Social Attitudes of AI Rebellion: A Framework

Human attitudes of objection, protest, and rebellion have undeniable potential to bring about social benefits, from social justice to healthy balance in relationships. At times, they can even be argued to be ethically obligatory. Conversely, AI rebellion is largely seen as a dangerous, destructive prospect. With the increase of interest in collaborative human/AI environments in which synthetic agents play social roles or, at least, exhibit behavior with social and ethical implications, we believe that AI rebellion could have benefits similar to those of its counterpart in humans. We introduce a framework meant to help categorize and design Rebel Agents, discuss their social and ethical implications, and assess their potential benefits and the risks they may pose. We also present AI rebellion scenarios in two considerably different contexts (military unmanned vehicles and computational social creativity) that exemplify components of the framework. Society, Ethics, and AI Rebellion In human social contexts, attitudes of resistance, objection, protest, and rebellion are not necessarily destructive and antisocial; they serve a variety of fundamentally positive, constructive social functions. At a macro-societal level, protest can support social justice. At a micro level, saying “no” in a constructive way can help maintain healthy balance in personal and professional relationships (Ury, 2007). In many cases, rebellious attitudes are arguably not merely acceptable, but ethically obligatory, e.g. an engineer refusing to continue working on a project if a number of safety issues are not addressed. In contrast, AI rebellion is generally perceived as being fundamentally destructive: not just antisocial, but antihuman, a narrative reinforced by numerous sci-fi depictions in which AI follows in the footsteps of various mythical creatures to play the part of the ominous “other”. Such manifestations of rebellion are generally attributed to post-singularity AI with mysterious but decidedly dangerous inner workings. Copyright © 2017, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. We believe that AI attitudes of constructive rebellion can in many ways contribute to “maximizing the societal benefit of AI”, an AI research priority expressed by Russell, Dewey, and Tegmark (2015), by enabling refusal of unethical behavior, supporting value alignment with human groups (e.g., through protest on behalf of humans), maintaining safety, supporting task execution correctness, enhancing social co-creativity, and providing or supporting diverse points of view. As we will show through two scenarios and various smaller examples, such instances of AI rebellion neither require human-level intelligence or superintelligence nor involve rebelling against humanity as a whole. We are especially interested in collaborative, human-AI interaction environments, such as the long-term collaborations envisioned by Wilson, Arnold, and Scheutz (2016). In such contexts, AI rebellion has benefits comparable to those it has in human social contexts and associated risks pertaining to the maintenance of the long-term collaborations. The two scenarios that we present are drawn from the fields of (1) military unmanned vehicles and (2) computational social creativity. The first scenario is based on preexisting work of established practical interest, while the second is largely speculative. To facilitate this discussion, we define AI Rebel Agents and propose an initial framework for their study. A reduced version of this framework is described in (Aha and Coman, 2017). Rebel Agents are AI agents that can develop attitudes of opposition to goals or courses of action assigned to them by other agents, or to the general behavior of other agents. These attitudes can result in resistance, objection, and/or refusal to carry out tasks, or in challenging the attitudes or behaviors of other agents. We use “rebellion” as an umbrella term covering reluctance, protest, refusal, rejection of tasks, and similar stances/behaviors. We call an agent against which one rebels an Interactor. We assume that the Interactor is in a position of power in relation to the Rebel Agent; the source(s) and nature of that power can vary. The Interactor can be human or synthetic, an individual or a group. A Rebel Agent is not intended to be permanently adversarial towards the Interactor or in a rebelling state by default. A Rebel Agent has potential for rebellion that may or may not manifest based on external and internal conditions. An AI agent can be specifically designed to be a Rebel Agent (rebel by design), but rebellious behavior can also emerge unintendedly from the agent’s autonomy model (emergent rebellion). Our proposed framework for AI rebellion includes types of rebellion and stages of the rebellion process. The framework is applicable to both types of rebellion introduced above: (1) it can be used to guide the development and implementation of intentionally Rebel Agents, and (2) to categorize and study the rebellion potential and ramifications of emergent rebels (including their dangerous AI potential: while we argue that AI rebellion can, in certain instances, be positive and beneficial, we do not claim that it is necessarily so). The framework also (3) facilitates discussion of social and ethics-related aspects and implications of rebellion: we demonstrate this by examining dimensions of AI rebellion with strong social implications (emotion and social capital: particularly, trust) and by including ethics-related questions pertaining to the framework throughout the paper. Our framework is meant to be generally applicable to AI agents, with no restrictions on agent architecture, paradigm, purpose, deployment context, or other factors. However, the type and features of the agent will affect how it instantiates the components of the framework.