Privacy as Protection of the Incomputable Self: From Agnostic to Agonistic Machine Learning

tags
Predictive Analytics Mireille Hildebrandt Privacy

Notes

understanding of privacy that is capable of protecting what is uncountable, incalculable or incomputable about individual persons

NOTER_PAGE: (1 . 0.3643939393939394)

relational nature of privacy

NOTER_PAGE: (1 . 0.421969696969697)

productive indeterminacy of human identity

NOTER_PAGE: (1 . 0.44166666666666665)

ecological understanding of privacy

NOTER_PAGE: (1 . 0.4621212121212121)

protection against overdetermination of individuals by machine inferences

NOTER_PAGE: (2 . 0.3431818181818182)

An environment either affords us privacy or it does not

NOTER_PAGE: (2 . 0.4492424242424242)

Including technological environments

magical thinking has taken hold of the public imagination

NOTER_PAGE: (3 . 0.15227272727272728)

behaviorist assumptions of machine learning (which necessarily reduces human interaction to behavioral data) impact human identity in relation to privacy

NOTER_PAGE: (3 . 0.478030303030303)

Relational Conception of Privacy and Identity

NOTER_PAGE: (5 . 0.17196969696969697)

Privacy relates to the foundational indeterminacy of human identity

NOTER_PAGE: (5 . 0.2037878787878788)

Arendt understands human freedom as a practice of speaking rather than calculating, of acting rather than behaving, and of facing the uncertainty of being (mis)understood in one way or another.

NOTER_PAGE: (7 . 0.3234848484848485)

continuous need to learn and to review what is learnt

NOTER_PAGE: (7 . 0.4409090909090909)

“Old programs do not learn, they simply fade away. So do human beings, their undebuggable programs replaced by younger, possibly less tangled, ones in other human heads.”

NOTER_PAGE: (8 . 0.11212121212121212)

performativity that cannot be understood in the computational terms of performance metrics or mathematical optimization

NOTER_PAGE: (8 . 0.5545454545454546)

Are machines capable of illocutionary speech?

human beings are always in the process of — incrementally and/or radically — reinventing themselves and their shared world

NOTER_PAGE: (9 . 0.15151515151515152)

Incomputability

NOTER_PAGE: (9 . 0.21742424242424244)

incomputability refers to a specific type of decidability, meaning that it is impossible to develop “a single algorithm that always leads to a correct yes-or-no answer.”

NOTER_PAGE: (9 . 0.2863636363636364)

the formalization that is necessary to turn real life events into machine-readable data (including programs) necessarily results in uncertainty at the level of mathematical decidability

NOTER_PAGE: (9 . 0.3643939393939394)

no machine learning algorithm will necessarily provide optimized output on new data

NOTER_PAGE: (9 . 0.4583333333333333)

whether real life events can be formalized in the first place

NOTER_PAGE: (9 . 0.49696969696969695)

as with every translation, something gets lost

NOTER_PAGE: (9 . 0.6318181818181818)

avoid mistaking the translation for what has been translated

NOTER_PAGE: (10 . 0.11515151515151516)

The Map Is Not the Territory

The temporality that grounds us and the natality it entails confronts machine learning with the fundamental uncertainty of the real world

NOTER_PAGE: (10 . 0.3795454545454545)

this particular first-person perspective cannot be formalized

NOTER_PAGE: (10 . 0.7068181818181818)

Ecological Understanding of Privacy

NOTER_PAGE: (11 . 0.27045454545454545)

incomputability of human identity is, however, not a bug but a feature.

NOTER_PAGE: (11 . 0.30303030303030304)

an important text by Philip Agre, on privacy and “capture.”

NOTER_PAGE: (12 . 0.26515151515151514)

data-driven systems reconfigure their environment to gain access to more data, turning both our environment and ourselves into data engines

NOTER_PAGE: (12 . 0.38106060606060604)

As human activities become intertwined with the mechanisms of computerized tracking, the notion of human interactions with a “computer” — understood as a discrete, physically localized entity — begins to lose its force. In its place we encounter activity systems that are thoroughly integrated with distributed computational processes.

NOTER_PAGE: (12 . 0.5431818181818182)

Since we are an important asset within the environment of these ICIs, we must be reconfigured in ways that enable the capture of behavioral and other data “from” us

NOTER_PAGE: (13 . 0.24696969696969698)

aims . . . . are not political but philosophical, as activity is reconstructed through assimilation to a transcendent (‘virtual’) order of mathematical formalism

NOTER_PAGE: (13 . 0.38257575757575757)

statement made by Mark Zuckerberg: “I’m also curious about whether there is a fundamental mathematical law underlying human social relationships

NOTER_PAGE: (14 . 0.13787878787878788)

Connection between rationalization and Facebook's Record Everything approach

There are elements of totalitarianism in some of these assumptions, notably where they prefer bits to atoms and mathematical theory to the reality we actually face

NOTER_PAGE: (14 . 0.3189393939393939)

right to privacy as the effective and practical remedy to protect what counts but cannot be counted

NOTER_PAGE: (14 . 0.42272727272727273)

collision of phrase with Counting the Countless

Agonistic Machine Learning

NOTER_PAGE: (14 . 0.6694502435629784)

no system can be trained on future data

NOTER_PAGE: (17 . 0.25191370911621436)

whereas algorithms can be optimized to learn specified tasks, this never implies that the optimization works on new data or with a view to another task

NOTER_PAGE: (17 . 0.42171189979123175)

issues of bias are inherent in machine learning and must be understood at the level of its methodological integrity

NOTER_PAGE: (20 . 0.30410577592205984)

we need some form of prejudice to even begin to understand whatever it is we face

NOTER_PAGE: (21 . 0.15309672929714685)

training data is necessarily biased

NOTER_PAGE: (21 . 0.34377174669450244)

uncover whether the bias is a computational artefact (bug) in the dataset, or a pattern in the world of atoms and meaning

NOTER_PAGE: (21 . 0.3827418232428671)

If no purpose was defined prior to the collection of data, then the data should not be used.

NOTER_PAGE: (23 . 0.11482254697286014)

His interest was the validity, relevance and accuracy of inferences

NOTER_PAGE: (23 . 0.16562282533054976)

methodological integrity of machine learning requires advance specification of the purpose

NOTER_PAGE: (23 . 0.2303409881697982)

elasticity, ex-centricity and ecological nature of the inner mind are what makes us human, but thereby also vulnerable to being hacked by an environment that is conducive to cognitive automation.

NOTER_PAGE: (23 . 0.651356993736952)

“agonistic machine learning,” i.e., demanding that companies or governments that base decisions on machine learning must explore and enable alternative ways of datafying and modelling the same event, person or action

NOTER_PAGE: (24 . 0.2860125260960334)

this type of machine learning or cognitive computing is parasitizing on human domain expertise or simply on human experience.

NOTER_PAGE: (25 . 0.11413454270597129)

the better the simulation, the higher the risk that it will follow the bias that is hidden in the ground truth

NOTER_PAGE: (25 . 0.2290249433106576)

“ground truth” itself is often contestable

NOTER_PAGE: (25 . 0.2668178382464097)

By requiring the specification of one or more legitimate purposes by the controller, data protection law unwittingly contributes to sustainable research designs

NOTER_PAGE: (30 . 0.5184411969380655)

Notes that link to this note