Presentation
Tools for Foresight and Divergent Thinking about AI in Sociotechnical Systems
DescriptionThere is a growing imperative to use artificial intelligence (AI) for high consequence work, including various applications in healthcare. While it is relatively straightforward to employ human factors methods to identify gaps, develop AI concepts, and elicit requirements from user communities, it can be difficult to predict how AI will behave “in the wild.” This is largely because emergent behaviors and activities often result from the integration of a new AI capability into a sociotechnical system (i.e., people, technology, and work).
We developed participatory tools to increase foresight of how AI may integrate into a sociotechnical system. Originally developed to support design efforts for engineering more trustworthy AI, these tools also offer value for education and training purposes by encouraging exploratory thinking about AI’s impacts on different users and stakeholders. We will provide an overview and instruction of how to employ two different tools to enable multidisciplinary teams to collaboratively explore the impact of an AI solution on the sociotechnical systems into which they will be integrated.
First, the Evidence-based List of Exploratory Questions for AI Trust Engineering (ELATE) is a deck of 50+ cards to be used by AI development teams, human-centered engineers, end users, and other stakeholders of a real or hypothetical AI-enabled system. Each ELATE card includes exploratory questions to drive discussion, and a real-world event where people gained or lost trust in AI (which serves as the basis for the questions).
Second, the Targeted Premortem (TPM) is a variant of Klein’s Premortem Technique that uses ELATE as a basis for discussion prompts (instead of a ‘blue sky’ approach). The TPM has the group collectively envision a future state where people lost trust in the AI under development, and brainstorm the reasons for losing trust.
We will discuss how both tools can be employed as a rapid, low-fidelity means to simulate future potential issues with AI, encouraging divergent thinking and consideration of perspectives that individuals otherwise might not have considered.
We developed participatory tools to increase foresight of how AI may integrate into a sociotechnical system. Originally developed to support design efforts for engineering more trustworthy AI, these tools also offer value for education and training purposes by encouraging exploratory thinking about AI’s impacts on different users and stakeholders. We will provide an overview and instruction of how to employ two different tools to enable multidisciplinary teams to collaboratively explore the impact of an AI solution on the sociotechnical systems into which they will be integrated.
First, the Evidence-based List of Exploratory Questions for AI Trust Engineering (ELATE) is a deck of 50+ cards to be used by AI development teams, human-centered engineers, end users, and other stakeholders of a real or hypothetical AI-enabled system. Each ELATE card includes exploratory questions to drive discussion, and a real-world event where people gained or lost trust in AI (which serves as the basis for the questions).
Second, the Targeted Premortem (TPM) is a variant of Klein’s Premortem Technique that uses ELATE as a basis for discussion prompts (instead of a ‘blue sky’ approach). The TPM has the group collectively envision a future state where people lost trust in the AI under development, and brainstorm the reasons for losing trust.
We will discuss how both tools can be employed as a rapid, low-fidelity means to simulate future potential issues with AI, encouraging divergent thinking and consideration of perspectives that individuals otherwise might not have considered.
Event Type
Oral Presentations
TimeMonday, March 232:30pm - 3:00pm EDT
LocationMorgan
Simulation and Education