AI transparency in practice: a report
by Lucy Smith
Fritzchens Fritz / Better Images of AI / GPU shot etched 5 / Licenced by CC-BY 4.0
A report, co-authored by Ramak Molavi Vasse’i (Mozilla’s Insights Team) and Jesse McCrosky (Thoughtworks), investigates the issue of AI transparency. The pair dig into what AI transparency actually means, and aim to provide useful and actionable information for specific stakeholders. The report also details a survey of current approaches, assesses their limitations, and outlines how meaningful transparency might be achieved.
The authors have highlighted the following key findings from their report:
- The focus of builders is primarily on system accuracy and debugging, rather than helping end users and impacted people understand algorithmic decisions.
- AI transparency is rarely prioritized by the leadership of respondents’ organizations, partly due to a lack of pressure to comply with the legislation.
- While there is active research around AI explainability (XAI) tools, there are fewer examples of effective deployment and use of such tools, and little confidence in their effectiveness.
- Apart from information on data bias, there is little work on sharing information on system design, metrics, or wider impacts on individuals and society. Builders generally do not employ criteria established for social and environmental transparency, nor do they consider unintended consequences.
- Providing appropriate explanations to various stakeholders poses a challenge for developers. There is a noticeable discrepancy between the information survey respondents currently provide and the information they would find useful and recommend.
Topics covered in the report include:
- Meaningful AI transparency
- Transparency stakeholders and their needs
- Motivations and priorities of builders around AI transparency
- Transparency tools and methods
- Awareness of social and ecological impact
- Transparency delivery – practices and recommendations
- Ranking challenges for greater AI transparency
You can read the report in full here. A PDF version is here.
, Managing Editor for AIhub.