Manish Bhide wrote this article on fairness or bias is one of the big issues that plagues the adoption of AI in enterprises.
Even if a model is known to act in an unbiased manner during training, it does not necessarily mean that it will continue to exhibit similar behaviour once it is deployed to production. IBM Watson OpenScale helps address this problem by monitoring the behaviour of AI Models in production and checking if they are acting in a biased manner.
The article then describes:
- What is Bias?
- Bias detection in Watson OpenScale
- Data Perturbation
IBM Watson OpenScale provides a mechanism for enterprises to monitor if their models are acting in a biased manner at runtime. Interesting read!
[Read More]