Why Worry about responsible Machine Learning?
Machine learning capabilities has empowered data scientists and developers to innovate, however it's always wise to take a step back, and the ask hard questions;
Are we innovating responsibly with machine learning ?
Are we deploying trustworthy machine learning solutions for our customers?
Are we enabling stakeholder confidence and regulatory compliance?
With great power comes great responsibility, we have read and heard that over and over again, Machine Learning is powerful , it’s a state-of-the-art technology that is helping the world’s most important institutions use their data to solve their most urgent problems.
We got to work for common good within our organization and with other organizations to make sure that our work is not only useful but safe.
In this articles let's ponder on worries about machine learning and how we can inculcate responsible AI principles in our daily ML development cycles and build trust throughout our the development circle.
Lets first look at the machine learning worries.
1.Increasing Inequality
2 Weaponization
3.Unintentional Bias like Showing high end jobs ads to men more than women
4.Adversarial attacks
5.Killer Drones
6.Deep Fake
7.Data poisoning especially Public data poisoning
8.Hype Definitely the biggest threats to Machine Learning that promises unrealistic expectations
How to get over the hump
One solution to mitigate some of the worries highlighted above is Microsoft AI principles. Microsoft is collaborating with research and academics around the globe in effort to advance responsible AI practices and technologies . They include;
Accountability
People should be accountable for Machine learning systems
Reliability and safety
Machine learning systems should perform reliably and safety
Fairness
Machine learning systems should treat all people fairly
Inclusiveness
Machine learning systems should empower everyone and engage people
Transparency
Machine learning systems should be understandable
Conclusion
To have the principles on your finger tips we use the acronym PARFIT, just remember that A PERFECT ML model is PARFIT .
Privacy & Security
Accountability
Reliability & Safety
Fairness
Inclusiveness
Transparency