BEST PRACTICES - for AI and Workplace Assessment Technologies

When properly designed and utilized, AI can and must process vast amounts of personal data fairly and ethically, and can create more equitable access for people with disabilities, and people from underrepresented, marginalized and multi- marginalized communities.1 Because of the complexity of both issues and equities within the hiring and employment space, there is a clear need for best practices. [...] Governance frameworks should: i) Establish and document processes that map, measure, and manage reasonably foreseeable risks and harms to individuals that AI tools that have Consequential Impacts can pose; ii) Address the full product life cycle of AI tools, recognizing the different roles and responsibilities of Developers and Deployers within the life cycle of a tool and relevant informational a. [...] Individuals or teams tasked with compliance should clearly articulate the potential risks of an AI tool, identify the equity and legal considerations of the AI tool with respect to possible Consequential Impacts, and establish that the AI tool is fit for its intended purpose and consistent with legal and ethical standards. [...] Impact Assessments may include, but are not limited to: (i) Identification of the purpose and intended uses of the AI tool; (ii) The types of data used to train the AI tool; (iii) Mitigation measures for potential bias or other high-risk behaviors; (iv) Identified risks of the AI tool to individuals, communities, or society; (v) Testing structures that are in place for the AI tool post- deployment. [...] i) Developers providing AI tools with Consequential Impacts should disclose to Deployers when applicable: (1) the intended purposes of the AI tool; (2) purposes for which the AI tool is not intended; (3) known efficacy limits of the AI tool; (4) how the AI tool was trained; (5) whether the AI tool was assessed for potential discriminatory bias; (6) whether the AI tool uses information from Deploye.