Author: Denis Avetisyan
New research reveals that communicating via AI-mediated video, especially with realistic avatars, diminishes feelings of trust and confidence in judgments, even if it doesn’t improve lie detection.
AI-mediated communication decreases perceived trustworthiness and confidence in judgments, potentially eroding relational aspects of online interactions.
While increasingly prevalent, computer-mediated communication raises concerns about its impact on fundamental social judgments. This research, titled ‘Through the Looking-Glass: AI-Mediated Video Communication Reduces Interpersonal Trust and Confidence in Judgments’, investigates how artificially-mediated video-including retouching, background replacement, and avatars-affects perceptions of trustworthiness and lie detection accuracy. Across two experiments with [latex]\mathcal{N}=2000[/latex] participants, findings reveal that AI mediation diminishes perceived trust and confidence in judgments, particularly when strong AI tools like avatars are employed, despite not affecting actual accuracy in detecting deception. Does this erosion of relational confidence signal a broader need for trustworthy design principles in AI-mediated communication technologies?
The Fragile Foundation of Digital Trust
Human evaluation of trustworthiness in communication is a deeply ingrained process, extending far beyond the literal meaning of words. Individuals instinctively process a complex interplay of verbal and nonverbal cues – encompassing factors like tone of voice, facial expressions, body language, and even subtle physiological signals – to gauge the sincerity and reliability of others. This assessment isn’t merely conscious deliberation; rather, it’s a rapid, often unconscious computation rooted in evolutionary pressures favoring accurate social perception. Successful social interaction historically depended on the ability to quickly distinguish cooperative individuals from those with deceptive intentions, and this reliance on multi-modal cues remains fundamental to human communication even today. Consequently, any alteration to the availability or authenticity of these cues-as is increasingly common in digital environments-can significantly disrupt established mechanisms for building trust and detecting dishonesty.
The rise of AI-mediated communication-encompassing everything from automated chatbots to deepfake videos-represents a significant shift in how humans perceive and establish trust. Traditionally, individuals assess credibility through a complex interplay of verbal content and nonverbal cues like facial expressions, tone of voice, and body language. However, AI systems often strip away or artificially construct these cues, presenting communication devoid of genuine emotional signals or laden with deliberately misleading ones. This disruption doesn’t simply reduce the amount of information available; it fundamentally alters the reliability of the signals themselves, potentially overriding evolved mechanisms for deception detection and trustworthiness assessment. Consequently, individuals interacting with AI-mediated communication may struggle to accurately gauge intent, leading to increased susceptibility to manipulation or a generalized erosion of trust in digital interactions.
The pervasive shift towards digital communication fundamentally challenges established methods of assessing truthfulness and forming reliable judgments. Historically, humans have relied on a complex interplay of verbal messaging, facial expressions, body language, and vocal tonality – all cues readily available in face-to-face interactions – to gauge the sincerity of others. However, interactions mediated by artificial intelligence often strip away these crucial nonverbal signals, or present them in distorted or artificial forms. This presents a significant cognitive hurdle, forcing individuals to recalibrate their deception detection strategies and potentially increasing susceptibility to manipulation. Consequently, research is now focused on understanding how these altered cue landscapes affect our ability to accurately assess trustworthiness, and whether new cognitive biases emerge in these increasingly digital exchanges.
The Architecture of Belief and Expectation
Truth-Default Theory, originating from the work of Kravitz and Plous (1992), posits that humans do not instinctively assume deception; instead, individuals begin with an assumption that statements are truthful. This cognitive bias functions as a heuristic, reducing the cognitive load associated with constant skepticism. The theory details that this baseline expectation of honesty requires compelling evidence to be overcome, meaning individuals will only suspect deception when presented with cues-verbal, nonverbal, or contextual-sufficiently strong to negate the initial assumption of truthfulness. Critically, the threshold for detecting deception is higher than that for accepting truth, contributing to a systematic error where deceptive attempts are often missed or underestimated.
Expectancy Violation Theory posits that individuals develop expectations regarding the behavior of others, particularly in communicative contexts; when these expectations are violated, it creates a psychological and communicative tension that can negatively impact trust. The magnitude of this impact is determined by the valence of the violation – whether it is perceived as positive or negative – and the strength of the prior expectation. Significant or repeated violations, especially those perceived as negative, lead to increased uncertainty and decreased trust, as the individual questions the communicator’s motives and credibility. This is due to a cognitive shift requiring increased mental processing to reconcile the observed behavior with pre-existing beliefs about appropriate conduct.
Trust and deception theories, specifically Truth-Default Theory and Expectancy Violation Theory, offer a framework for evaluating the impact of AI-driven alterations to communication patterns. Because humans typically assume honesty absent compelling evidence to the contrary, even subtle deviations from established communicative norms – such as those introduced by AI in phrasing, response time, or emotional expression – can activate suspicion. These alterations, perceived as violations of expected behavior, necessitate increased cognitive processing to assess credibility and may trigger negative emotional responses, ultimately leading to distrust even if no intentional deception is present. Consequently, understanding these theoretical underpinnings is crucial for anticipating and mitigating potential skepticism towards AI-mediated communication.
The Erosion of Signals: How AI Alters Perception
Avatar-mediated communication introduces discrepancies in nonverbal cues due to the inherent limitations of digitally recreating human expression. These digital representations often fail to accurately convey the full range of subtle cues – such as micro-expressions, nuanced body language, and realistic vocal intonation – that individuals rely on during face-to-face interactions. This mismatch between expected and received cues can violate established social norms and lead to perceptions of artificiality or insincerity, consequently hindering the development of trust between communicators. The degree to which these discrepancies impact trust is further influenced by the communication context, specifically whether the interaction occurs in a mixed or homogeneous environment.
Video filters, commonly used in digital communication, introduce alterations to visual cues such as facial expressions and skin tone. These manipulations, even when subtle, can raise concerns regarding the authenticity of the displayed information and increase the recipient’s suspicion. The introduction of artificial modifications to naturally occurring visual signals disrupts established expectations regarding nonverbal communication, potentially leading to decreased trust and altered perceptions of the sender’s credibility. While not necessarily indicative of deception, these filtered cues introduce ambiguity that can negatively impact judgment and interpretation of communicated messages.
Communication environment significantly moderates the effect of avatar use on trust. Research indicates that avatar-mediated communication in mixed environments – those where some participants are physically present and others are represented by avatars – results in a statistically significant decrease in trust. Specifically, trust levels were reduced by 0.08 (95% Confidence Interval: [-0.09, -0.03]) in these mixed environments, demonstrating a quantifiable negative impact on interpersonal trust compared to homogeneous environments where all participants are either physically present or represented by avatars.
Evaluations of deception accuracy across all tested conditions demonstrated performance remained at chance levels, ranging from 52% to 54%. However, the use of avatars in mixed communication environments – where some participants interacted face-to-face and others via avatar – resulted in a statistically significant reduction in confidence regarding these deception judgments. Specifically, confidence levels decreased by -0.046 (95% Confidence Interval: [-0.06, -0.03]), indicating that while participants were no worse at detecting lies, they reported substantially less certainty in their assessments when evaluating avatars within a mixed interaction setting.
Toward a Calculus of Confidence: Statistical Insights and Future Directions
Recent advancements leverage Response Error Modeling (REM) as a robust statistical approach to disentangle the multifaceted relationships between artificially altered communication cues, levels of interpersonal trust, and the accuracy of detecting deception. This framework moves beyond simple accuracy metrics by explicitly accounting for errors in judgment arising from both perceptual distortions – how AI alterations impact cue processing – and response biases – systematic tendencies to over- or underestimate deception. By modeling these error components separately, REM offers a nuanced understanding of how AI-mediated cues influence judgment, rather than merely whether they do. The technique allows researchers to isolate the specific contributions of altered cues to both correct and incorrect deception assessments, revealing whether changes in cues lead to genuine improvements in detection or simply shift response patterns. Ultimately, REM provides a powerful tool for evaluating the efficacy and potential pitfalls of employing AI in contexts where accurate assessment of trustworthiness is paramount, and for informing the design of more reliable and transparent communication systems.
The accuracy with which individuals assess deception is significantly influenced by their level of confidence in those judgments, acting as a critical intermediary between manipulated communicative cues and overall performance. Research indicates that alterations to typical behavioral signals – those normally associated with truthful or deceptive communication – don’t simply impair accuracy directly, but rather erode a person’s confidence in their ability to correctly identify lies. This diminished confidence, in turn, leads to poorer deception detection rates, even when individuals might otherwise possess the capacity to accurately assess truthfulness. Consequently, interventions aimed at bolstering judgment confidence, or those that focus on preserving the integrity of natural communicative cues in AI-mediated interactions, may prove vital in maintaining reliable deception detection capabilities.
Investigations must now turn toward actionable strategies for safeguarding interpersonal trust in an age increasingly shaped by artificial intelligence. Current research highlights the potential for AI-mediated alterations – even subtle ones – to erode confidence and impair accurate judgment of truthfulness. Future studies should prioritize the development of communication protocols that emphasize transparency regarding AI involvement, potentially through clear disclosures or verifiable authenticity markers. Furthermore, exploring methods to enhance human cognitive resilience against manipulative cues, such as training programs focused on critical thinking and emotional awareness, is paramount. Ultimately, fostering a framework of responsible AI implementation, where technology serves to augment rather than undermine human connection, will be crucial for maintaining social cohesion and effective communication in the years to come.
The study illuminates a concerning paradox within AI-mediated communication. While proponents often emphasize efficiency, this research suggests a diminishment of crucial relational factors-specifically, interpersonal trust. This erosion isn’t a matter of detecting deception more effectively, but of fostering an environment where judgments are made with reduced confidence. As Robert Tarjan aptly stated, “A program is a good idea that has become a program.” This article serves as a cautionary example; a technological ‘solution’-AI-mediated communication-introduces unforeseen consequences regarding trust and credibility, demonstrating that elegant code doesn’t necessarily translate to sound social dynamics. The core finding-decreased confidence-highlights a need for a more holistic evaluation of these technologies, extending beyond mere functional performance.
Further Reflections
The observed decrement in perceived trustworthiness is not a mere curiosity. It suggests a fundamental recalibration is occurring in how humans assess one another through digital channels. The study isolates the effect of AI mediation, but the broader context is a world increasingly populated by synthetic proxies. Future work must move beyond detecting falsehoods-a task at which humans consistently underperform regardless-and toward understanding the subtle costs imposed by these intermediaries on relational integrity.
A particularly austere challenge lies in disentangling the effect of strong AI – convincingly human avatars – from the more general anxieties surrounding computer-mediated communication. Does the aversion stem from a distrust of the technology itself, or from a subconscious recognition of the manipulated signal? This distinction is crucial, as the former implies a solvable engineering problem, while the latter points to a more deeply rooted psychological phenomenon.
The pursuit of perfect fidelity – the creation of avatars indistinguishable from real people – appears increasingly misguided. Perhaps the goal should not be to eliminate the trace of mediation, but to establish new, transparent conventions for digital interaction-a sort of ‘digital honesty’-that acknowledge the inherent artificiality of the channel. The erosion of trust is not inevitable; it is a design failure waiting to be corrected.
Original article: https://arxiv.org/pdf/2603.18868.pdf
Contact the author: https://www.linkedin.com/in/avetisyan/
See also:
- Gold Rate Forecast
- Seeing in the Dark: Event Cameras Guide Robots Through Low-Light Spaces
- American Idol vet Caleb Flynn in solitary confinement after being charged for allegedly murdering wife
- Hatch Dragons Beginners Guide and Tips
- Magicmon: World redeem codes and how to use them (March 2026)
- eFootball 2026 is bringing the v5.3.1 update: What to expect and what’s coming
- eFootball 2026 Epic Italian Midfielders (Platini, Donadoni, Albertini) pack review
- HEAVENHELLS: Anime Squad RPG WiTCH Tier List
- Ethereum’s Wild Ride: From $1,750 to $2,200 – A Comedic Tale of Liquidity and Losses!
- Disgraced ex-radio star Marty Sheargold doubles down on his sexist Matildas outburst as he is grilled on ACA in first television interview since sacking… as Allison Langdon says she was ‘offended’ by his behaviour
2026-03-22 09:38