Scholars from the Berkman Klein Center for Internet & Society of Harvard University point out that (positive and negative) human rights impacts of machine learning come from at least three main sources:
1. *Quality of training data:* This is known as the “garbage in, garbage out” problem where even the best algorithms will give skewed outputs when the data that it trains on is biased.
2. *System design:* The human designers of the AI systems may incorporate their own values on the design, such as prioritising certain variables to be optimised, over others.
3. *Complex interactions:* The AI system may interact with its environment in a way that leads to unpredictable outcomes. https://coconet.social/2019/human-rights-artificial-intelligence-southeast-asia/