130 likes | 152 Views
3 is not a crowd, it ’ s an anecdote. Jon Crowcroft, C2B(I)D ’ 15 http://www.cl.cam.ac.uk/~jac22. Vertices and Edges. We know vertex is person – what ’ s edge?. Relationship Kin, friend, colleague, ship-in-the-night External evidence (registry, dna, HR) Co-lo
E N D
3 is not a crowd, it’s an anecdote Jon Crowcroft, C2B(I)D’15 http://www.cl.cam.ac.uk/~jac22
We know vertex is person – what’s edge? • Relationship • Kin, friend, colleague, ship-in-the-night • External evidence (registry, dna, HR) • Co-lo • Shared air, drink/food, touch? • In same record (video/photo/ticket?) • Communicated • Messaged, liked/mention/comment etc • Wrote a letter…
3 things • Sampling the Crowd? • Lost in the Crowd? • Same old Crowd?
Samples • Selection bias • Has an Interweb Gadget • WEIRD • Recruitment smallworld re-enforcement • Demographic baseline • Reward (and punishment) • Halo effects, Focussing Illusion • Other behavioural economic fails…
Ground Truth of Sample • Takes old fashioned social science • Diaries/Interviews etc, but beware • Representativeness Bias • Priors, sample size • Regression • Retrievability • Imaginability • exponentials, long range dependence etc • General Anchoring effects… http://psiexp.ss.uci.edu/research/teaching/Tversky_Kahneman_1974.pdf
Sample Skew • Sometimes, you’re not allowed … • FluPhone – H1N1 Epidemic • Self report symptoms • Phone app tracks location/encounters • Build spatial/temporal model of SIR…. • IRB ruled • No location (privacy risk) • No kids (informed consent “hard”) • Nulls most the experiment… … …
Hiding in Plain Sight • Or seeing the wood for the trees • Human’s not good at this, but s/w v. good • Re-identification from power of 4 • External data sets • Yellowcab driver and celeb passengers • Massachusetts health records • Fb loan advice • Other stuff already out there….be aware • Human’s are wired to infer stuff from 3 • But society used to be wired to hide… • Stuff that crowdsourcing may reveal • But humans are very bad at some stuff
Hiding in plain sight • Humans are bad at getting • Small World (6 degrees) • Exponentials (2x grains rice on next square) • Large deviations (black swan) • Makes informed consent nonsense • Violate principal of least astonishment • How to train the public? • C.f. Thinking fast, and slow… • Cooling off period & Examples…?
Crowds from both sides now (looking at) • Reproduceable, not repeatable • Science and Business want evidence • To make better decisions in future • So model from data has to have persistence • Hence, needs to be checked • Need to assume no observer bias • as well as sample problems discussed before • So Big Data isn’t necessarily big • Its lots of small, representative, samples • Over time… … … and yet the world changes
Psychohistory • Knowledge you’re being observed can • Lead to change of behaviour • Well known in stock market • Also google flu search term #2 • Also in Crowd funding… • We built tool to “predict” campaign success • If investors all use that tool, I predict: • It won’t work as well • Oh, outcome was roughly what you expect
4 arguments for TV elim… • conditions us to accept someone else’s authority • facilitates consolidated power through the colonization of experience. • physically conditions us for authoritative rule • inherent biases of TV https://en.wikipedia.org/wiki/Four_Arguments_for_the_Elimination_of_Television
Conclusions, Discussion • 4 arguments (for the elimination of crowd*) - I hope you agree these are not independent: • You really can't summarize complex information • Nielson->$ • Interweb->TV • Crowding • Live long and prosper