Transparency in Language Generation: Levels of Automation
This addresses transparency issues for consumers and stakeholders in AI-generated content, but it is incremental as it adapts an existing taxonomy framework.
The paper tackles the problem of misleading consumers due to advanced language models by proposing a taxonomy based on SAE driving automation levels to establish shared terms for describing automated language, aiming to increase transparency in the field.
Language models and conversational systems are growing increasingly advanced, creating outputs that may be mistaken for humans. Consumers may thus be misled by advertising, media reports, or vagueness regarding the role of automation in the production of language. We propose a taxonomy of language automation, based on the SAE levels of driving automation, to establish a shared set of terms for describing automated language. It is our hope that the proposed taxonomy can increase transparency in this rapidly advancing field.