Thinking difference and indifference produced through ... file1. Classification through machine...
73
Katja de Vries – PhD student at Vrije Universiteit Brussel – 3 May 2012 4th ICTs and Society-Conference 2012 Uppsala Image source: http://icep.wikispaces.com/Cluster+Analysis+and+Diversity+Analysis Some are more equal than others. Thinking difference and indifference produced through automated, algorithmic profiling.
Transcript of Thinking difference and indifference produced through ... file1. Classification through machine...
Katja de Vries – PhD student at Vrije Universiteit Brussel – 3 May 2012
4th ICTs and Society-Conference 2012 Uppsala
Image source: http://icep.wikispaces.com/Cluster+Analysis+and+Diversity+Analysis
Some are more equal than others. Thinking difference and indifference produced through
automated, algorithmic profiling.
Out of place: Heidegger’s Sein und Zeit as night stand book
Out of place: legal philosopher (<-> communication scientists & political scientists)
Similarities: green left
Similarities: female
Similarities: philosophical question as point of departure
1. Classification through machine learning/ data mining 2. Concerns about machine profiling of human behavior 3. Back to the old philosophical questions 4. How machine learning sheds a light on some old philosophical questions & opens new possibilities of thought 5. A double route: philosophical experience of ambivalence & protection of the grounds of taboo
“THE MNIST DATABASE” See: http://yann.lecun.com/exdb/mnist/ and LeCun et al (1998), Gradient-based learning applied to document recognition, Proceedings of the IEEE. Available online at: http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf Image source: exerpt from the MNIST database as reproduced in MacCormick, J. (2012). Nine Algorithms That Changed the Future. The Ingenious Ideas That Drive Today's Computers. Princeton: Princeton University Press, p.83
Classification or pattern recognition based on machine learning/data mining: statistical model generated by a machine-exectued algorithm
Image source: http://www.nec-labs.com/research/machine/ml_website/
Algorithm: a “precise recipe that specifies the exact sequence of steps required to solve a problem” (MacCormick, 2012).
Top Ten Algorithms in Data Mining 1. C4.5 2. K-Means 3. Support vector machines (SVM) 4. Apriori (association) 5. Expectation Maximalization (EM) 6. PageRank 7/8/9. (tie) AdaBoost, k-nearest neighbours (kNN), Naïve Bayes 10. Classification and Regression Tree (CART) International Conference on Data Mining (ICDM) in December 2006: http://www.cs.uvm.edu/icdm/algorithms/index.shtml
Descriptive/unsupervised learning*: Clustering K-means Association Expectation maximization (PageRank) Predictive/supervised learning*: Classification C4.5 CART K Nearest Neighbours (kNN) Naive Bayes SVM AdaBoost * This distinction is up for debate (predictive methods are often also descriptive, equating “description” with clustering around an unknown class attribute might be a bit of a stretch sometimes, Pagerank is just very different from the rest, etc., etc. ) but it offers a first conceptual tool to think about the field of statistical modelling algorithms.
Distinction descriptive –predictive is not so sharp: Classification tree (probability of surviving the Titanic)
Image source: http://en.wikipedia.org/wiki/File:CART_tree_titanic_survivors.png
Clustering intuition: Minimize the variation within cluster, maximize the variation between clusters
K-means
Image source: Wu et al (2008). Top 10 algorithms in data mining, Knowl Inf Syst 14:p. 1–37 Online available at: http://www.cs.uvm.edu/~icdm/algorithms/10Algorithms-08.pdf
Descriptive/unsupervised learning: Clustering K-means Association Expectation maximization (PageRank) Predictive/supervised learning: Classification C4.5 CART K-Nearest Neighbours (kNN) Naive Bayes SVM AdaBoost
Image source: MacCormick, J. (2012). Nine Algorithms That Changed the Future. The Ingenious Ideas That Drive Today's Computers. Princeton: Princeton University Press, p. 87 Original data source: http://fundrace.huffingtonpost.com/
1. Classification through machine learning/ data mining 2. Concerns about machine profiling of human behavior 3. Back to the old philosophical questions 4. How machine learning sheds a light on some old philosophical questions & opens new possibilities of thought 5. A double route: philosophical experience of ambivalence & protection of the grounds of taboo
Sharma, R., Moon, H., & Jung, N. (Filing date: Dec 17, 2007 Publication date: 8 May, 2008). U.S. Patent, Application No. 12/002,398, “Automatic detection and aggregation of demographics and behavior of people”.
Customers Who Bought This Item Also Bought
Reassembling the Social:
An Introduction t... by
Bruno Latour
(1)£14.24
Conversations on
Science, Culture, and
Tim... by Michel Serres
(2)£12.95
Inspired by Your Shopping Trends
The Roads to Modernity:
The British... Paperback by
Gertrude Himmelfarb
£7.19
Harry Price: The Psychic
Detective Hardcover by
Richard Morris
£18.99 £14.49
The Secret Life of Houdini
Hardcover by William
Kalush, Larry...
£18.99 £12.34
Image source: http://www.cs.ucf.edu/~ahakeem/projects.html
Image source: Bo Wu, Haizhou Ai, Chang Huang, "Facial Image Retrieval Based on Demographic Classification," icpr, vol. 3, pp.914-917, 17th International Conference on Pattern Recognition (ICPR'04) - Volume 3, 2004 Online available at: http://media.cs.tsinghua.edu.cn/~imagevision/papers/204_WU_B.pdf
Image source: http://www.videomining.com/technologies/main.html (Website of VideoMining Corporation)
Categorization
Image source: http://www.videomining.com/technologies/main.html (Website of VideoMining Corporation)
Resistance: 1. Fundamental rights (EU) 2. Social science
1. Fundamental rights (EU)
EU Anti-Discrimination Law
• Treaty level: – Chapter III (art. 20-26) of the EU Charter of
Fundamental Rights (2000)
– art. 18-25 Treaty on the Functioning of the European Union
General principle of equal treatment: Arcelor v. Prime Minister, C-127/07, Judgment of 16 December 2008.
EU Anti-Discrimination Law
• Treaty level: – Title III of the EU Charter of Fundamental Rights
– art. 18-25 Treaty on the Functioning of the European Union
• Directive level: – 2000/43/EC: race in employment, social welfare,
access to goods & services
– 2000/78/EC: religion, belief, age, disability, sexual orientation in employment.
– 2006/54/EC: gender in employment
– 2004/113/EC: gender in access G&S.
– CFD 2008/913/JHA: on combating certain forms and expressions of racism and xenophobia by means of criminal
Anti-discrimination regarding protected grounds ECJ, Test Achats v. Council, C-236/09, Judgment of 1 March 2011. Huber v. Germany, C-524/06, Judgment of 16 December 2008
Image source: http://manipuronline.com
2. Social Science
Image source: http://sassafras.typepad.com/.a/6a00e551a9430888340133f35deb6f970b-500wi
Machine learning is thorougly dialectic – appropriating negativity, incorporating resitance: “the dialectical process is set in motion when an experience loses its threatening negativity through being conceptualized and incorporated [….] to confront fresh experiences which are incorporated in their turn. Learning through experience….”
Resistance against a dialectic-learning mechanism?
Oudemans, T. C. W., & Lardinois, A. P. M. H. (1987). Tragic ambiguity: anthropology, philosophy and Sophocles' Antigone. Leiden: Brill, p.46
1. Classification through machine learning/ data mining 2. Concerns about machine profiling of human behavior 3. Back to the old philosophical questions 4. How machine learning sheds a light on some old philosophical questions & opens new possibilities of thought 5. A double route: philosophical experience of ambivalence & protection of the grounds of taboo
Postpone the resistance for a while (we can return to that later) and ask:
What can machine learning algorithms tell us about the ancient philosophical conundrum of how to think difference and equality?
Classification or pattern recognition based on: statistical model generated by a machine-exectued algorithm
Source image: http://www.nec-labs.com/research/machine/ml_website/
1770-1831 1646-1716
I. Dialectics
A = A
A ≠ A A=A
A=A =
"The modern idea of equality and the normative obligation toward individuality are subject to an irresolvable dialectic: they only exist in their transition into their opposite“ Christoph Menke, Reflections of Equality. Translated by Howard Rouse and Andrei Denejkine. Stanford, California: Stanford University Press, 2006, pp. 7-8.
Difference and equality (≈similarity): always a dialectic
a. We never perceive pure difference – just opposition of equalities. b. Equality (identity, similarity) is always constructed against difference.
A = A
A A
a. We never perceive pure difference – just opposition of equalities. b. Equality (identity, similarity) is always constructed against difference. c. As such the act of harmonization is violent.
“Hegel's dialectics resemble a procedure of controlled ambiguity. As in the case of every other ambiguous ritual, the adjustment it brings about can also be considered a violent expulsion”.
Oudemans, T. C. W., & Lardinois, A. P. M. H. (1987). Tragic ambiguity: anthropology, philosophy and Sophocles' Antigone. Leiden: Brill, p.216
II. Probabilistic ratio (or: reading Leibniz against the grain)
a. Principle of non-contradiction b. Principium rationis (principle of reason)
A ≠ A A=A
Every predicate is already in the subject: The apple is red
A = R
Textbook Leibniz – the Leibniz of hard determinations
Uti minus malum habet rationem boni, ita minus bonum habet rationem
mali.
A lesser evil has a proportion (‘ratio’) of goodness, so a lesser good has a
proportion (‘ratio’) of evil.
Leibniz, Discourse on Metaphysics, section 3
Nihil est sine ratione sufficiente, cur potius sit, quam non sit
Nothing is without a sufficient ratio why it rather exists than non-exists
Christian Wolff, Philosophia prima sive Ontologia (Frankfurt, 1730), § 70
Partial, probabilistic implication
1. Classification through machine learning/ data mining 2. Concerns about machine profiling of human behavior 3. Back to the old philosophical questions 4. How machine learning sheds a light on some old philosophical questions & opens new possibilities of thought 5. A double route: philosophical experience of ambivalence & protection of the grounds of taboo
Source image: http://www.nec-labs.com/research/machine/ml_website/
2
=
1
3
0.1
0.7
0.2
3
Image source: http://www.un.org/democracyfund/images/open%20window.jpg
Principle of non-contradiction Determining principle of ground
Principle of non-contradiction Probabilistic ratio
Image source: http://members.chello.nl/~c.vandalen1/Lenig%20worden%20kan%20alleen%20maar%20door%20iedere%20dag%20op%20de%20juiste%20manier%20te%20rekken_bestanden/image020.jpg
Double route that allows politics to re-enter: a. The route of taboo-grounds
b. The philosophical possibility of experiencing the constructed, violent nature of the difference-equality dialectic – and its ambiguity “outside” the dialectic.
Out of place: Heidegger’s Sein und Zeit as night stand book
Image source: http://www.videomining.com/technologies/main.html (Website of VideoMining Corporation)
2
=
1
3
0.1
0.7
0.2
3
There are plenty of “technical” problems that border with classical philosophical questions on how to abstract/generalize/induce: Concept shift, overfitting, etc.
