Arthur now also offers bias detection capabilities for NLP models, allowing data science teams to uncover differences in accuracy and other performance measures across different subgroups to identify—and fix—unfair model bias. The Arthur platform offers performance-bias analysis for natural language and tabular models, as well as the ability to partition by multiple attributes at a time to provide you more granular insights into potential biases.
The Arthur team has also released a new set of explainability tools for NLP models, providing token-level insights for language models. Organizations can now understand which specific words within a document contributed most to a given prediction, even for black-box models.
Arthur’s customers, including Fortune 100 companies like Humana and AI-driven startups like Expel and Truebill, are using the platform to ensure that they can catch and fix any issues with their production AI systems before they become billion-dollar problems. With Arthur’s new NLP monitoring capabilities, the platform can now offer advanced support for an entirely new class of models that are rapidly becoming fixtures in the enterprise.