Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Democra(zing Data Science
Sophie Chou @mpe(tchou
William Li @williampli
Ramesh Sridharan @tweetsbyramesh
{soph,wpli,rameshvs}@mit.edu
Some Links • Paper: bit.ly/DDSpaper
• Cathy O’Neil’s Blog (@mathbabedotorg): bit.ly/DDSblog
• TwiMer: @mpe3tchou, @williampli, @tweetsbyramesh
2
[insert technology] for Social Good
Technology is a force mul(plier, for beMer or worse
3
What is Data Science? • Our working defini(on: transforming data into insights/solu(ons/products 1. collec(on & storage 2. cleaning & structuring 3. analyzing & finding paMerns 4. visualizing & communica(ng results
4
What is “Democra(zing Data Science”?
The applica(on of data science is undemocra(c: problems that promote the common good
receive insufficient aMen(on.
5
Ford/MacArthur Founda(on, 2013 “Technology talent is a key need in
government and civil society, but the current state of the pipeline is
inadequate to meet that need.” Source: hMp://bit.ly/FordMacArthurReport
6
Why?
misalloca3on of data science
lack of diversity in data science
incen(ves for data scien(sts
data science educa(on
7
Outline • Incen(ves for data scien(sts
• Democra(zing data science educa(on
• Poten(al solu(ons
8
Sources of Power in Data Science
capital data
people 9
Human exper(se “...[salaries] between $200,000 and $300,000 a year…100 recruiter emails a day” “...working for a consumer Internet firm can be surprisingly rewarding.”
Source: hMp://bit.ly/WSJDataScience 10
Sources of Power in Data Science
capital data
people 11
Outline • Structural inequali(es in data science
• Democra(zing data science educa(on
• Poten(al solu(ons
12
Nega(ve feedback loop Lack of
diversity in data scien(sts
Lack of diversity in educa(on
Lack of diversity in applica(on
13
Why care? • Diversity is key to innova(on (Forbes Insights, 2011)
• Lack of diversity perpetuates misalloca(on
14
Racial inequality in science and engineering
• Misrepresenta(on isn’t going away
• Fewer minori(es receive degrees
15 Source: NSF Science and Engineering Indicators 2014
Women in Compu(ng • 1985: 37% 2000: 29% 2014: 18%
• 2000 to 2014: – 5% decrease in math – 2% decrease in engineering
16 Source: NSF Science and Engineering Indicators 2014
Graduate degrees • Women fare even worse
• 2x as many white males receiving degrees as all minori:es combined
• Only 1 in 5 PhDs female
Source: NSF Science and Engineering Indicators 2014 17
Unlocking the Clubhouse: A Case Study
• 2014 incoming class: 40% women
Source: Unlocking the Clubhouse, 2002 18
Triggering Posi(ve Change
“insuring science and technology are considered in their social context may be the most important
change that can be made in science teaching for all people, both male and female.”
Source: Taggart & O’Gara, 2000 19
Posi(ve feedback loop Relevant
applica(ons
Increase educa(onal opportuni(es
AMract diverse demographics
20
Outline • Structural inequali(es in data science
• Democra(zing data science educa(on
• Poten(al solu(ons
21
Technologies • Recent technologies target broader audiences
• Ooen require significant technical literacy
• Can we broaden access further s(ll?
22
Crowd-‐based efforts • Machine learning compe((ons – Can promote meaningful problems – Low barrier to entry
• Ooen run by for-‐profit en((es • Can we encourage more ini(a(ves like KDD Cup 2014?
23
Educa(on: MOOCs • Tremendous poten(al for reaching students
• Most ooen taken by professionals and people with graduate degrees
24
Private Sector Opportuni(es • Reaching out to underserved groups – Tap new markets
• Pro bono work could service groups and bring in new customers
• Meaningful “small data” to serve the long tail 25
Academic Research • Research promo(ng social good is par(cularly accessible to academics – Social welfare problems ooen rely on public data – Academia is well-‐suited to interdisciplinary research
• Need for focus on meaningful problems
26
Your Solu(on Here! • We believe the community has a responsibility to solve these problems
• Exper(se in policy, business, sta(s(cs, healthcare, computer science will all be crucial
• Undemocra(c inequali(es persist in data science applica(ons
• All of us can be part of the solu(on @mpe(tchou @williampli @tweetsbyramesh 27