Responsible AI Flashcards
Ai in medicine and health care
-Diagnosis
-Prognosis
-Choice of therapy
-Therapeutic decisions
-Assisting during health care or operations
-Development of new drugs or drug repurposing
-Fraud detection
Digital tools for skin cancer detection
Been developed since 2000s
- Apps claim to be better than dermatologists
-Aims to enable people to check their skin themselves relieving the need for dermatology
This specific app was found to less accurate for dark skinned individuals
AI and Bias
-Bias can be unintentional or intentional
-Historical bias: a system learning from historical data
-Technology acts supposedly neutral
-Intransparency of the outcomes due to AI being a black box
Reproduction of inequality through AI
Technology trained with biased datasets can reproduce existing discrimination
Biases via false stereotyping in labels (confirmation bias or social bias)
Lack of completeness of data (sampling bias or representation bias)
Implementation of bias due to context of use
Implementation bias
using wrong data for an outcome, e.g. Obermeyer case where less money spent on healthcare for black people was used as proxy for health instead of care needs. (Goal prioritize patients for additional care)-> less documentation of healthcare needs might wrongly be assumed to be more healthy
How developers influence technology
Bias from pattern programming:
-race/ethnicity/class
-capability of user
-Designed out of the developers cultural worldview
-Developers focus on product, not context of use
-No/little involvement of end-users in design
-Behavioralist assumptions about future use
-No attention for unanticipated consequences and position in broader political economy
Why technology is not neutral
-Society shapes humans
-Humans shape technology
-Technology changes society
Technology is culture
-developed by people in certain place, time, and society
-Situated in larger political and societal structures
-Used by different people in their own lifeworlds
-Has unintended consequences
-Changes the way we live together and think
-Reproduces societal inequalities
AI from an ethical perspective: values
-[Justice
-Environmental sustainability
-Accountability
-Transparency
-Privacy
-Reliability
(AI ethics impact group 2020)