AI2030 Responsible AI Tools Marketplace

Key Takeaways

Different stakeholders need different types of explanations: The paper identifies four distinct types of explanations—Diagnostic (how the model works internally), Explication (what the output means), Expectation (how the model generalizes), and Role (why we should use this model)—each serving different purposes for different audiences, meaning a single explanation method cannot satisfy all needs.
"Explanation" means different things in AI contexts: The paper addresses a critical ambiguity in XAI terminology, where practitioners and stakeholders often use the word "explanation" to mean fundamentally different things, leading to miscommunication about what XAI methods actually deliver and whether they're fit for purpose.
Explanations should identify intervention points: The paper frames explanations as tools for identifying where we can intervene in AI systems to create desired changes, suggesting that a good explanation is one that reveals actionable leverage points rather than simply providing information.
Role-explanations address social context: Beyond technical explanations of how models work, the paper emphasizes that stakeholders need "Role-explanations" that justify why a model should be used in a particular social context—a crucial consideration for responsible AI deployment that goes beyond technical transparency.
A pluralistic framework prevents misuse of XAI methods: By providing a taxonomy that matches explanation types to appropriate XAI methods, the paper offers practitioners and stakeholders a template to avoid using the wrong explanation tool for their actual need, reducing the risk of false confidence in AI systems.

Abstract

The increasingly widespread application of AI models motivates increased demand for explanations from a variety of stakeholders. However, this demand is ambiguous because there are many types of 'explanation' with different evaluative criteria. In the spirit of pluralism, I chart a taxonomy of types of explanation and the associated XAI methods that can address them. When we look to expose the inner mechanisms of AI models, we develop Diagnostic-explanations. When we seek to render model output understandable, we produce Explication-explanations. When we wish to form stable generalizations of our models, we produce Expectation-explanations. Finally, when we want to justify the usage of a model, we produce Role-explanations that situate models within their social context. The motivation for such a pluralistic view stems from a consideration of causes as manipulable relationships and the different types of explanations as identifying the relevant points in AI systems we can intervene upon to affect our desired changes. This paper reduces the ambiguity in use of the word 'explanation' in the field of XAI, allowing practitioners and stakeholders a useful template for avoiding equivocation and evaluating XAI methods and putative explanations.

To be published in CD-MAKE 2021 conference proceedings

Key Takeaways

Different stakeholders need different types of explanations: The paper identifies four distinct types of explanations—Diagnostic (how the model works internally), Explication (what the output means), Expectation (how the model generalizes), and Role (why we should use this model)—each serving different purposes for different audiences, meaning a single explanation method cannot satisfy all needs.
"Explanation" means different things in AI contexts: The paper addresses a critical ambiguity in XAI terminology, where practitioners and stakeholders often use the word "explanation" to mean fundamentally different things, leading to miscommunication about what XAI methods actually deliver and whether they're fit for purpose.
Explanations should identify intervention points: The paper frames explanations as tools for identifying where we can intervene in AI systems to create desired changes, suggesting that a good explanation is one that reveals actionable leverage points rather than simply providing information.
Role-explanations address social context: Beyond technical explanations of how models work, the paper emphasizes that stakeholders need "Role-explanations" that justify why a model should be used in a particular social context—a crucial consideration for responsible AI deployment that goes beyond technical transparency.
A pluralistic framework prevents misuse of XAI methods: By providing a taxonomy that matches explanation types to appropriate XAI methods, the paper offers practitioners and stakeholders a template to avoid using the wrong explanation tool for their actual need, reducing the risk of false confidence in AI systems.

Abstract

To be published in CD-MAKE 2021 conference proceedings

Explanatory Pluralism in Explainable AI

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links

Explanatory Pluralism in Explainable AI

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links