People Perceive More Phantom Costs From Autonomous Agents When They Make Unreasonably Generous Offers

TL;DR

This study investigates phantom costs, defined as perceived hidden drawbacks, in both human–human and human–robot car sales scenarios. Using a 2×2×2 between-subjects design with agent type (human vs robot), autonomy (autonomous vs non-autonomous), and discount size (small vs large), the authors show that larger discounts increase phantom-cost perceptions and purchase intentions, while robots are perceived as less self-interested, reducing phantom costs and bolstering trust in some conditions. Phantom costs toward the car, the seller, and the manager originate from different sources and interact with discount and autonomy in nuanced ways, with self-interest mediating some, but not all, relationships. The findings have implications for ethical AI design, transparency in agent rationale, and marketing strategies that leverage robot representations to balance consumer trust and perceived risk in generous offers.

Abstract

People often reject offers that are too generous due to the perception of hidden drawbacks referred to as "phantom costs." We hypothesized that this perception and the decision-making vary based on the type of agent making the offer (human vs. robot) and the degree to which the agent is perceived to be autonomous or have the capacity for self-interest. To test this conjecture, participants (N = 855) engaged in a car-buying simulation where a human or robot sales agent, described as either autonomous or not, offered either a small (5%) or large (85%) discount. Results revealed that the robot was perceived as less self-interested than the human, which reduced the perception of phantom costs. While larger discounts increased phantom costs, they also increased purchase intentions, suggesting that perceived benefits can outweigh phantom costs. Importantly, phantom costs were not only attributed to the agent participants interacted with, but also to the product and the agent's manager, highlighting at least three sources of suspicion. These findings deepen our understanding of to whom people assign responsibility and how perceptions shape both human-human and human-robot interactions, with implications for ethical AI design and marketing strategies.

Figures (6)

• Figure 1: Pictures Depicting both a Robot (a) and Human (b) Sales Agent at a Car Dealership.
• Figure 2: Perceived Autonomy as a function of Agent, Discount, and Autonomy.
• Figure 3: Estimated Marginal Means of Phantom Costs Perception as a function of Agent, Discount, and Autonomy.
• Figure 4: Post Hoc Comparisons for Perception of Phantom Costs towards the Car, Agent, and Manager
• Figure 5: Estimated Marginal Means of Buying the Car as a function of Agent, Discount, and Autonomy.