Social decisions and fairness change when people’s interests are represented by autonomous agents

Abstract

In the realms of AI and science fiction, agents are fully-autonomous systems that can be perceived as acting of their own volition to achieve their own goals. But in the real world, the term "agent" more commonly refers to a person that serves as a representative for a human client and works to achieve this client's goals (e.g., lawyers and real estate agents). Yet, until the day that computers become fully autonomous, agents in the first sense are really agents in the second sense as well: computer agents that serve the interests of the human user or corporation they represent. In a series of experiments, we show that human decision-making and fairness is significantly altered when agent representatives are inserted into common social decisions such as the ultimatum game. Similar to how they behave with human representatives, people show less regard for other people (e.g., exhibit more self-interest and less fairness), when the other is represented by an agent. However, in contrast to the human literature, people show more regard for others and increased fairness when "programming" an agent to represent their own interests. This finding confirms the conjecture by some in the autonomous agent community that the very act of programming an agent changes how people make decisions. Our findings provide insight into the cognitive mechanisms that underlie these effects and we discuss the implication for the design of autonomous agents that represent the interests of humans.

Evolution of Collective Fairness in Hybrid Populations of Humans and Agents

“…For each combination of M = {1, N − 1} with N = {5, 10} we collected 100 pairs of strategies (p, q). We followed the strategy method (de Melo, Marsella, and Gratch 2018), informing people about the game rules, group size and group decision rule, and asking participants to submit their proposal as a Proposer (p) and the minimum acceptable offer as a Responder (q). We provide an eliminatory test, used to guarantee that individuals actually understood the rules of the MUG.…”

Human Cooperation When Acting Through Autonomous Machines

“…These machines are changing the traditional ways we engage with others, and it is important to understand those changes. Recent research confirms that interacting through machines can affect the decisions we make with others (5). These findings showed that in ultimatum, impunity, and negotiation games, people were less likely to accept unfair offers if asked to program a machine to act on their behalf compared with direct interaction with proposers.…”

“…However, to increase experimental control, participants always engaged with a computer script that simulated the other participants. Similar experimental manipulations have been used in other experiments studying behavior involving intelligent machines (5,38,39). Participants were fully debriefed about this experimental procedure at the end of the experiment.…”

The Effects of Experience on Deception in Human-Agent Negotiation

“…There is some evidence that individuals instruct their representatives to perform more fairly than they themselves would due to reputation (Ramchurn et al, 2003) or temporal effects (Pronin et al, 2008)due to a desire to be perceived by others as good or fair-but these effects persist even into anonymized scenarios (de Melo et al, 2016). Some theories maintain that, when considering what instructions to provide to one's representative, principals engage in higher level thinking about fairness and equity and other broad social goals than they might otherwise do in the heat of the moment (de Melo et al, 2018;Giacomantonio et al, 2010). Indeed, this may be the goal of some people who value indirect interactions-allowing "cooler heads to prevail" can often have direct benefits for all parties.…”

“…In a recent provocative paper in the Academy of Sciences, de Melo and colleagues examined the norm of fairness and showed that people treated each other more fairly when they interacted through AI systems (de Melo et al, 2018). Using several domains, including negotiation, they found that when people "programmed" the behavior of an agent to act on their behalf, the agent treated others more fairly than they would have themselves.…”