When a human first encounters a social robot, the interaction is charged with curiosity. Users are highly engaged, deeply expressive, and remarkably forgiving of technical flaws. In human-robot interaction (HRI) literature, this honeymoon phase is known as the novelty effect.
The fundamental problem in HRI is that most studies are brief, single-session encounters. As Micol Spitale et al. point out in Past, Present, and Future: A Survey of The Evolution of Affective Robotics For Well-being, relying on these short-term interactions creates an illusion of success. When a robot moves out of the lab and into a multi-week or multi-month deployment, this initial charm decays. To build robots that can genuinely assist in long-term contexts—like healthcare, education, or wellness—we must understand the underlying mechanics of engagement decay and design systems that survive the end of novelty.
The Mechanics of Engagement Decay
Engagement decay is not a random loss of interest; it is a predictable psychological and behavioral shift driven by three distinct mechanisms.
1. Behavioral Predictability and Habituation
Humans are highly efficient pattern-recognition engines. When a robot operates on fixed, heuristic-based rules, users quickly decode its underlying logic. In Motivating Physical Activity via Competitive Human-Robot Interaction, Boling Yang et al. observe that as soon as humans predict a robot's behavioral patterns, the interaction loses its cognitive challenge. This rapid habituation causes users to lose interest, causing a sharp drop-off in the quality and frequency of the interaction.
2. The "Muting" Effect
As the novelty of a physical robot fades, human social behavior itself undergoes a structural stabilization. By tracking long-term interactions, Sidney Richardson et al. in REACT: Two Datasets for Analyzing Both Human Reactions and Evaluative Feedback to Robots Over Time discovered a "muting" effect: over time, humans naturally become less nonverbally expressive.
[Initial Encounter] ───> High Expressiveness, Exaggerated Gestures (Novelty)
[Repeated Exposure] ───> Behavioral Stabilization, Subdued Cues (Muting Effect)
If a robot’s affective computing models are static, they will misinterpret this natural drop in expressiveness as a sign of user dissatisfaction or system failure, when it is actually a sign of habituation.
3. Social Strategy Fatigue
Social mechanisms that delight a user during a first encounter can become actively counterproductive over extended horizons. For example, Cong Shi 0001 demonstrates in Reverse Psychology in Trust-Aware Human-Robot Interaction that while manipulative social tactics like reverse psychology can successfully drive short-term human performance, their long-term application erodes trust and undermines the human-robot relationship.
A similar fatigue occurs with robotic apologies. In "Oh, Sorry, I Think I Interrupted You": Designing Repair Strategies for Robotic Longitudinal Well-being Coaching, Minja Axelsson et al. found that highly emotional, empathic conversational repairs work well initially but quickly begin to feel scripted and disingenuous when repeated across multiple sessions. Over time, users prefer brief, transparent technical explanations over simulated empathy.
The Evolution of User Boundaries
Long-term exposure does more than just lower engagement; it fundamentally changes what users expect and tolerate from a robotic companion.
Initially, users approach a robot with open-ended curiosity. However, longitudinal in-home studies show that prolonged exposure forces users to ground their expectations. In Promising directions for human-robot interactions defined by older adults, Anastasia K. Ostrowski et al. demonstrate that long-term interaction does not universally increase technology acceptance. Instead, it makes users far more certain about their boundaries. Over months of exposure, users actively reject invasive emotional or physical monitoring, shifting their preferences toward strict, transactional utility.
Concurrently, the way users perceive system failures changes. In Longitudinal Evolution of Coachees' Behavioural Responses to Interaction Ruptures in Robotic Positive Psychology Coaching, Micol Spitale et al. found that as users habituate to a robot's social scripts and technical limitations over multiple sessions, their perception of interaction ruptures decreases. They learn to adapt to the machine's flaws, treating it less like an intelligent social agent and more like a routine tool.
Designing for the Long Horizon
To build robotic systems whose efficacy survives the decay of novelty, researchers have converged on three core design strategies.
Injecting Strategic Unpredictability
If deterministic behavior causes rapid habituation, the solution is to design architectures that maintain a degree of behavioral entropy. Shahabedin Sagheb et al. tackle this in Towards Robots that Influence Humans over Long-Term Interaction by optimizing for "belief entropy." By intentionally making the robot's coordination strategies less predictable, the system prevents the human from completely decoding and bypassing its behavior, thereby sustaining long-term influence.
Dynamic Personalization and Memory
A robot cannot maintain a social bond if it treats a user like a stranger in every session. As noted by Nathalia Céspedes et al. in A Socially Assistive Robot for Long-Term Cardiac Rehabilitation in the Real World, avoiding engagement decay requires personalization mechanisms that actively recall individual user history and track specific longitudinal progress. This memory layer transforms repetitive scripts into an evolving, continuous dialogue.
Anchoring to High-Value Tasks
Social charm alone is insufficient to sustain long-term HRI. Cory D. Kidd argues in A Robotic Weight Loss Coach that the long-term integration of a robot into daily life depends heavily on the presence of a high-value, structured task (such as chronic health management).
When a robot addresses a genuine, ongoing human need and pairs it with adaptive communication—such as large language models that tailor interactions based on a user's fluctuating stress levels, as shown by Guy Laban et al. in What People Share With a Robot When Feeling Lonely and Stressed and How It Helps Over Time—the therapeutic and functional utility of the system remains stable long after the novelty has vanished.