Mind Perception and Morality of Artificial Intelligence in Social Interaction
Abstract
Self-driving vehicles, autonomous military drones, rescue robots, dating websites, smart home security systems, targeted advertising, chatbots, medical recommender systems and a host of other artificially intelligent (AI) technologies may behave in ways that could be considered moral. How and when are the behaviors of and toward AIs judged as moral? According to current research the answer to the how is that the AI must be perceived to have mind and the behavior must be a social interaction that causes harm to another. Unlike humans who are the paradigm of fully minded entities having both agency and experience, AIs are cryptominds Ð somewhere in-between having and not having mind. Therefore, how behaviors are judged as moral depends on perceptions of the AIsÕ mind, but when this occurs depends on peopleÕs personal interaction. When do people perceive mind and morality in their personal interactions with AIs? To answer this, a qualitative survey will ask people to recall a time they personally interacted with an AI which was perceived to have mind or to be acting morally and the generalizability of these perceptions will be tested with others who did not experience this interaction (Study S1). How do different levels of perceived mind alter moral judgments? An experiment will answer this question by manipulating whether AIs are moral actors or recipient of moral action and possess either high or low mind (S2.1). How do moral judgments differ between AIs and humans? An experiment will answer this question by manipulating the focal agent as an AI or a human and as a moral actor or recipient of moral action (S2.2). Therefore, collectively these studies will indicate when people perceive personal moral interactions with AIs and demonstrate how different levels of agentic and experiential mind changes these perceptions, separating AI morality from human morality. Research find that perceptions of mind and morality judgments reciprocally influence each other. For cryptominds like AIs, this leads to a puzzle of whether people first perceive mind and then make moral judgments or vice versa. I argue that this puzzleÕs solution in real world social interactions involves peopleÕs personal and cultural sentiments about the AIs preceding and enabling both mind and morality perceptions. As people interact with a new AIs, they develop affective meanings, or sentiments, toward it. They may develop personal sentiments if the technology is not yet popularized through a technology diffusion process, or cultural sentiments Ð affective meanings that are shared at a cultural level Ð if it has been more widely diffused. But how do cultural and personal sentiments of AIs alter social interactions and mind and morality perceptions? Two surveys which compare AIs sentiments, perceived mind, and morality will answer this by having people rate a wide range of AIs in general Ð that is out of a situational context Ð and also in a variety of moral situations (S3). These studiesÕ findings about the morality of AIs can be generalized, codified, and therefore extended to AIs not studied in this project, and even not yet created. These findings will be integrated into affect control theory, a mathematical model that predicts affect, moral perceptions, emotion, behavior, and labeling based on human social interaction and sentiments using grounded equations. Based on the insights and examples of mind perceptions and moral interaction (S1, S2) and the grounded sentiment data (S3) an affect control theory module will be built to make predictions about morality perceptions of AIs. The theory and moduleÕs general predictions will enable simulations of emergent and future AIs based on ever-changing sentiments of them. As military robots, smart home systems, wearables and other AIs are created and diffused into cultural knowledge, this system can offer up-to-date predictions of general moral judgment and blame for their social behavior.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 06, 2019
- Source ID
- W911NF1910246
Entities
People
- Daniel Shank
Organizations
- Army Contracting Command
- Missouri University of Science and Technology
- United States Army