Amir H. Payberah

�payberah@kth.se

� Oct. 10, 2025

Data Colonization

Civilization

2

https://tinyurl.com/bdhbexwf

Colonialism and Land Grabbing

• Explore�
• Expand�
• Exploit�
• Exterminate

3

https://tinyurl.com/mry959j7

Data Grab - Ulises A. Mejias and Nick Couldry

4

https://tinyurl.com/yc62yz9h

Data Colonialism

• Data is often harvested without consent, reinforcing existing inequalities and creating new forms of exploitation.

5

https://tinyurl.com/39w2zfkf

Impact of Data Colonialism

• Silences marginalized voices�
• Erases narratives�
• Perpetuates oppression

6

https://tinyurl.com/5xe26xx3

Challenge These Practices

• Embrace pluralism in data science�
• Value and include the marginalized communities viewpoints at all stages of the process.�
• Recognizing a multiplicity of voices for a complete picture.

7

Anti-Eviction Mapping Project

8

Anti-Eviction Mapping Project (AEMP)

• AEMP: an example of how data can be used to resist the 4X model and serve the interests of marginalized communities.�

​

9

Housing Crisis and Evictions in San Francisco

• High income gap between the richest and poorest in San Francisco.�
• Wealth gap has racial and gender dimensions.�
• Escalating rate of evictions since 2003.

10

https://tinyurl.com/ycx2fjz2

AEMP Community

• A collective of housing justice activists, researchers, artists, and historians (founded in 2013).�
• With many designers and nonexperts (local residents) to map evictions.

11

Feminist Counterdata and Countervisualization

• AEMP documents the displacement of residents through gentrification and eviction, particularly in communities of color and low-income neighborhoods.�
• A set of counterdata collection and countervisualization strategies.
• Where people go after they are evicted?
• How many of those people end up homeless?
• Which landlords are responsible for evicting?�

12

https://tinyurl.com/ywnzbnxx

https://antievictionmap.com

• Maps of displaced residents
• Maps of evictions
• Maps of tech buses
• Maps of property owners
• Maps of the Filipino diaspora
• Maps of the declining numbers �of Black residents in the city
• And more …

13

https://antievictionmap.com

Embracing Pluralism in Data Science

• AEMP exemplifies the embrace pluralism principle of data feminism.�
• Valuing various perspectives at all stages of the process, from data collection to cleaning to analysis to communication. �

​

14

https://tinyurl.com/2kks39b6

Questions for Discussion

• Can pluralism and profit coexist in the field of data science, or are they fundamentally at odds?�
• How effective is collaborative data science in truly shifting power to marginalized communities?�
• What are the potential risks of using these methods in ways that still benefit those in power?

15

Embracing Pluralism in Data Science

• Also means recognizing how data science methods can unintentionally silence voices for clarity, cleanliness, and control.

16

https://tinyurl.com/39yzjt99

Data Cleaning

17

Cleaning and Preparing Data

• Data cleaning is considered a crucial part of data science.�
• What might be lost in this process?�
• Whose perspectives might be lost in this process?�
• Whose perspectives might be additionally imposed?

18

Historical Roots and Ethical Considerations of Cleanliness

• Ideas of cleanliness in data have historical ties to eugenics.
• Improve the genetic quality of human populations through selective breeding.�
• Early statisticians like Pearson and Galton were also leaders in the eugenics movement.�

​

19

https://tinyurl.com/ydv28yh7

https://tinyurl.com/4b9j47yd

Pearson Correlation Coefficient

Ghost Stories for Darwin - Banu Subramaniam

• Argues that scientific practices are deeply embedded in cultural and political contexts.�
• Critiques how the desire for purity and uniformity in scientific data has often mirrored broader societal efforts to control and categorize human populations.

20

https://tinyurl.com/ynky768r

The TESCREAL Bundle

• Transhumanism�
• Extropianism�
• Singularitarianism�
• Cosmism�
• Rationalism�
• Effective Altruism�
• Llongtermism

21

The Cleaning Paradigm and Its Critique

• Katie Rawson and Trevor Muñoz, in Against Cleaning argued that cleaning assumes an underlying correct order.�
• Rich information can be lost during cleaning.�
• Diversity-hiding trick

22

Epistemic Violence - Gayatri Chakravorty Spivak

​

​

• The harm caused silencing or invalidating�marginalized knowledge and voices.

​

23

Situated Knowledges - Donna Haraway

• All knowledge is partial, no single person or group can claim an objective view of the Truth.�
• People make knowledge from a particular standpoint: from a situated, embodied location in the world.

24

https://tinyurl.com/y6zzjhj5

All Data Are Local - Yanni Loukissas

• Data is inherently local: it reflects the specific conditions, practices, and power structures of the settings in which it is created.
• Considering data settings rather than datasets.
• E.g., Clemson University Library's "upstate" term is clear locally, but confusing to outsiders.

25

https://lmc.gatech.edu/featured-people/yanni-loukissas

• How do we start gaining a deeper understanding in data science?

26

Transparency and Reflexivity

• Transparency: revealing technical details.�
• Reflexivity: who is doing the work and the process behind it.

27

https://tinyurl.com/5n99sjuj

Beyond Transparency and Reflexivity

• Actively and deliberately inviting other perspectives into the data analysis and storytelling process.

28

https://tinyurl.com/mrxy76mz

From Data for Good (Data Ethics) to Data for Co-liberation (Data Justice)

29

Questions for Discussion

​

• Is it possible to achieve both technical accuracy and ethical responsibility in data cleaning, or are these goals sometimes in conflict?�
• How should data scientists approach situations where prioritizing one might compromise the other?

30

Situating Data on the Wild Web

31

32

https://tinyurl.com/3p9r2fnr

Context and Metadata in Data

• Many datasets lack context or metadata.�
• Lack of context makes data exploration difficult.�
• Without local knowledge, understanding�power dynamics is challenging.

33

Context-Free Data Analysis

• Chris Anderson's "The End of Theory" claims data speak for themselves, eliminating the need for context.�
• Statistical inference relies on sampling, but big data suggests using all data directly.

34

Is Correlation Enough?

• Anderson insists: correlation is enough.�
• E.g., Google search�
• But what happens when the number of�links is also highly correlated with sexist,�racist, and pornographic results?�
• Correlation can reinforce societal biases.

35

"Raw Data" Is an Oxymoron - Lisa Gitelman

• “The numbers speak for themselves” is the premise that data are a raw input.�
• Data is anything but raw.

36

https://tinyurl.com/2f7v5325

Raw Data, Cooked Data, Cooking

• Data are not raw inputs; they are cooked through social, political, and historical contexts.�
• One strategy for considering context is to consider the cooking process that produces “raw” data.�
• Exploring and analyzing what is missing �from a dataset.

37

https://tinyurl.com/yvhs3v6w

Word Embeddings Quantify 100 Years of Gender and Ethnic Stereotypes

​

• Word embeddings�
• Explore gender and ethnic stereotypes
• Across Google Books and the New York Times�
• Words like intelligent, logical, and thoughtful were predominantly associated with men until the 1960s.�
• From the 1960s onwards, these words increasingly became associated with women.�
• Attributed to the women's movement in the 1960s and 1970s.

38

Formerly Gang-Involved Youth as Domain Experts for Analyzing Unstructured Twitter Data

• SAFELab: use Twitter data to understand and prevent gang violence.�
• Deep listening and contextual analysis of social media.�

​

39

Consider Context

40

Consider Context

​

• Where did it come from?�
• Who collected it? �
• When? �
• How was it collected? �
• Why was it collected?

41

Consider Context in Data

• Datasheets for datasets�
• Dataset nutrition label�
• A Participatory approach for �data work

42

https://tinyurl.com/3a4av458

Datasheets for Datasets

• Address the needs of two stakeholder groups: dataset creators and consumers�
• Objectives for dataset creators
• Encourage careful reflection on the creation, distribution, �and maintenance of a dataset, considering assumptions,�risks, and potential impacts.�
• Objectives for dataset consumers
• Ensure they have the information they need to make �informed decisions about using a dataset.

43

https://www.dair-institute.org/projects/documentation-accountability/

Datasheets for Datasets - Key Components

• Motivation
• Why was the dataset created? What is its intended use?�
• Composition
• What does the dataset consist of? Who are the subjects?�
• Collection process
• How was the data collected? Any ethical concerns?�
• Preprocessing
• Any cleaning or transformations applied to the data?�
• Distribution
• How can the dataset be accessed or shared?�
• Maintenance
• Who is responsible for maintaining the dataset?

44

Datasheets for Datasets - Example (Health Records)

• Motivation
• Collected to improve healthcare outcomes�
• Composition
• Contains anonymized patient records from 10 hospitals�
• Collection process
• Data collected through surveys and hospital records�
• Preprocessing
• Data anonymized and cleaned for missing entries�
• Distribution
• Available to researchers under a strict data-sharing agreement�
• Maintenance
• Updated annually by the hospital consortium

45

Dataset Nutrition Label

• A structured approach to provide key information about datasets.�
• Inspired by food nutrition labels: offers transparency on data content and quality.

46

https://tinyurl.com/2dx7xe7x

Dataset Nutrition Label - Key Components

• Data provenance
• Information about the origin and source of the data.�
• Data collection methods
• How the data was gathered, including any ethical �concerns.�
• Limitations and biases
• Acknowledging data biases, gaps, or potential issues.�
• Data use and misuse
• Guidelines for appropriate use and potential risks of misuse.�
• Maintenance and updates
• Who manages the data and how often it’s updated.

47

Dataset Nutrition Label - Example (Health Records)

​

• Data provenance
• Collected from multiple healthcare providers.�
• Data collection methods
• Patient data from electronic health records.�
• Limitations and biases
• Underrepresentation of rural populations and minority groups.�
• Data use and misuse
• For predictive healthcare models, but should avoid use in determining insurance premiums due to potential biases.�
• Maintenance and updates
• Maintained by a health consortium, updated annually.

48

Participatory Documentation

• Involves multiple stakeholders (data creators, curators, users) in the documentation process.�
• Encourages diverse perspectives and shared ownership over data production practices.

49

https://tinyurl.com/4ndhe9hm

Participatory Documentation - Key Steps

• Involve stakeholders early
• Engage data collectors, curators, and users from the beginning of the project.�
• Map data creation
• Document the entire data production process, from collection to final output, noting key decisions.�
• Record rationales
• Explain why certain data collection, cleaning, or curation choices were made.�
• Highlight biases and gaps
• Identify and openly document any biases or limitations in the data.

50

Questions for Discussion

​

• Who should provide context?
• End users?
• Data publishers?
• Data intermediaries (e.g., librarians, journalists, educators)?

51

Summary

52

53

• Data Feminism, (ch. 4-5)
• Data colonialism: Rethinking big data’s relation to the contemporary subject, Nick Couldry et al., 2019
• Against Cleaning, Katie Rawson and Trevor Muñoz [link]
• All Data Are Local: Thinking Critically in a Data-Driven Society, Yanni Loukissas (ch. introduction)
• Situated knowledges: The science question in feminism and the privilege of partial perspective, Donna Haraway, Feminist Studie, 1988
• Social media for large studies of behavior, Derek Ruths and Jürgen Pfeffer, 2014
• Artificial Intelligence and Inclusion: Formerly Gang-Involved Youth as Domain Experts for Analyzing Unstructured Twitter Data, William R. Frey et al., 2020
• Datasheets for datasets, Timnit Gebru et al., Communications of the ACM, 2021
• The Dataset Nutrition Label (2nd Gen), Kasia S. Chmielinski et al., 2018
• Documenting Data Production Processes: A Participatory Approach for Data Work, Milagros Miceli et al., 2022

54