Categories
Artificial Intelligence

Artificial Intelligence problem: Interpretability

++ All booklets now selling at 70% discount and fully available for free through the KDP unlimited circuit. Get them while it lasts!

Tailored content can be provided upon request. Submit your technical specification through the contact form in the About page at https://www.tenproblems.com/about/ and get your quote.

Literature Review: Interpretability problem for Artificial Intelligence

This “Ten Problems for Artificial Intelligence in the 2020s” booklet identifies ten relevant areas from very recent contributions put forward at academic level in the form journal articles, conference proceedings and students theses. Ten freely accessible internet references have been selected for each area and direct links are provided at the end of each chapter for own consultation. Our selected references do not intend to mirror ranking indexes nor establish novel classifications. On the contrary, they are meant to represent peer-reviewed, diverse and scientifically-sound case studies for vertical dissemination aimed at non-specialist readers. They will also be able to scoop even more references through the bibliography that is reported at the end of each selected reference.

Without further ado, these are the ten problems that we are going to introduce in this booklet:

  1. computer games, 
  2. intelligent optimization,
  3. accountability,
  4. human intervention,
  5. ethical issues,
  6. bias,
  7. interpretability,
  8. software development,
  9. open source,
  10. operations research.

Each problem has its own dedicated chapter made of an introductory section, a short presentation of the ten selected references and a conclusions section.

The final chapter of this booklet will report the conclusions from each chapter again in order to provide a complete executive summary.


7 Interpretability

THE PROBLEM — Definitions of interpretability vary, but at fundamental level it is the ability to explain or present in understandable terms to humans. Users want and frequently need to understand how decisions impacting them are made. The way forward is to design models that are inherently interpretable and theoretical, legal and practical frameworks are needed. Problems, in fact, arise from non-alignment of practices and targets.

CASE STUDIES — … buy this booklet from Amazon …

CONCLUSIONS — Practitioners can design tasks-specific interpretation methods with their expertise and insights. Initial steps have been taken to use explainable AI beyond validation to arrive at better and more predictive models. White-box highly performing models are very hard to create, especially in computer vision and natural language processing. Review papers have universally failed even to truly distinguish between the basic concepts of explaining a black box and designing an interpretable model. Given a model and an input instance, a novel model-agnostic approach uses a custom genetic algorithm to generate counterfactuals: instances close to the input that change the prediction of the model. More in general, in order to make Machine Learning trustworthy, computer scientists need to consider measures beyond average predictive accuracy. The question is whether we can have a self-adaptive machine learning system that is able to interpret and explain its data model in order for it to be controlled. A synthesis between the once popular k-nearest neighbors approach and information theory may provide a clear path towards models that are innately interpretable and auditable. Machine Learning methods based on Evolutionary Fuzzy Systems also preserve comprehensibility also boosting their modeling abilities. Simplicity, stability, and predictivity are minimum requirements for interpretable models.

TEN FREE REFERENCES FROM THE INTERNET — … buy this booklet from Amazon …


ArtifIntel
“Ten Problems for Artificial Intelligence in the 2020s” booklet for Amazon Kindle, 2020; click on the cover to go to the dedicated Amazon listing page

booklet updated on 19 Jun 2021, now on sale as version 1.2

By TenProblems

Literature Reviews for Inquisitive Minds