RDS Paper Presentations
David Schouten
RDS Paper Presentations 1
P01 Ethical dilemmas and moral obligations for visualizations 2
P02 Man is to Computer Programmer as Woman is to Homemaker? 2
P03 Critical Questions for Big Data 3
P05 ‘‘We are all Different’’: Statistical Discriminationand the Right to be Treated as
an Individual 4
P06 Ethical Implications of Embodied Artificial Intelligence in Psychiatry,Psychology,
and Psychotherapy 4
P07 On the Dangers of Stochastic Parrots: Can language models be too big? 5
P08 The curse of knowledge in visual data communication 6
P09 Is it possible to grant legal personality to artificial intelligence software
systems? 6
P10 The State of the Art in Enhancing Trust in Machine Learning Models with the
Use of Visualizations 7
P11 FA*IR: A Fair Top-k Ranking Algorithm 8
P12 The ethics of algorithms: Mapping the debate 9
P13 Ten simple rules for responsible big data research 9
P14 Surfacing Visualization Mirages 11
P15 The ethics of big data as a public good: which public? Whose good? 12
P17 Why we should have seen that coming: comments on Microsoft’s Taw
experiment, and wider implications 12
1
, P01 Ethical dilemmas and moral obligations for visualizations
Correll, M. (2019, May). Ethical dimensions of visualization research. In Proceedings
of the 2019 CHI Conference on Human Factors in Computing Systems (pp. 1-13).
Visualizations are often not neutral. Their sources are biased, data is absent, data is
misrepresented or based on wrong perceptions. Some concerning trends and design
dilemmas is the increased use of automated analysis, the increase use of machine
learning and provenance (When we visualize the end result of a visualization design,
but not the process by which it was created, we risk propagating false, misleading, or
unreproducible findings).
As a visualization researcher, you may be the only contact someone has with the
presented data. Therefore, you should adhere to three ethical obligations:
1. Visibility
a. Visualize hidden labour
b. Visualize hidden uncertainty
c. Visualize hidden impacts
2. Privacy
a. Encourage “small data” (you don’t always need a lot of data (“big data”)
for your research)
b. Anthropomorphize (humanize) data
c. Obfuscate data to protect privacy
3. Power
a. Support data “due process” (only use data for the purposes it was
intended for)
b. Act as data advocates
c. Pressure or slow unethical analytical behavior
This paper is the synthesis of existing critical and ethical views of data and
visualization, and a call to action to be mindful o f the ethical implications of
visualisation researcher’s work.
P02 Man is to Computer Programmer as Woman is to Homemaker?
Bolukbasi, T., Chang, K. W., Zou, J., Saligrama, V., & Kalai, A. (2016). Man is to
computer programmer as woman is to homemaker? debiasing word embeddings.
arXiv preprint arXiv:1607.06520.
Word embeddings are a way to represent text data as vectors which are used by
many machine learning and natural language processing applications. It turns out
that word embeddings can be very biased. The Word2vec dataset used in the paper
(based on Google News) was also used in the RDS workshops.
When asked “Japan is to Tokyo what France is to X”, the model obviously
responds X = Paris. However, when asked “Man is to Computer Programmer what
2
David Schouten
RDS Paper Presentations 1
P01 Ethical dilemmas and moral obligations for visualizations 2
P02 Man is to Computer Programmer as Woman is to Homemaker? 2
P03 Critical Questions for Big Data 3
P05 ‘‘We are all Different’’: Statistical Discriminationand the Right to be Treated as
an Individual 4
P06 Ethical Implications of Embodied Artificial Intelligence in Psychiatry,Psychology,
and Psychotherapy 4
P07 On the Dangers of Stochastic Parrots: Can language models be too big? 5
P08 The curse of knowledge in visual data communication 6
P09 Is it possible to grant legal personality to artificial intelligence software
systems? 6
P10 The State of the Art in Enhancing Trust in Machine Learning Models with the
Use of Visualizations 7
P11 FA*IR: A Fair Top-k Ranking Algorithm 8
P12 The ethics of algorithms: Mapping the debate 9
P13 Ten simple rules for responsible big data research 9
P14 Surfacing Visualization Mirages 11
P15 The ethics of big data as a public good: which public? Whose good? 12
P17 Why we should have seen that coming: comments on Microsoft’s Taw
experiment, and wider implications 12
1
, P01 Ethical dilemmas and moral obligations for visualizations
Correll, M. (2019, May). Ethical dimensions of visualization research. In Proceedings
of the 2019 CHI Conference on Human Factors in Computing Systems (pp. 1-13).
Visualizations are often not neutral. Their sources are biased, data is absent, data is
misrepresented or based on wrong perceptions. Some concerning trends and design
dilemmas is the increased use of automated analysis, the increase use of machine
learning and provenance (When we visualize the end result of a visualization design,
but not the process by which it was created, we risk propagating false, misleading, or
unreproducible findings).
As a visualization researcher, you may be the only contact someone has with the
presented data. Therefore, you should adhere to three ethical obligations:
1. Visibility
a. Visualize hidden labour
b. Visualize hidden uncertainty
c. Visualize hidden impacts
2. Privacy
a. Encourage “small data” (you don’t always need a lot of data (“big data”)
for your research)
b. Anthropomorphize (humanize) data
c. Obfuscate data to protect privacy
3. Power
a. Support data “due process” (only use data for the purposes it was
intended for)
b. Act as data advocates
c. Pressure or slow unethical analytical behavior
This paper is the synthesis of existing critical and ethical views of data and
visualization, and a call to action to be mindful o f the ethical implications of
visualisation researcher’s work.
P02 Man is to Computer Programmer as Woman is to Homemaker?
Bolukbasi, T., Chang, K. W., Zou, J., Saligrama, V., & Kalai, A. (2016). Man is to
computer programmer as woman is to homemaker? debiasing word embeddings.
arXiv preprint arXiv:1607.06520.
Word embeddings are a way to represent text data as vectors which are used by
many machine learning and natural language processing applications. It turns out
that word embeddings can be very biased. The Word2vec dataset used in the paper
(based on Google News) was also used in the RDS workshops.
When asked “Japan is to Tokyo what France is to X”, the model obviously
responds X = Paris. However, when asked “Man is to Computer Programmer what
2
