Algorithmic bias

Data Science in a Box

datasciencebox.org

1 / 22

First a bit of fun...

The Hathaway Effect

2 / 22

Oct. 3, 2008: Rachel Getting Married opens, BRK.A up 0.44%
Jan. 5, 2009: Bride Wars opens, BRK.A up 2.61%
Feb. 8, 2010: Valentine’s Day opens, BRK.A up 1.01%
March 5, 2010: Alice in Wonderland opens, BRK.A up 0.74%
Nov. 24, 2010: Love and Other Drugs opens, BRK.A up 1.62%
Nov. 29, 2010: Anne announced as co-host of the Oscars, BRK.A up 0.25%

Dan Mirvish. The Hathaway Effect: How Anne Gives Warren Buffett a Rise.
The Huffington Post. 2 Mar 2011.

3 / 22

Algorithmic bias and gender

4 / 22

Google Translate

5 / 22

Amazon's experimental hiring algorithm

Used AI to give job candidates scores ranging from one to five stars -- much like shoppers rate products on Amazon
Amazon's system was not rating candidates for software developer jobs and other technical posts in a gender-neutral way; it taught itself that male candidates were preferable

Gender bias was not the only issue. Problems with the data that underpinned the models’ judgments meant that unqualified candidates were often recommended for all manner of jobs, the people said.

Jeffrey Dastin. Amazon scraps secret AI recruiting tool that showed bias against women.
Reuters. 10 Oct 2018.

6 / 22

Algorithmic bias and race

7 / 22

Facial recognition

Ian Tucker. 'A white mask worked better': why algorithms are not colour blind.
The Guardian. 28 May 2017.

8 / 22

Criminal Sentencing

There’s software used across the country to predict future criminals.
And it’s biased against blacks.

Julia Angwin, Jeff Larson, Surya Mattu, and Lauren Kirchner. Machine Bias. 23 May 2016. ProPublica.

9 / 22

A tale of two convicts

10 / 22

A tale of two convicts

10 / 22

“Although these measures were crafted with the best of intentions, I am concerned that they inadvertently undermine our efforts to ensure individualized and equal justice,” he said, adding, “they may exacerbate unwarranted and unjust disparities that are already far too common in our criminal justice system and in our society.”

Then U.S. Attorney General Eric Holder (2014)

11 / 22

ProPublica analysis

Data:

Risk scores assigned to more than 7,000 people arrested in Broward County, Florida, in 2013 and 2014 + whether they were charged with new crimes over the next two years

12 / 22

ProPublica analysis

Results:

20% of those predicted to commit violent crimes actually did
Algorithm had higher accuracy (61%) when full range of crimes taken into account (e.g. misdemeanors)
Algorithm was more likely to falsely flag black defendants as future criminals, at almost twice the rate as white defendants
White defendants were mislabeled as low risk more often than black defendants

13 / 22

How to write a racist AI without trying

Thomas Lumley. How to write a racist AI in R without really trying.
Biased and Inefficient. 27 September 2018.

14 / 22

↑, ←, Pg Up, k	Go to previous slide
↓, →, Pg Dn, Space, j	Go to next slide
Home	Go to first slide
End	Go to last slide
Number + Return	Go to specific slide
b / m / f	Toggle blackout / mirrored / fullscreen mode
c	Clone slideshow
p	Toggle presenter mode
t	Restart the presentation timer
?, h	Toggle this help

Algorithmic bias

Data Science in a Box

datasciencebox.org

First a bit of fun...

The Hathaway Effect

Algorithmic bias and gender

Google Translate

Amazon's experimental hiring algorithm

Algorithmic bias and race

Facial recognition

Criminal Sentencing

A tale of two convicts

A tale of two convicts

ProPublica analysis

Data:

ProPublica analysis

Results:

How to write a racist AI without trying

Further reading

Machine Bias

Ethics and Data Science

Weapons of Math Destruction

Algorithms of Oppression

Parting thoughts

Do good with data

Further watching

First a bit of fun...

The Hathaway Effect

Help