The document discusses gender segregation and diversity in knowledge production. It notes that while women have made progress in entering male-dominated fields, horizontal and vertical segregation persist. This segregation restricts opportunities and is inefficient. To promote equity in data science, an institute should "fix the numbers" by supporting women's education and careers, "fix the institutions" by reforming cultures and policies, and "fix the knowledge" by addressing bias in research and design. The document outlines various initiatives IVADO is taking in talent recruitment and training, counteracting biases, and hosting a summer school on bias in AI to enhance excellence through an equity lens.
3. In 2019, gender segregation is still a deeply entrenched feature of
our education system and occupations across North America
4. Women are working in nearly all occupations that once were
exclusively the domain of men, and many are in prominent
leadership roles.
Yet social norms continue to restrict occupational choices
by women and men
6. Vertical Segregation
Concentration of one gender in certain grades, levels of responsibility or
positions
The leaky pipeline: share of women in higher education and research, 2013 (%)
Ref : UNESCO, 2015
7. Gender segregation has detrimental effects on women’s and men’s
chances in the labour market and in society in general.
8. … restricts aspirations and occupational
choices by women and men
… distorts labor markets
… is a major cause for the persistent wage gap
… Hurts business innovation and productivity
Occupational gender segregation is unfair and inefficient
9. What can a research
institute in data science
do to promote equity?
Fix
the
numbers
Fix
the
institutions
Fix
the
knowledge
11. Fix the numbers
Increasing women’s participation in
Data Science by supporting girls’
education and women’s careers
Ref: Londa Schiebinger
12. Diversity should not be an end or a box to check off:
Diversity ⍯ Equity
Diversity ⍯ Inclusion
Diversity ⍯ Belonging
13. While critically important, this approach is insufficient
because it doesn’t address the structural or institutional
barriers that women face through their scientific careers.
Once women (and other under-represented groups) have chosen to study in
STEM or entered a STEM job, institutions need to develop policies and practices
to keep them there.
15. Fix the institutions
Increasing women’s participation in
Data Science by reforming
scientific and educational
institutions
Ref: Londa Schiebinger
16. A culture is more than institutions, legal regulations […] It consists in the unspoken
assumptions and values of its members.
Despite claims to objectivity and value-neutrality, academic institutions have
identifiable cultures that have developed over time—and, historically, in the absence
of women […]
Much remains to be done to restructure research and educational institutions to
remove barriers that limit women’s full participation in academic life.
Londa Schiebinger
Institutions and culture
17. This approach focuses on restructuring institutions while assuming that what goes
on inside institutions —research and knowledge production— is free of bias (and
gender neutral).
Restructuring institutions is important, but should be supplemented by
efforts to eliminate bias from research and design.
20. Potential harms from algorithmic decision-making
Ref: Kate Crawford
Harm of representation
Harm of allocation
When a system allocates or
withholds certain groups an
opportunity or resource
Allocation is immediate and readily
quantifiable
When systems reinforce the
subordination of certain groups
along the lines of identity like race,
class, gender etc
Harder to formalize and track
21. Classification is ALWAYS a reflexion of culture that divides
the world into parts.
Datasets reflects the culture but also the hierarchy of the world
that they were made in. Who is powerful is gonna appear more
frequently than who is not.
Kate Crawford, NIPS 2017
Bias as a socio-technical issue
22. Fix the knowledge
Enhancing excellence by taking
into account bias and equity into
Data Science research
Ref: Londa Schiebinger
24. What IVADO is doing to fix the numbers…
Inspiring the next generation (projet SEUR)
25. What IVADO is doing to fix the numbers…
Boosting the integration of young female graduates into the
labour market (AI for Social Good Summer Lab)
26. What IVADO is doing to fix the numbers…
Supporting African young scientists in data science
By 2030, AI will add nearly $16 trillion
to the global economy, but 70% of this
new wealth will be captured by North
America and China (IDRC)
28. What IVADO is doing to fix the institutions…
Collecting data and consulting the community
Counteracting subtle biases in hiring practices
Taking into consideration career breaks
Supporting Quebec
Interuniversity EDI Network
31. IVADO-MILA SUMMER SCHOOL: BIAS IN AI
Fernando Diaz
Principal Researcher, Microsoft FATE,
Deborah Raji
Student, UfT
Margaret Mitchell
Senior Research Scientist, Google
Petra Molnar
Human Rights Researcher, UfT
Pedro Saleiro
Post-Doc, Aequitas, University of Chicago
Moritz Hardt
Assistant Professor, UC Berkeley
Cynthia Savard Saucier,
Director UX, Shopify
32. IVADO-MILA SUMMER SCHOOL: BIAS IN AI
Registration will open on March 11th at 12h00
Ivado.ca
Training > IVADO Schools > Bias in AI
33. Our approaches are interrelated:
Increasing the participation of under-represented
groups in Data Science will not be successful until our
institutions are restructured and take bias and equity
into account into knowledge production.