Hands-On Explainable AI (XAI) with Python by Denis Rothman

Author:Denis Rothman
Language: eng
Format: epub
Tags: COM004000 - COMPUTERS / Intelligence (AI) & Semantics, COM037000 - COMPUTERS / Machine Theory, COM044000 - COMPUTERS / Neural Networks
Publisher: Packt
Published: 2020-07-29T11:38:43+00:00

We recommend that you spend as much time as necessary drawing exciting what-if conclusions that will boost your XAI expertise.

In this section, we loaded the dataset and transformed and defined the new features. We then ran a linear estimator and a DNN. We displayed the data points in WIT's visualization module to investigate factual and counterfactual data points.

Summary

In this chapter, we first explored the moral and ethical issues involved in an AI project. We asked ourselves whether using certain features could offend the future users of our program.

We examined the legal issues of some of the information in a dataset from a legal perspective. We discovered that the legal use of information depends on who processes it and how it is collected. The United States government, for example, is entitled to use data on certain features of a person for a U.S. census survey.

We found that even if a government can use specific information, this doesn't mean that a private corporation can use it. This is illegal for many other applications. Both U.S. and European legislators have enacted strict privacy laws and make sure to apply them.

The ML perspective showed that the key features, such as age and the level of education, provide interesting results using a KMC algorithm. We built a KMC algorithm and trained on age and level of education. We saved the model and ran it to produce our labels. These labels provided a logical explanation. If you have a high level of education, such as 14+ years of overall education, for example, you are more likely to have a higher income than somebody with less education.

The KMC algorithm detected that higher education in age groups with some years of experience provides a higher income. However, the same customized XAI approach showed that additional features were required to predict the income level of a given person.

We suppressed the controversial features of the original dataset and inserted random cultural features.

We trained the transformed data with a linear classifier and a DNN, leaving the initial labels as they were.

Finally, we used WIT's visualization modules to display the data. We saw how to use WIT to find explanations for factual and counterfactual data points.

In the next chapter, AI Fairness with Google's What-If Tool (WIT), we will build an ethical XAI investigation program.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.