine distinct?
• Perhaps not ontologically or epistemologically viable to refer to
individuals’ goals on a large scale, as something that feeds specification.
• Outward behaviour.
• Intentions may be useful in describing the work of these machines but
may not help define characteristics that enable us to predict which
machines may go viral.
• Look at the overall behaviour of the machine itself as something
ontologically distinct from the inner states of its users.
• The machine’s goal can perhaps be specified as something that is
emergent; defined via network characteristics of users’ behaviour en-
masse.
• Network Science aligns itself easily with large–scale phenomena, accounts
for “genetic” variation and can analyse behaviour of those using social
machines.
• Goals mapped out as emergent exogenous behaviours defined via
network characteristics.
24. Some factors
• Understanding of network characteristics.
• Efficiency.
• Omnivorous use of data, sometimes at scale, sometimes
hugely aggregated.
• Aligning incentives between the social and the machine -
depends to some extent on understanding the element of
intention as defined above.
• Strong and weak – as in AI?
• So is it possible to balance a meaningful discussion of
incentive against the behavioural network science
approach advocated above?
25. Small-scale empirical experiments?
• Explore issues around using Network Science in order to
make predictions about success factors?
• Interviews and discourse-based methods to understand
more the feelings, and “goals” of victims using such a
system that creates knowledge of crime to offset current
open crime data or victim survey data?
• Could such a system show that crowdsourced data can feed
discussions about accountability without becoming mired
in statistics which can often found to be meaningless or
even dangerous?
26. Web Science
• Social machines mediated by Philosophy,
Computer science, Network Science, Psychology,
Criminology , Behavioural economics , Sociology.
• Policy formation mediated by technological
strategists.
Can we create new architectures shaping the spaces of crime and crime reporting?
Build up society’s knowledge of crime and feed decision-making on crime policy.
Discussions on crowdsourcing accountability data to offset statistics generated by
those under scrutiny.
Consider the impacts of what such technologies can do?
Is there hubris in attempting to define large-scale human phenomena?
Goals and intentions in users.
In reducing these phenomena to nodes and edges and make predictions about
success?
How can we do justice? Both to crime victims and people who try to define social
machines? Have social machines really helped with social constraint ?