and consist of expressing rebellion. Rebellion can be
expressed through verbal or nonverbal communication (Briggs, McConnell, and Scheutz 2015). It can be
expressed behaviorally (for example, physically
resisting incorrect movements [Gregg-Smith and
Mayol-Cuevas 2015]). Or it can be expressed through
an internal change in the agent’s attitudes: inaction
(for example, acquiring the belief that the alter’s
behavior jeopardizes the team’s goals).
Post-rebellion covers behavior in the aftermath of a
rebellion episode, as the agent responds to the alter’s
or other witnesses’ reactions to rebellion. Post-rebellion can consist of reaffirming one’s objection or rejection (for example, the robot’s objection to an assigned
task becoming increasingly intense in the experiments
of Briggs, McConnell, and Scheutz ) or ceasing
to rebel. It may also consist of assessing and managing
inverse trust (Floyd and Aha 2016).
Rebel agents are not necessarily cognitively complex.
When they are, however, this creates interesting
challenges and opportunities pertaining to the
sociocognitive dimensions of their rebellion. We now
explore two such dimensions: social awareness and
counternarrative intelligence. Further sociocognitive
mechanisms involved in rebellion include emotion
and trust, which we briefly explored in previous work
(for example, Coman et al. 2017).
Rebellion and Social Awareness
Rebellion-aware agents can reason about rebellion
(their own and that of others) and its implications,
such as social risks. Rebellion-aware agents are not
necessarily rebels themselves. Such an agent might
attempt to assess, for example, whether a human or
AI teammate is inclined to rebel, or whether a human
alter is likely to interpret the rebellion-aware agent’s
own behavior as being rebellious (irrespective of
whether the agent is actually rebelling or not). Patil
et al. (2012) use machine learning techniques to predict which members will leave World of Warcraft
guilds and the potential impact of their departures.
One can imagine an AI agent using similar techniques to anticipate whether another agent will
rebel. A rebellion-unaware agent could conceivably
become rebellion-aware through various, possibly
human-inspired, processes (for example, by examining its own beliefs, interpreting the reactions of others to its behavior, or otherwise acquiring and applying social knowledge).
Naive rebel agents are rebellion-unaware rebels:
they deliberate on whether to trigger rebellion, but
do not reason about the social implications, consequences, and risks of rebellious attitudes. Apker,
Johnson, and Humphrey’s (2016) agent is a naive
rebel: it deliberates on whether it should rebel based
purely on its rules for activating contingency behavior, not on any social implications of rebellion.
Conflicted rebel agents are rebellion-aware rebels:
they can both rebel and reason about the implications
and consequences of rebellion. This capability can cre-
Table 2. Stages of Rebellion: Examples for the Three Scenarios.
Mover Personal Assistant
Pre-rebellion In addition to executing the
alter’s orders, the rebel agent
monitors the environment for
potential obstacles and threats.
The rebel agent monitors the alter’s
product-ordering and exercise-scheduling behavior for any unhealthy
patterns of behavior.
The rebel agent observes the social
interactions between the members of
the hiring committee to determine who
has high social capital (thus affording
to express their opinions freely) and
who does not (and may need support).
After each command, the rebel
agent projects future states to
determine if any are undesirable
to the alter or the agent itself.
The rebel agent checks whether its
threshold for tolerance of negative
health-related behavior (for example, a
maximum number of orders of highly-processed food per month) has been
The rebel agent assesses whether
committee member E (see table 1) has
low social capital and whether E’s
suggestion appears to have merit.
command to push the table
would endanger the alter’s
The rebel agent verbally informs
the alter that obeying the
The rebel agent challenges the alter
about the order he or she intends to place.
The rebel agent interrupts the
discussion to highlight the merits of E’s
Post-rebellion If the alter insists that the rebel
agent should push the table, the
agent re-assesses the danger and,
if appropriate, reiterates the
The rebel agent monitors the alter’s trust
in it after the rebellion episode.
The rebel agent monitors social
interactions to detect any ill will that
might be developing towards E as a
result of the intervention.