Intelligent machines as social catalysts

Technical and Ethical Challenges

A robot that is a better social catalyst by being less socially suave than a human offers new and fascinating perspectives. Consider the technical and ethical challenges raised by robots that attempt at being social catalysts by imitating human social behavior as closely as possible. First, this strategy requires the robot to be endowed with sophisticated abilities to navigate social contexts, emotional states, and face-saving strategies and more generally, to approximate the theory of mind-related reasoning that humans are so very good at. Although social robotics is certainly making strides in that direction, we know that this strategy comes with formidable challenges (3): for example, overcoming the uncanny valley (4). A robot that attempts to be a good social catalyst by being intentionally less socially sophisticated than a human bypasses at least some of these challenges.

Second, a robot with abilities and behavior that are designed to make it as human like as possible raises specific ethical concerns. As some ethicists argued, this humanizing strategy may amount to unethical deception (5) and may muddle our perception of who is really accountable for the robot’s behavior (6). Similarly, some privacy concerns may arise when social robots are humanized to the extent that people form emotional bonds with them. In particular, these emotional bonds may prompt humans to interact in a more open or intimate way and can encourage them to share sensitive information that they would have otherwise kept private (7). In all of these cases, a robot with behavior that is a clear reminder that it is not trying to display human-level social skills can mitigate a variety of risks.

Consider now the technical and ethical challenges raised by robots that try to be social catalysts by behaving in a nonhuman way. First, they may succeed by using unconventional means at an unconventional scale. For example, robots may have access to vast amounts of data about us (beyond what would be required by a human-like approach), enabling it to personalize interventions to push people’s right buttons. This is not only a technical challenge, but also a privacy challenge, echoing current debates in the field of human resources analytics (8) and Surveillance Capitalism more broadly (9). In other words, being nonhuman may entail having dangerous (superhuman) influence capabilities.

Traeger et al. show that a robot uttering “vulnerable” statements can promote engagement within a human team both in absolute terms (total speaking time increases) and in relative terms (speaking time is more equally distributed).

Second, the efficacy of this approach may require (when possible) hiding the fact that a machine is acting as a social catalyst to avoid defensive reactions from humans who may be concerned about being manipulated by an opaque algorithm. This creates a tension between the pursuit of performance and the ethical pursuit of transparency and personal autonomy (10). Third, to seek social catalyzation through uncharted, nonhuman means creates the risk of adverse indirect effects, which are hard to anticipate precisely because there might be little precedent for what the machine is attempting.

Here too, programming a robot to use techniques that are common to human social catalysts but intentionally not as smooth may mitigate all of these risks. This strategy requires less sensitive data when the robot performs its role (since it does not need a subtle understanding of its teammates), it is by design transparent (since it works precisely because it is performed by a robot rather than a human), and using well-charted human techniques narrows down the possibility of totally unexpected indirect effects.

Future Directions

There is much to be done in the wake of the research of Traeger et al. (1). Indeed, our interpretation of their findings is entirely speculative as are most of the perspectives we sketched in this comment. An important step would be to identify other contexts in which a robot, merely by being a robot, can put to good use techniques that would not be optimal for a human social catalyst. For example, there is research showing that a robot can helpfully call out hostile or offensive behavior directed by a human teammate to another (11). This is a delicate role when performed by a human because calling out the behavior of another can be perceived as judgmental and increase tension and hostility as a result, adversely affecting the social dynamics of the team. Accordingly, it requires subtle attention to emotional shifts and a sophisticated mastery of politeness protocols. It is fascinating to think that a robot may be able to bluntly perform that role without the need for subtle conversational moves precisely because a blunt call out can emphasize that the robot is a nonjudgmental machine.

Another area of great potential is interaction with children. Recent work showed that robots that interact with children as peers could promote growth mindsets in children, thus facilitating learning (12). Simple robots were also used in therapy for children with Autism Spectrum Disorder (13). There is evidence that, when interacting with such robots, the children directed more speech at the robot than humans (14). Robots may offer a new role that facilitates all manners of learning and therapy in children precisely because they are not human like and thus, may not evoke negative feelings, such as fear of competition or disappointment.

Even more broadly, humans increasingly interact with machines in an ecology that contains other humans and other machines (Fig. 1). This means that we might not only engineer individual robots, but also machine collectives that are able to influence humans (and each other) on vast scales. This raises many questions about understanding the emergent behavior of these machines and the emergent hybrid human–machine system (15). The vulnerable robot of Traeger et al. (1), with its corny jokes and bogus anecdotes, puts a friendly face on this social order; however, even that robot may face its own ethical issues: for example, fairness. Inclusive and culturally sensitive social catalyzation arguably requires greater social sophistication. If we intentionally simplify the social capabilities of a machine for the purpose of greater average performance, we may inadvertently reduce its ability to recognize that different people may face different challenges or have different expectations when contributing to a team effort. This is not a criticism of the research of Traeger et al. (1), who never suggested downgrading the social capabilities of the robot teammate. It is merely a pointer for those who, like us, believe that this research is opening fascinating perspectives about when and why robots could paradoxically end up being better social catalysts by doing things that would not work well for humans.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Scales of modeling human and machine behavior.