An AI with an Agenda - How Our Cognitive Biases Leak Into Machine Learning

This talk examines how human assumptions and cognitive biases can enter machine-learning systems. It discusses using probability, representative data, robust optimization, and ethical checks to reduce harmful effects.

Syllabus

Intro
REAL CONSEQUENCES
KNOW WHAT QUESTION YOU'RE ASKING UP FRONT
USE CONDITIONAL PROBABILITY OVER CORRELATION
MORTGAGE LENDING ANALYSIS
BUILD A BETTER DATA SET!
BEWARE SHADOW COLUMNS
MAKE SURE YOUR SAMPLE SET IS REPRESENTATIVE
KEEP IN MIND YOU NEED TO KNOW WHO CAN BE AFFECTED IN ORDER TO UN-BIAS
PRICING ALGORITHMS
WHAT IF AMAZON BUILT A SALARY TOOL INSTEAD?
THE BRATWURST PROBLEM
MORE COMPLEX ALGORITHMS THAT INCLUDE OUTSIDE INFLUENCE
VERIFY AND CHECK SOLUTIONS DERIVED FROM SIMULATION
MANY AI/ML TOOLS ARE TRAINED TO MINIMIZE AVERAGE LOSS
REPRESENTATION DISPARITY
DISTRIBUTIONALLY ROBUST OPTIMIZATION
WHAT HAPPENS TO PEOPLE WHO USE DIALECT?
PREDICTIVE POLICING
LOOK TO CONTROL ENGINEERING
ABIDE BY ETHICS GUIDELINES
Transparency of Use Transparency of Algorithms