1. XAI techniques
Well done on the exercises so far! We will now explore ways in which XAI aims to explain an AI system in a transparent and understandable way.
2. XAI explanations
There are different ways in which XAI aims to explain the decision making of an AI system. We can categorize the approaches as follows. Model-specific explanations, model-agnostic explanations, local explanations, and global explanations. We will go over these throughout the course as well in more detail.
3. Model-specific explanations
An example of a model-specific explanation is a decision tree. We can use AI to automatically generate a decision tree from a dataset. This decision tree can be visualized to understand how the data is categorized.
This is how a visualization of decision tree is a model-specific explanation.
4. Model-agnostic explanations
Model-agnostic explanations are like a swiss army knife. They're flexible and adaptable, capable of working with any AI model, whether it's a simple linear model or a complex neural network. These explanations don't need to know the inner workings of the model. They can provide insights just by looking at the input and output, much like how a swiss army knife can help you without needing to be custom-made for each task.
5. Local explanations
Imagine using a zoom lens on a camera to focus closely on a single detail in a vast landscape. Local explanations in XAI do something similar. They zoom in on a single decision made by an AI model, like why it recommended a particular movie or flagged a transaction as fraudulent. This close-up view helps us understand the reasoning behind specific, individual outcomes, providing clarity and transparency for those particular cases.
6. Global explanations
Contrast that zoom lens with a panoramic photo that captures the entire landscape in one wide shot. Global explanations offer a comprehensive overview of an AI model's overall behavior, similar to that panoramic view. They help us see the big picture, showing us how the model tends to make decisions across a wide range of scenarios. This broad perspective is crucial for understanding the general principles guiding the model's behavior, ensuring it aligns with our values and expectations.
7. Applying the right explanation
Choosing the right type of explanation depends on your specific needs and context.
Use model-agnostic explanations when dealing with multiple models or when simplicity and flexibility are paramount. Opt for model-specific explanations when depth and detail are required to understand or refine a particular model.
Apply local explanations when you need to dissect and justify individual decisions, especially in critical applications.
Employ global explanations to gain and provide a comprehensive understanding of the model's behavior and to ensure alignment with broader goals and standards.
8. Let's practice!
As we move forward, we'll explore these concepts in more depth, applying them to real-world AI scenarios to see how they explain the inner workings of AI, making it more transparent, understandable, and trustworthy.