Better Transparency: Introduction to Contextual Transparency for Automated Decision Systems

LinkedIn Recruiter, a search tool used by professional job recruiters to find candidates for open positions, would work better if recruiters knew exactly how LinkedIn generates their responses to search queries, possible through a framework called “contextual transparency.” .

That’s what a team of researchers led by Mona Sloane of NYU’s Tandon School of Engineering, Principal Research Scientist at NYU’s Center for Responsible AI and Research Assistant Professor in the Department of Technology, Culture, and Society, advances in a provocative new study published in Nature Machine Intelligence.

The study is a collaboration with Julia Stoyanovich, associate professor at the Institute of Computer Science and Engineering, associate professor of data science, and director of the Center for Responsible AI at New York University, as well as Ian René Solano-Kamaiko. , Ph.D. student at Cornell Tech; Aritra Dasgupta, assistant professor of data science at the New Jersey Institute of Technology; and Jun Yuan, Ph.D. Candidate at the New Jersey Institute of Technology.

It introduces the concept of contextual transparency, essentially a “nutritional label” that would accompany the results delivered by any Automated Decision System (ADS), a computer system or machine that uses algorithms, data, and rules to make decisions without human intervention. The label would expose explicit and hidden criteria (ingredients and recipe) within the algorithms or other technological processes used by the ADS in specific situations.

LinkedIn Recruiter is an example of real-world ADS: it “decides” which candidates best fit the recruiter’s desired criteria, but different professions use ADS tools in different ways. The researchers propose a flexible model for constructing contextual transparency, the nutrition label, making it highly context-specific. To do this, they recommend three “contextual transparency principles” (CTPs) as the basis for building contextual transparency, each of which is based on an approach related to an academic discipline.

• CTP 1: Social Sciences for Stakeholder Specificity: This is intended to identify the professionals who rely on a particular ADS system, how exactly they use it, and what information they need to know about the system to do their job better. This can be accomplished through surveys or interviews.
• CTP 2: Engineering for ADS Specificity: This is intended to understand the technical context of the ADS used by the relevant interested parties. Different types of ADS operate with different assumptions, mechanisms, and technical restrictions. This principle requires an understanding of both the input, the data that is used in decision making, and the output, how the decision is delivered.
• CTP 3: Design for transparency and specificity of results: This is to understand the link between the transparency of the process and the specific results that the ADS system would ideally deliver. In recruiting, for example, the result could be a more diverse pool of candidates facilitated by an explainable ranking model.

The researchers looked at how contextual transparency would work with LinkedIn Recruiter, in which recruiters use Boolean searches, AND OR NOT typed queries, to receive ranked results. The researchers found that recruiters do not blindly trust ADS-derived rankings and typically double-check ranking results for accuracy, often by going back and adjusting keywords. Recruiters told investigators that ADS’s lack of transparency challenges efforts to recruit for diversity.

To address the transparency needs of recruiters, the researchers suggest that the contextual transparency nutrition label include both passive and active factors. Passive factors comprise information that is relevant to the overall operation of the ADS and professional contracting practice in general, while active factors comprise information that is specific to the Boolean search string and therefore changes.

The nutrition label would be inserted into the typical workflow of LinkedIn Recruiter users, providing them with information that would allow them to assess the degree to which ranked results match their original search intent and refine the boolean search string accordingly to generate better results. results.

To assess whether this ADS transparency intervention achieved the change that could reasonably be expected, the researchers suggest using interviews with stakeholders about the potential change in ADS use and perception along with participant diaries documenting the practice. Professional and A/B testing (if possible).

Contextual transparency is an approach that can be used for AI transparency requirements that are mandated in new and future AI regulation in the US and Europe, such as New York City Local Law 144 of 2021 or the EU AI Law.