Bias in Algorithms
Sarah Dean
CS 194/294 Guest Lecture, 3/1/21
About Me
 PhD student in EECS
 "Reliable Machine Learning in Feedback Systems"
 formally reasoning about safety, stability, equity, wellbeing
 Cofounded GEESE (geesegraduates.org)
Algorithms are consequential and ubiquitous
Bias in Algorithms
Bias: prejudice, usually in a way considered to be unfair
(mathematical) systematic distortion of a statistical result
Prejudice: preconceived opinion
(legal) harm or injury resulting from some action or judgment
Algorithm: a set of rules to be followed, especially by a computer
Promises of algorithmic screening
 explicit criteria
 equally applied
 auditable
Ex: Algorithmic Resume Screening
"I usually look for candidates who will be a good fit"
Ex: Algorithmic Resume Screening
Promises of algorithmic datadriven screening

explicit criteria"superhuman" accuracy  equally applied
 auditable
Ex: DataDriven Resume Screening
Handcoded algorithms are hard!
Machine learning extracts statistical patterns from employment records to automatically generate accurate and optimal classifications
Does DataDriven Screening Discriminate?
Federal laws make it illegal to discriminate on the basis of race, color, religion, sex, national origin, pregnancy, disability, age, and genetic information.
Important quantities:
Formalizing Algorithmic Decisions
features \(X\) and label \(Y\)
(e.g. resume and employment outcome)
classifier \(c(X) = \hat Y\)
(e.g. interview invitation)
 Accuracy: \(\mathbb P( \hat Y = Y)\)
 Positive rate: \(\mathbb P( \hat Y = 1)\)
 Among truly (un)qualified (conditioned on \(Y\))
 Among positive (negative) decisions (conditioned on \(\hat Y\))
Statistical Classification Criteria
Accuracy
\(\mathbb P( \hat Y = Y)\) = ________
Positive rate
\(\mathbb P( \hat Y = 1)\) = ________
False positive rate
\(\mathbb P( \hat Y = 1\mid Y = 0)\) = ________
False negative rate
\(\mathbb P( \hat Y = 0\mid Y = 1)\) = ________
Positive predictive value
\(\mathbb P( Y = 1\mid\hat Y = 1)\) = ________
Negative predictive value
\(\mathbb P( Y = 0\mid\hat Y = 0)\) = ________
\(X\)
\(Y=1\)
\(Y=0\)
\(c(X)\)
\(3/4\)
\(9/20\)
\(1/5\)
\(3/10\)
\(7/9\)
\(8/11\)
Formal Nondiscrimination Criteria
features \(X\) and label \(Y\)
(e.g. resume and employment outcome)
classifier \(c(X) = \hat Y\)
(e.g. interview invitation)
individual with protected attribute \(A\)
(e.g. race or gender)
Independence: decision does not depend on \(A\)
\(\hat Y \perp A\)
e.g. applicants are accepted at equal rates across gender
Separation: given outcome, decision does not depend on \(A\)
\(\hat Y \perp A~\mid~Y\)
e.g. qualified applicants are accepted at equal rates across gender
Sufficiency: given decision, outcome does not depend on \(A\)
\( Y \perp A~\mid~\hat Y\)
e.g. accepted applicants are qualified at equal rates across gender
Ex: Pretrial Detention
 COMPAS: a criminal risk assessment tool used in pretrial release decisions
 \(X\) survey about defendant
 \(\hat Y\) designation as high or lowrisk
 Audit of data from Broward county, FL
 \(A\) race of defendant
 \(Y\) recidivism within two years
“Black defendants who did not recidivate over a twoyear period were nearly twice as likely to be misclassified. [...] White defendants who reoffended within the next two years were mistakenly labeled low risk almost twice as often.”
Ex: Pretrial Detention
“In comparison with whites, a slightly lower percentage of blacks were ‘Labeled Higher Risk, But Didn’t ReOffend.’ [...] A slightly higher percentage of blacks were ‘Labeled Lower Risk, Yet Did ReOffend.”’
\(\mathbb P(\hat Y = 1\mid Y=0, A=\text{Black})> \mathbb P(\hat Y = 1\mid Y=0, A=\text{White}) \)
\(\mathbb P(\hat Y = 0\mid Y=1, A=\text{Black})< \mathbb P(\hat Y = 0\mid Y=1, A=\text{White}) \)
\(\mathbb P(Y = 0\mid \hat Y=1, A=\text{Black})\approx \mathbb P( Y = 0\mid \hat Y=1, A=\text{White}) \)
\(\mathbb P(Y = 1\mid \hat Y=0, A=\text{Black})\approx \mathbb P( Y = 1\mid \hat Y=0, A=\text{White}) \)
COMPAS risk predictions do not satisfy separation
Ex: Pretrial Detention
COMPAS risk predictions do satisfy sufficiency
If we use machine learning to design a classification algorithm, how do we ensure nondiscrimination?
Achieving Nondiscrimination Criteria
Attempt #1: Remove protected attribute \(A\) from features
Attempt #2: Careful algorithmic calibration
Achieving Nondiscrimination Criteria
 Preprocessing: remove correlations between \(A\) and features \(X\) in dataset. Requires knowledge of \(A\) during data cleaning

Inprocessing: modify learning algorithm to respect criteria.
Requires knowledge of \(A\) at training time 
Postprocessing: adjust thresholds in groupdependent manner.
Requires knowledge of \(A\) at decision time
Limitations of Nondiscrimination Criteria
 Tradeoffs: It is impossible to simultaneously satisfy separation and sufficiency if populations have different base rates
 Observational: since they are statistical criteria, they can measure only correlation; intuitive notions of discrimination involve causation, but these can only be measured with careful modelling
 Unclear legal grounding: antidiscrimination law looks for disparate treatment or disparate impact; while algorithmic decisions may have disparate impact, achieving criteria involves disparate treatment
 Limited view: focusing on risk prediction might miss the bigger picture of how these tools are used by larger systems to make decisisons
Preexisting Bias: exists independently, usually prior to the creation of the system, with roots in social institutions, practices, attitudes
Bias in Computer Systems
Technical Bias: arises from technical constraints and considerations; limitations of formalisms and quantification of the qualitative
Emergent Bias: arises in context of use as a result of changing societal knowledge, population, or cultural values
Classic taxonomy by Friedman & Nissenbaum (1996)
individual \(X\)
Machine Learning Pipeline
training data
\((X_i, Y_i)\)
model
\(c:\mathcal X\to \mathcal Y\)
prediction \(\hat Y\)
measurement
learning
action
Bias in Training Data
training data
\((X_i, Y_i)\)
Existing inequalities can manifest as preexisting bias
measurement
Bias in Training Data
training data
\((X_i, Y_i)\)
Technical bias may result from the process of constructing a dataset:
 Sample bias: who and where we sample
 Feature/label bias: which attributes we include
 Measurement bias: how we quantify attributes
measurement
Bias in Training Procedure
training data
\((X_i, Y_i)\)
model
\(c:\mathcal X\to \mathcal Y\)
Further technical bias results from formulating learning task and training the model: optimization bias
learning
e.g. optimizing for average accuracy will prioritize majority groups
Ex: Miscalibration in Recommendation
Designing recommendations which optimize engagement leads to overrecommending the most prevalent types (Steck, 2018)
romcom 80% of the time
horror 20% of the time
optimize for probable click
recommend romcom 100% of the time
Validity of Prediction Tasks
Small Is Beautiful: Economics as if People Mattered, E.F. Schumacher:
 Full predictability exists only in the absence of human freedom
 Relative predictability exists with regard to the behavior pattern of very large numbers of people doing 'normal' things
 Relatively full predictability exists with regard to human actions controlled by a plan which eliminates freedom
 Individual decisions by individuals are in principle unpredictable
Bias in Deployment
Emergent bias results from realworld dynamics, including those induced by the decisions themselves
individual \(X\)
model
\(c:\mathcal X\to \mathcal Y\)
prediction \(\hat Y\)
e.g. Goodhart's law: "When a measure becomes a target, it ceases to be a good measure."
action
Predictive policing models predict crime rate across locations based on previously recorded crimes
Ex: Emergent Bias in Policing
PredPol analyzed by Lum & Isaac (2016)
recorded drug arrests
police deployment
estimated actual drug use
Bias Beyond Classification
image cropping
facial recognition
information retrieval
generative models
Main References
 Barocas, Hardt, Narayanan. Fairness and machine learning: Limitations and Opportunities, Intro and Ch 1.
 Friedman & Nissenbaum. Bias in computer systems. TOIS, 1996.
 Dobbe, Dean, Gilbert, Kohli. A broader view on bias in automated decisionmaking: Reflecting on epistemology and dynamics. FAT/ML, 2018.
Bias in Algorithms
By Sarah Dean
Bias in Algorithms
 945