From Big Data to Banality of Evil
1 like1,178 views
This document discusses big data and some of the opportunities and challenges it presents. It covers topics like how big data involves the collection and analysis of large, diverse datasets to find patterns and correlations. However, it also notes that there are ethical issues to consider regarding how data is collected and used, and models built from data can be wrong and should be viewed critically. The document advocates thinking carefully about how data is applied to individuals given the statistical nature of models.
From Big Data to Banality of Evil
- 1. Kavé Salamatian Professeur à l’université de Savoie
- 2. Big Data : great expectations
- 3. Big Data is the name for what ? Capture storage of massive amount of data Diverse, heterogeneouse, imperfect Data processing Sanitisation Statistical models Co-occurrences/correlations Uncertainty structure Specialisation Model application to specific individuals Action Intelligent behaviour
- 6. What is new Processing and storage capacity Hardware, databases, langages Data science Statistics, data fusion, Absence of purpose “From application to data” vs. “From data to applications” Massive application to individuals Statistics is for groups But motivatiuon remains the same "scientia potentia est” Money is power
- 7. What about ethics ? Where is ethics ? In code writing? In executing the code? In modelling ?
- 8. On the use of computers in real world
- 9. About statistics There is no statistics but statistical models Some are useful, all are wrong
- 10. From power to “power laws” P(k) ~ k -γ
- 11. S.cerevisiae k n(k) 1 2 3 4 6 8 11 16 23 33 47 67 95 141 220 1 2 3 5 8 16 33 56 111 210 535 1248 Figure 1: Yeast protein interaction data (o) and best-fit probability distribu-tions:
- 12. Seneca already did criticisms of Big data Is not Big Data the haruspicine of nowadays etruscans
- 13. Should we regress when we do regression ?
- 14. A fairy tale Once upon a time there was a complex reality … To understand it people make fable out of it … After a while people believed the fable and forget about the reality Same story frequently happen in science Once upon a time there was a complicated reality After a while people build a model to understand it Later people forgot about the reality and thought that the model is the reality The model became real … and the troll eat the princess J You always need to have critical view about what a model predict Sometimes it is relevant to reality, sometimes not. The role of the scientist is to filter out what is relevant from what is not
- 15. The quest for objectivity Finally what is objective We throw a coin The probability of the model is p Pr{10th throw is face}=p Application to individuals can only be subjective
- 16. This is risky to go in bed with procuste
- 17. Hannah Arendt and the ideosyncrate …
- 18. Hannah Arendt and the ideosyncrate … What is totalitarianism?... A power-hungry group achieves power -- using demagogic means to maintain and expand its unlimited and unchecked power. These means are particularly terrifying and efficient and our time because of technical progress and inventions which permits swift concentration and application of power and manipulation as well as production of public opinion. There are no truly representative subleaders, there are no independent social groups, no traditional rights, no historical contributions and merits which can check the modern absolute totalitarian rulers; it must be emphasized that the totalitarian system differs from old-fashioned despotism and autocracy in its use of economic and technological pressures and manifestations of so-called public opinion. The totalitarian tyrannies of our time do not appeal to the "divine right," but they claim to represent the true will of the masses and of the people. They like to characterize themselves as representing the "real country" against legalistic fictions of democracies which destroy the unity of the people and paralyze it by parties, the instruments of parasitic and egoistic particular interests. They present themselves as "true" democracies because, for instance, only the communistic party (which allegedly is identical with the working productive masses) exists; whereas all exploiters and their representatives are suppressed. . . .
- 19. Invisible hand ? “Ever& individual... neither intends to promote the public interest, nor knows how much he is promoting it... he intends only his own securit&; and by directing that indust=& in such a manner as its produce may be of the g=eatest value, he intends only his own gain, and he is in this, as in many other cases, led by an invisible hand to promote an end which was no par? of his intention.” The wealth of Nations, Adam Smith
- 20. So what’s economy ? “Economy is the stAdy of the use of scarce resources which have alterBative usages “ Lionel Robbins (1898-1984)
- 21. 1 billion $ question How are scarce resources assigned between alternative users ? Price Parity between offer and demand
- 22. Foundations ! Walras Pioneered equilibrium theory Conjectured existence of equilibrium prices Stated that network should be at equilibrium Prices such that supply match demand What happens when ? multiple goods multiple buyers Different desires Different buying power
- 23. Equilibrium Two problems Price are set, how much I can buy ? Such that total utility is maximized Quantities are defined, what is the equilibrium price ? Total utility maximization pi pi pi
- 24. Arrow-Debreu Theorem Established existence of equilibrium in a very general setting Competitive equilibrium is Pareto optimal Every Pareto optimal allocation comes from a Competitive equilibrium
- 25. What about data ? Economy is the science of sharing rare resources Information is different from other goods Follow the second law of thermodynamic No conservation law
- 26. The devil is the details Assumption about goods and services They are rare and finite One cannot reproduce them and negligible cost The added value utility is predictable Unless for innovations or monopoly Information lack all of these It is not rare Creating information is costly but one can reproduce it at low cost The added value is unpredictable Information is very flexible
- 27. 1 Million $ question How can we evaluate the value of a .com ? How can we evaluate the value of private information ? How can we predict network growth based on economical premices?
- 28. A new micro-economical theory ? The insider traders view the value of money that somebody can gain by knowing this piece of information The thermodynamic model the value of information as the price of the energy needed to create it The information market view You can only trade information with information No monetary value
- 29. What is a network ? Seems a trivial question, …, but no clear consensus on the answer. Each community has it own definitions Still network is so universal that they should be commonalities Challenge Can we make a generic definition of network that Integrates all network Still let enough freedom for their differences.
- 30. What is transported by a network ? Some network clearly transport goods Road network, silk road, grid network, metabolic network, etc. We will name them goods network Some network do not transport goods Citation network, social network, etc A lot of these network transport “Information” Internet Indeed, but also citation networks, social network, etc We call them Information network Frequently intertwined with goods network power grid where the network transporting energy is coupled with a control network used manage the energy flow commercial network (like the Silk Road) where goods are transported along with the seller handling the commercial information. Sometime no clear information network Electrical circuit
- 31. Logistic vs. Information networks ? Any Logistic network have or had an associated information network that was used for growing it For some case the information network associated with a goods network disappear later when the network get ossified Where the network is still growing and dynamic the information network is still here. One has to analyse the associated information network to understand the goods network resulting from it. I believe that one can consider as a network, only the information network.
- 32. So … what is a network A network consists of a set of distributed elements that are cooperating to exchange information An information exchange market Several ill defined term Cooperation Information … Internet is a network that strictly exchange information in Shannon terms Other network might transport other type of information Migration network transporting DNA
- 33. Information theory … Is not a theory of coding information Does not tell how much information is send It is the theory of cooperation It limits the rate of uncertainty reduction that a use of a communication channel provides Side information : get the bit that complete the picture Simplicity theory : what is interesting ? What reduce the most the Kolmogorov complexity simplicity corresponds in a drop in Kolmogorov complexity Information theory is a theory of cooperation Error Control Coding tells how to arrange transmission such that each one cooperate with others Separation principle : one can separate transmission from processing Foundation of layering and net neutrality Valid for point to point, not valid for multi-point Cross-layering needed when resources are scarce
- 34. Cooperation ? Full Cooperation ¨ Do the best possible behavior to reach a goal ¤ Assuming full collaboration Is the goal achievable ? How to achieve the goal ? Multi-user information Theory Non–cooperative Selfish behavior Different rational goal How to mitigate conflicting rational goal ? Game theory Malicious behavior Harmful goal How to contain irrational objectives ? Behavioral inference
- 35. What about privacy ? • Privacy is an economic problem… even when privacy issues may not have direct economic interpretation • Privacy is about trade-offs: pros/cons of revealing/ accessing personal information – Individuals – Organizations • … and trade-offs are the realm of economics
- 36. Privacy… But at the price of accuracy § You don’t want to like things that have absolutely nothing to do with your real interests. § In fact, your data are also used by your friends, advertisements, recommender systems, etc… § Preserve usability of the applications that run on your data § If we do not know what application the data will be used for: Need general distortion constraints! Privacy-utility trade-off 36
- 37. Prisoners dilemma A simple game that has become the dominant paradigm for social scientists since it was invented about 1960. Game theoretic problems: payoffs for each player depend on actions of both Two possible strategies: A party cooperates when he performs value-increasing promises, and defects when he breaches While cooperation is collectively rational, defection is individually rational. Player 2 Cooperate Defect Cooperate Both cooperate Player 1 cooperates, Player 2 defects Defect Player 1 defects, Player 2 cooperates Both defect Player 1 Cooperate Defect Cooperate 3, 3 -1, 4 Defect 4, -1 0, 0
- 38. Iterated Prisoner’s Dilemma IPD n-stages; n is known or not Backward induction; Nash equilibrium; Strategies Tit for Tat Tit for Two Tat Suspicious Tit for Tat Free Rider Always Cooperate Axelrod’s Tournament Altruistic strategies does better Networks extend the possibility of cooperation between N actors from O(N2 ) to O(2N) 38
- 39. Net Neutrality Question Net Neutrality is not just a question of access network It has to deal with cooperation between all actors in Internet Content provider/ISP ISP/ISP End Users/ISP End Users/Content Provider End Users/End Users Opacity generate Prisoner Dilemma Defection is the default How to improve cooperation Improve transparency
- 40. Is surveillance an addition to networks or is it intrinsic ?
- 41. Conclusions There is a new science emerging :Internet Science We are still searching for its fundaments and principles This new science is highly multidisciplinary Multidisciplinary research is tough and odd A large part of this work has resulted from discussions and exchange with members of the European Network of Excellence EINS funded under FP7 Fire activity. www.internet-science.eu