SDS 513: Transformers for Natural Language Processing — with Denis Rothman

Model Agnostic

Denis Rothman emphasizes the importance of model-agnostic methods in explainable AI, highlighting tools like SHAP and LIME. These methods allow users to understand AI models by focusing on inputs and outputs rather than the complex inner workings of models like GPT-3. Rothman uses a cake analogy to explain this concept, comparing it to identifying a missing ingredient by adjusting inputs and observing outputs.

The best explainable AI is model agnostic, unless you're a developer and you want to see what's going on inside.

--- Denis Rothman

This approach simplifies the process of understanding AI behavior without needing to dissect the intricate layers of the model 1 2.

   

Practical Ethics

In practical applications, Rothman discusses the challenges of implementing explainable AI while maintaining ethical standards. He shares an example using US census data, where removing biased fields like race and gender improved model outcomes. Rothman argues that focusing on relevant data, such as age and education, can lead to more ethical and effective AI models.

I took every field out of there, and I just left two fields in there. Age, years of education.

--- Denis Rothman

This approach not only enhances model performance but also aligns with ethical guidelines by eliminating unnecessary biases 3 4.