Self-imposed Algorithmic Thoughtlessness and the Automation of Crime Control

Name: Self-imposed Algorithmic Thoughtlessness and the Automation of Crime Control | A study of person-based predictive policing and the algorithmic turn
Brand: Nomos
Price: 74.0 EUR
Availability: OnlineOnly

A study of person-based predictive policing and the algorithmic turn

Lucia M. Sommerer(Autor*in)

Nomos (Verlag)

1. Auflage

Erschienen am 11. März 2022

260 Seiten

E-Book

PDF mit Adobe-DRM

Systemvoraussetzungen

978-3-7489-2955-0 (ISBN)

74,00 €inkl. 7% MwSt.

Systemvoraussetzungen

für PDF mit Adobe-DRM

E-Book Einzellizenz

Als Download verfügbar

Beschreibung

Weitere Details

Weitere Ausgaben

Person

Inhalt

Cover
Chapter I. Foundations of person-based predictive policing
A. Introduction
B. Chapter structure
C. Terminology
I. Types of predictive policing
1. Person-based and location-based
2. Suspicion confirming and suspicion generating approaches
II. Crime algorithms beyond preventive policing: court proceedings, the penitentiary system and ex post police investigations
D. Research design
1. Topicality
2. Transformative potential
3. Research gap
E. Technological background
I. Traditional statistical methods
1. Compiling database
2. Defining target variable
3. Defining predictive input variables
4. Deducing a predictive model
5. Risk translation and risk communication
a) Percentages
b) Probability of crime in comparison to the average population
c) Base rate
d) Error rate
e) Cut-off
6. Application to individual cases
a) People affected by crime prognoses
b) Prognosis tool operators
II. Algorithmic predictions
1. Rule-based algorithmic predictions
2. Machine learning algorithmic predictions
a) Differences: Compiling training dataset (step 1)
b) Differences: Defining target variables (step 2)
c) Differences: Defining predictive input variables (step 3)
d) Differences: Deducing a predictive model (step 4)
e) Differences: Risk translation and risk communication (Step 5)
f) Differences: Application to the individual case (step 6)
aa) People affected by crime prognoses
bb) Prognosis tool operators
(1) From prognosis experts to prognosis laypeople
(2) Automation bias
3. Conclusion: technological background
F. Examples of predictive policing
I. US and UK
1. Beware
2. HART (Harm Assessment Risk Tool)
3. Strategic Subject List
4. Summary US and UK
II. Germany
1. RADAR-iTE (Rule-based analysis of potentially destructive perpetrators to assess acute risk - Islamist terrorism)
a) Prognosis tool development
b) Prognosis tool application
aa) Prognosis tool operators
bb) People affected by crime prognoses
cc) Lack of transparency
2. Palantir Gotham - hessenDATA
3. DyRiAS (Dynamic Risk Analysis System)
a) Prognosis tool development
b) Prognosis tool application
aa) Prognosis tool operators
bb) People affected by crime prognoses
4. Flight Data Pattern Recognition
5. Summary Germany
G. Factual limits of person-based predictive policing
I. No overall assessment - big data does not mean all data
II. No causality
III. No neutrality
1. Data selection
a) Lack of data quality
b) The dark figure of crime
c) Lack of training data representativity
d) Perpetuation of societal discrimination
e) Conclusion on data selection and possible solutions
2. Definition of target variables
3. Selection of input variables
4. Calibration and monitoring of the learning process
a) Error rate
b) Overfitting and spurious correlations
5. Summary of no neutrality
H. Conclusion on Chapter I.
Chapter II. Legal limits of person-based predictive policing
A. Constitutional framework
I. Privacy
1. Scope
a) Autonomous decisions about the disclosure and use of personal data
b) Protected data
c) Prohibition of personality profiling
d) Public interest dimension and chilling effects
2. Limits and proportionality
3. Specific limits placed on person-based predictive policing by the right to privacy
a) Presence of high-intensity infringements
aa) Secrecy and lack of transparency
bb) Scope
cc) Automation
dd) Proximity to personality profiles
ee) Stigmatization
b) Consequences of a high infringement intensity
aa) Sufficiently serious offences
bb) Concretely foreseeable danger
(1) Requirements of the BVerfG for dragnet investigations
(2) No lowering of the intervention threshold below a concrete danger
c) Organizational and procedural safeguards
4. Summary of privacy
II. Non-discrimination
1. Introduction
2. Sources of algorithmic discrimination
a) Compiling the training database
b) Defining target variables
c) Defining predictive input variables
d) Calibration and monitoring of the learning process
3. Constitutional law view on algorithmic discrimination
a) Detection of unequal treatment
b) Statistical justifications
aa) Argument of 'rational racism'
bb) Argument of 'lesser evil'
4. Conclusion on non-discrimination
III. Transparency
1. Janus-headedness of the black box terminology
2. Three layers of non-transparency
a) Terminology
aa) Intentional lack of transparency
bb) Illiterate lack of transparency
cc) Intrinsic lack of transparency
(1) Machine learning algorithms
(2) Rule-based algorithms
(3) Examples
(4) Attempt at a solution: Explainable AI
b) Conclusion on transparency layers
3. Transparency fostering mechanisms
a) First dimension of transparency: moment of transparency
b) Second transparency dimension: transparency-entitled entity
aa) Affected citizens
(1) Single individuals
(2) Collaborative crowdsourcing
bb) Government control body
(1) Possible configurations
(2) Personnel resources
(3) Organizational-structural considerations
(4) Conclusion on a government control body
cc) General public
dd) Summary of the people authorized to access information
c) Third dimension of transparency: scope of transparency
aa) Abstract information
(1) Existence and use
(2) General principles of action
(3) Source code
(4) Output testing
(5) Logging obligations in the design phase
bb) Explanation in individual cases
cc) Conclusion on transparency mechanisms
4. Arguments against transparency
a) Impossibility of transparency
b) Unnecessity of transparency
aa) Human in the loop
bb) Comparison with existing non-transparencies
c) Preponderance of competing interests
aa) Trade secrets
bb) Spying on the algorithm
cc) Third-party data protection
dd) Summary: Preponderance of competing interests
d) Conclusion on arguments against transparency
5. The foundations of transparency in constitutional law
a) Human dignity as the core of the right to informational self-determination
b) Democracy principle
aa) Chain of democratic legitimation
bb) Democratic formation of opinion
cc) Democratic control
c) Rule of law
6. Summary of transparency requirements
IV. Presumption of innocence
B. Summary of Chapter II: Legal limits of person-based predictive policing
Chapter III. The algorithmic turn in crime control - a criminological, sociological and legal theoretical analysis
A. Risk: a definition
I. Ambiguity
II. Negative connotations of risk in the risk society
III. Unequal distribution of risks
IV. Social assignment of meaning
V. Quantification of risks: the lure of numbers
VI. Risk perception
VII. Tolerable risks
VIII. Risk vs. danger
IX. Conclusion on risk
B. Status quo: risk-based crime control
I. Examples of risk-oriented norms
II. Security society
III. The actuarial turn
IV. Summary of status quo
C. Development: algorithmic crime control and the algorithmic turn
I. Continuation of risk-based crime control
1. Focus on risk well in advance of actual harm
2. Individualized risk ascription on the basis of group characteristics
3. Enemy perspective
4. Disguising political value judgements
5. Summary of risk-based crime control
II. New challenges associated with the algorithmic turn
1. Loss of legitimacy due to lack of procedural justice
a) Legitimacy and legal compliance
b) Basic concept of procedural justice
c) Procedural justice in algorithmic decision-making
aa) Participation
bb) Neutrality
cc) Respectful treatment
dd) Trust in the motives
d) Transparency as a response to a lack of procedural justice
e) Summary on legitimacy and procedural justice
2. Crime control without criminology
a) Algorithmic criminology
b) Summary of crime control without criminology
3. Self-imposed algorithmic thoughtlessness
a) The social psychology perspective on 'thoughtlessness'
b) Moral disengagement algorithmic decision support systems
aa) Dehumanization - the 'moral buffer'
bb) Displacement of responsibility - from tool to authority figure
cc) Conclusion on moral disengagement
c) Self-imposed nonage - willingly entering into a situation of 'not being able to understand'
aa) Deskilling
bb) Non-transparent algorithms
cc) 'Supra-human' knowledge mediation
d) Conclusion on self-imposed algorithmic thoughtlessness
4. From the logic of the law to the logic of algorithms - 'machine logic'
D. Conclusion on the algorithmic turn
Chapter IV. Minimum requirements for algorithmic crime prediction
A. Development modalities
I. Transparency
II. Impartiality
B. Deployment modalities
I. Statutory basis
II. Risk communication
C. Control modalities
D. Checklist
Chapter V. Closing remarks: Tyrannicide and a new beginning
Chapter VI. Theses
Bibliography

Systemvoraussetzungen

Als PDF speichern Als Link merken