3 min
Perpetuating Harm - How AI Can Reinforce and Spread Bias 52 Weeks of Cloud
-
- Technology
Description:
We explore an alarming potential outcome with AI - taking biased data or engagement incentives around sensitive issues as input can lead to exponentially amplifying and propagating harm across populations. Without oversight, a focus strictly on profit could promote extremism, discrimination, and violence over time even if unintended initially.
Key problems span:
Creating bias from tainted datasets
Optimization driving selective exposure
Enriching engagement at the cost of marginalized groups
Social costs far exceeding private short term gains
Understanding these dynamics is key to policy reforms including algorithmic accountability and ethical AI training. Companies have an obligation to assess the true impact of their systems.
Description:
We explore an alarming potential outcome with AI - taking biased data or engagement incentives around sensitive issues as input can lead to exponentially amplifying and propagating harm across populations. Without oversight, a focus strictly on profit could promote extremism, discrimination, and violence over time even if unintended initially.
Key problems span:
Creating bias from tainted datasets
Optimization driving selective exposure
Enriching engagement at the cost of marginalized groups
Social costs far exceeding private short term gains
Understanding these dynamics is key to policy reforms including algorithmic accountability and ethical AI training. Companies have an obligation to assess the true impact of their systems.
3 min