Thread by @Geek_Manager

I love the @TheLeadDev crowd so much. They just gave @carlaprvieira applause and whooping for explaining it’s her first time in the US and her first international conference talk here at #LeadDevSanFrancisco 🥰🥰🥰💖💖💖

“Bias is like a virus that travels and is replicated by machine learning models.”

@carlaprvieira #LeadDevSanFrancisco

Potential Harms Caused by AI Systems:
1) Bias & discrimination
2) Denial of individual autonomic rights
3) Non-transparent, unexplainable or unjustifiable outcomes
4) Invasions of privacy
5) Unreliable, unsafe or poor quality outcomes

@carlaprvieira #LeadDevSanFrancisco

Bias sneaks in to machine learning in a number of places

@carlaprvieira #LeadDevSanFrancisco

Types of Bias in ML:
- historical
- representation
- measurement
- learning
- aggregation
- evaluation
- deployment

@carlaprvieira #LeadDevSanFrancisco

“Deployment bias arises when there is a mismatch between the problem a model is intended to solve and the way in which it is actually used”

@carlaprvieira #LeadDevSanFrancisco

“Algorithms cause the illusion of neutrality”

“This is called Mathwashing. When power and bias hide behind the facade of “neutral” math” — @fredbenenson

@carlaprvieira #LeadDevSanFrancisco

Possible solutions:
- fairness (statistical analysis)
- explainable and interpretable AI (explainability issues growing with advent of deep learning)
- product thinking approach (it’s gonna be a product at some point)
- ethical to proceed?

@carlaprvieira #LeadDevSanFrancisco

“Data science is a team sport that is highly interdisciplinary.

Diversity of perspective matters!

1) Technology is not free of humans
2) Math can obscure the human element & give illusion of objectivity
3) Every single human is biased”

@carlaprvieira #LeadDevSanFrancisco