Leiden University Scholary Publications

Social, Cognitive, and Affective Decision Makin

Radovanović, K.

Lorskens, A.

Schütte, S.

Bräuer, J.

Call, J.

Haun, D. B. M.

Leeuwen, E. J. C. van

Bonobos respond aversively to unequal reward distributions

Diana, F.

Hortensius, R.

Kret, M.E,

Can robots trust us? Neuropsychophysiological insights into honesty
towards artificial agents
Diana, F.; Hortensius, R.; Kret, M.E,
Citation
Diana, F., Hortensius, R., & Kret, M. E. ,. (2025). Can robots trust us?:
Neuropsychophysiological insights into honesty towards artificial agents.
doi:10.1109/HRI61500.2025.10974055
 
Version: Publisher's Version
License: Licensed under Article 25fa Copyright Act/Law (Amendment Taverne)
Downloaded from: https://hdl.handle.net/1887/4254687
 
Note: To cite this publication please use the final published version (if applicable).
Can Robots Trust Us? Neuropsychophysiological 
Insights into Honesty Towards Artificial Agents 
Fabiola Diana 
Department of Cognitive Psychology 
Leiden University 
Leiden, The Netherlands 
d.fabiola@fsw.leidenuniv.nl 
Ruud Hortensius 
Department of Psychology 
Utrecht University 
Utrecht, The Netherlands 
r.hortensius@uu.nl 
Mariska E. Kret 
Department of Cognitive Psychology 
Leiden University 
Leiden, The Netherlands 
m.e.kret@fsw.leidenuniv.nl 
Abstract— While research often examines human trust in robots, 
humans frequently cheat in interactions—even simple ones. This 
raises a neglected question: can robots trust us? Our 
interdisciplinary, cross-cultural research explores honesty in 
human-robot interaction by capturing real-time 
psychophysiological signals through a comparative robotics 
framework. We investigate whether social cues, like pupil size, that 
influence human-human decisions similarly impact human-robot 
interactions. Future work will expand this to study how human 
physiological and neural responses shape behavior in social 
dilemmas with human vs artificial partners, guiding the 
development of agents attuned to human signals. 
Keywords—psychophysiology, HRI, honesty, network analysis 
I. INTRODUCTION 
Trust underpins human collaboration and societal progress. 
This is no different when the interaction partner is artificial: to 
collaborate or have a meaningful relationship with social robots, 
chatbots, or other forms of artificial agents, people must trust 
them in the first place, a question widely explored in literature 
[1]. Yet, trust is a two-way street. Our research explores a 
complementary, overlooked question: can artificial agents “trust” 
us? People are often tempted to cheat, [2] as evidenced by 
dishonest behaviors that entail a cost to society, like tax fraud, 
downloading illegal software and music, or bypassing self-
checkouts. As artificial agents play an increasingly important role 
in our society, a deep understanding of (dis)honest behaviors 
towards robots, and of how mutual trust can be established 
between humans and robots is a significant challenge that must 
be addressed. 
Deciding to act honestly or not is always a dilemma between 
what is good for oneself and what is beneficial for the other 
individual. This can be influenced by cognitive processes – past 
experiences, cultural norms, and cost-benefit analyses – but also 
by more intuitive unconscious processes that often override 
cognition-driven evaluations [3], especially in absence of 
previous information. These rapid, automatic judgments about 
others are based on a complex interplay between a variety of 
conscious behavioral signals (e.g., facial expressions, posture) 
[4,5], unconscious autonomic cues (e.g., pupil size, blushing) [6] 
and bodily responses (e.g., heart rate, skin conductance, brain 
activity) [7], which facilitate the intuitive prediction of each other 
behavior and shape subsequent interactions [8]. During social 
interaction, these signals, cues, and responses interplay in a 
continuous mutual exchange that complements our cognitive 
processes. For instance, pupil size seems to have a substantial 
effect on the way we perceive others [9], [10]: human faces with 
large pupils may be perceived as more attractive [11] and 
trustworthy [12]. Moreover, people seem to be more honest when 
interacting with a partner with large pupil size [10]. Like 
language, non-verbal communication varies significantly across 
cultures and contexts, adding layers of complexity to human 
interaction. 
When interacting through or with technological mediums, 
humans may (or may not!) apply the same heuristics they would 
use with other living beings [13]. For example, in previous work 
[14], we demonstrated that people mimic non-verbal cues to a 
similar extent in video calls as in face-to-face interactions. 
Although this study focused on select cues, it suggests that people 
might employ the same intuitive judgments when interacting via 
technology as they do in real life. Of course, while mimicking 
cues from a person on the other side of a video call is one thing, 
it’s another when the interaction partner is an artificial agent. This 
raises a fundamental question: how do humans interpret these 
cues when they come from a non-human? Artificial agents are 
designed with specific visual elements—eyes, faces—that 
convey certain impressions [15, 16], yet we understand little 
about how these design choices impact human behavior 
unconsciously, especially given the unfamiliarity of such 
interactions. Further, since non-verbal behaviors vary across 
cultures, it’s essential to investigate whether these cues, when 
applied to artificial agents, elicit culturally distinct responses. 
Building on these insights, our current studies aim to address 
three central research questions: RQ1. Are people more dishonest 
when interacting with artificial agents than with humans? RQ2. 
To what extent do unconscious cues (e.g., pupil size) shape 
interactions and perceptions of artificial agents as they do in 
human interactions? RQ3. How do these dynamics vary across 
cultures, revealing potential cultural sensitivities in human-
artificial agent interactions? 
979-8-3503-7893-1/25/$31.00 ©2025 IEEE
HRI Pioneer HRI 2025, March 4-6, 2025, Melbourne, Australia
1838
20
25
 2
0t
h 
AC
M
/I
EE
E 
In
te
rn
at
io
na
l C
on
fe
re
nc
e 
on
 H
um
an
-R
ob
ot
 In
te
ra
ct
io
n 
(H
RI
) |
 9
79
-8
-3
50
3-
78
93
-1
/2
5/
$3
1.
00
 ©
20
25
 IE
EE
 |
 D
O
I: 
10
.1
10
9/
HR
I6
15
00
.2
02
5.
10
97
40
55
Authorized licensed use limited to: Universiteit Leiden. Downloaded on June 25,2025 at 15:10:17 UTC from IEEE Xplore.  Restrictions apply. 
II. CURRENT WORK 
Our research sought to investigate whether individuals 
would exhibit greater dishonesty when interacting with artificial 
agents compared to humans, and if unconscious cues would 
influence this behavior. To this end, our current work employs a 
modified coin-toss game adapted from [10], which is well-known 
to elicit dishonest behaviors. Participants engage with three types 
of agents: a humanoid avatar on a screen, a machine-like robot, 
and a human confederate acting as a genuine participant. Across 
all agent types, we manipulate pupil size to be either dilated or 
constricted. For human confederates, this is achieved using 
customized contact lenses. Assessing pupil size over other signals 
has two main advantages: it is an unconscious cue beyond the 
participant's control, reducing potential bias; it can be 
consistently manipulated across diverse agent types while 
minimizing uncanny valley effects, which is crucial given the 
significant differences between our interaction agents. Besides, 
we continuously measured pupil size changes in the participants’ 
eyes. To ensure a diverse sample and environment, we recruited 
European participants both from a controlled laboratory (N=97) 
setting and in the wild from the general public (N=103). This 
approach allows us to gather data from a relatively heterogeneous 
sample, enhancing the validity and generalizability of our 
findings. Our preliminary results suggest that our participants 
show a clear difference in dishonesty between human and 
artificial agents, being more honest toward the human. Large 
pupils seem to amplify these differences, whereas small pupils 
tend to equalize the levels of dishonesty across agent types. 
Given that social cues like eye gaze and pupil size are 
culturally contextual, we replicated this study in Japan to examine 
potential cross-cultural differences. Previous literature shows that 
European and Japanese participants hold distinct explicit 
attitudes toward robots and avatars – with Japanese being more 
favorable to artificial agents compared to Europeans – but no 
difference was found at the implicit level [17]. These cultural 
nuances prompted us to replicate the study with a Japanese 
sample (N=75), who completed the same experiment as their 
European counterparts. Results, albeit preliminary, show that 
Japanese participants seem to behave equally dishonestly 
towards the human and artificial agents, with one notable 
exception: when playing with agents displaying constricted 
pupils, a slight difference emerged between avatars and human 
agents. Collectively, our findings suggest that subtle 
physiological cues like pupil size can influence dishonesty, with 
different effects across cultures and agent types. 
III. FUTURE WORK 
A. Network Analysis of Neuropsychophysiological Signals 
for Human-Centered AI 
Our current studies on the influence of cues on dishonesty 
highlighted a critical asymmetry in HRI research: while humans 
leverage social and emotional intelligence when assessing 
the trustworthiness of robot systems, current artificial agents do 
not (yet) possess the same sophisticated capabilities to fully 
interpret and respond to these non-verbal and often unconscious 
(dis)honesty signals. A key research question, then, is what 
honesty signals can be identified in human-robot interaction that 
would help AI systems respond meaningfully, and how do they 
interplay together (RQ4)? This question is central to our future 
work, where we will integrate methods from developmental 
psychology, social neuroscience, and human-robot interaction to 
perform a multi-modal network-based analysis. Sticking to our 
current approach, participants will engage in a real-life honesty 
game with a robot and with another human. We will continuously 
record facial expressions, heart rate, skin conductance, and pupil 
dilation, in combination with brain activity – measured through 
mobile functional near-infrared spectroscopy (fNIRS) over the 
temporoparietal junction. These signals, crucial for assessing 
emotional arousal [18], decision-making [19], and social 
cognition [20], have mostly been collected in isolation [21], with 
only limited insights gathered during real interaction with 
artificial agents. To better contextualize the observed behavior 
and physiological responses, we will also collect multiple self-
report measures from the participants. These will include 
attitudes and previous experience with AI, baseline honesty traits, 
reward sensitivity, and the level of anthropomorphism they 
ascribe to the agents. Our comparative network analysis approach 
will identify specific patterns and dynamics between the bio-
behavioral data (facial expressions, physiology, neural activity, 
self-report), showing similarities and differences in the 
interaction with robots and humans. 
B. Generational AI 
Our current cross-cultural research highlights the importance 
of individual differences to understand honesty towards artificial 
agents. While culture plays a role, generational differences are 
also a key. Dishonest behavior may vary across generations [22], 
as well as attitudes toward AI [10]: baby boomers may view 
robots as distant or fictional, millennials as an emerging reality, 
and Gen Z as part of daily life. We have no guarantee that an 
artificial agent developed and optimized for one group of users 
will work for users of different ages as well. How do these 
generational differences influence honesty towards artificial 
agents (RQ5)? To address this, we plan to invite individuals from 
different generations (20-, 40-, and 60-year-olds) to play with the 
robots or other humans. We will build network models for each 
age group offering similarities and differences in bio-behavioral 
signatures of dishonesty towards artificial agents and providing a 
direct and mechanistic input for human-robot interactions. Our 
project will explore how these cues vary across ages, revealing 
generational similarities and barriers, and facilitating artificial 
agent designs that accommodate a range of users, especially older 
adults at risk of digital exclusion. 
IV. TOWARDS REAL SOCIAL ROBOTS 
The interdisciplinary journey from understanding honesty in 
human-human interaction to decoding patterns of biobehavioral 
HRI Pioneer HRI 2025, March 4-6, 2025, Melbourne, Australia
1839
Authorized licensed use limited to: Universiteit Leiden. Downloaded on June 25,2025 at 15:10:17 UTC from IEEE Xplore.  Restrictions apply. 
signals in human-robot interaction put our research at the cutting 
edge of HRI. By merging social neuroscience, 
neuropsychophysiology, and comparative robotics, we will shed 
light on how honesty is perceived and reciprocated in human-
robot interactions. In addition, albeit limited, we have the 
opportunity to understand what circumstances trigger dishonest 
behavior in people. Integrating these insights into artificial 
systems could substantially improve human-robot collaboration, 
making future artificial agents more intuitive and human-
centered. 
Our holistic approach aligns with the broader conceptual 
framework proposed in our recent paper [23], where we call for 
a shift toward a “bottom-up” approach in social robotics that 
emphasizes spontaneous sociality emerging from robot-robot 
interactions. By exploring the fundamental, spontaneous 
mechanisms of social behaviors in robots, we can gain a more 
comprehensive understanding of the core principles that drive 
sociality across species and environments. This perspective 
complements our research focus by suggesting that 
understanding how fundamental behaviors emerge in robots 
could spotlight why certain low-level cues (e.g., pupil size) are 
used in complex social decisions. Through evolutionary robotics, 
we can experiment with and accelerate the conditions that foster 
these cues, potentially revealing how honesty indicators develop 
in both humans and artificial agents. 
ACKNOWLEDGMENT 
Past and present work is supported by ERC grant (804582 to 
M.E.K. and 101117045 to R.H.) and by the JSPS Short Term 
Fellowship awarded to F.D (PE22041). 
REFERENCES 
[1] E. Glikson and A. W. Woolley, “Human trust in artificial intelligence: 
Review of empirical research,” Acad. Manag. Ann., vol. 14, no. 2, pp. 
627–660, 2020, doi: 10.5465/annals.2018.0057. 
[2] C. Jacobsen, T. R. Fosgaard, and D. Pascual-Ezama, “Why Do We Lie? a 
Practical Guide To the Dishonesty Literature,” J. Econ. Surv., vol. 32, no. 
2, pp. 357–387, 2018, doi: 10.1111/joes.12204. 
[3] E. Prochazkova and M. E. Kret, “Connecting minds and sharing emotions 
through mimicry: A neurocognitive model of emotional contagion,” 
Neurosci. Biobehav. Rev., vol. 80, no. October 2016, pp. 99–114, 2017, 
doi: 10.1016/j.neubiorev.2017.05.013. 
[4] K. U. Likowski, A. Mühlberger, A. B. M. Gerdes, M. J. Wieser, P. Pauli, 
and P. Weyers, “Facial mimicry and the mirror neuron system: 
simultaneous acquisition of facial electromyography and functional 
magnetic resonance imaging,” Front. Hum. Neurosci., vol. 6, no. July, pp. 
1–10, 2012, doi: 10.3389/fnhum.2012.00214. 
[5] B. Tia, A. Saimpont, C. Paizis, F. Mourey, L. Fadiga, and T. Pozzo, “Does 
observation of postural imbalance induce a postural reaction?,” PLoS One, 
vol. 6, no. 3, 2011, doi: 10.1371/journal.pone.0017799. 
[6] M. E. Kret, “Emotional expressions beyond facial muscle actions. A call 
for studying autonomic signals and their impact on social perception,” 
Front. Psychol., vol. 6, no. May, pp. 1–10, 2015, doi: 
10.3389/fpsyg.2015.00711. 
[7] R. V. Palumbo et al., “Interpersonal Autonomic Physiology: A Systematic 
Review of the Literature,” Personal. Soc. Psychol. Rev., vol. 21, no. 2, pp. 
99–141, 2017, doi: 10.1177/1088868316628405. 
[8] R. T. Boone and R. Buck, “Emotional Expressivity and Trustworthiness: 
The Role of Nonverbal Behavior in the Evolution of Cooperation,” J. 
Nonverbal Behav., vol. 27, no. 3, pp. 163–182, 2003. 
[9] M. E. Kret, “The role of pupil size in communication. Is there room for 
learning?,” Cogn. Emot., vol. 32, no. 5, pp. 1139–1145, 2018, doi: 
10.1080/02699931.2017.1370417. 
[10] J. A. van Breen, C. K. W. De Dreu, and M. E. Kret, “Pupil to pupil: The 
effect of a partner’s pupil size on (dis)honest behavior,” J. Exp. Soc. 
Psychol., vol. 74, no. November 2017, pp. 231–245, 2018, doi: 
10.1016/j.jesp.2017.09.009. 
[11] S. Tombs and I. Silverman, “Pupillometry - A sexual selection approach,” 
Evol. Hum. Behav., vol. 25, no. 4, pp. 221–228, 2004, doi: 
10.1016/j.evolhumbehav.2004.05.001. 
[12] M. E. Kret and C. K. W. De Dreu, “Pupil-mimicry conditions trust in 
partners: Moderation by oxytocin and group membership,” Proc. R. Soc. 
B Biol. Sci., vol. 284, no. 1850, pp. 1–10, 2017, doi: 
10.1098/rspb.2016.2554. 
[13] K. J. Kim, “Heuristics in digital communication media: theoretical 
explications and empirical observations,” Qual. Quant., vol. 49, no. 5, pp. 
2187–2201, 2015, doi: 10.1007/s11135-014-0103-y. 
[14] F. Diana, O. E. Juárez-mora, W. Boekel, R. Hortensius, and M. E. Kret, 
“How video calls affect mimicry and trust during interactions,” Phil. 
Trans. R. Soc. B, vol. 378, 2023. 
[15] S. Woods, “Exploring the design space of robots: Children’s 
perspectives,” Interact. Stud., vol. 18, pp. 1390–1418, 2006, doi: 
10.1016/j.intcom.2006.05.001. 
[16] E. Phillips, D. Ullman, M. M. A. De Graaf, and B. F. Malle, “What does 
a robot look like?: A multi-site examination of user expectations about 
robot appearance,” Proc. Hum. Factors Ergon. Soc., vol. 2017-Octob, pp. 
1215–1219, 2017, doi: 10.1177/1541931213601786. 
[17] F. Diana, M. Kawahara, I. Saccardi, R. Hortensius, A. Tanaka, and M. E. 
Kret, “A Cross-Cultural Comparison on Implicit and Explicit Attitudes 
Towards Artificial Agents,” Int. J. Soc. Robot., vol. 15, no. 8, pp. 1439– 
1455, 2023, doi: 10.1007/s12369-022-00917-7. 
[18] H. A. Shehu, M. Oxner, W. N. Browne, and H. Eisenbarth, “Prediction of 
moment-by-moment heart rate and skin conductance changes in the 
context of varying emotional arousal,” Psychophysiology, vol. 60, no. 9, 
pp. 1–15, 2023, doi: 10.1111/psyp.14303. 
[19] G. Forte, M. Morelli, B. Grässler, and M. Casagrande, “Decision making 
and heart rate variability: A systematic review,” Appl. Cogn. Psychol., 
vol. 36, no. 1, pp. 100–110, 2022, doi: 10.1002/acp.3901. 
[20] M. Ahmad and A. Alzahrani, “Crucial Clues: Investigating 
Psychophysiological Behaviors for Measuring Trust in Human-Robot 
Interaction,” ACM Int. Conf. Proceeding Ser., pp. 135–143, 2023, doi: 
10.1145/3577190.3614148. 
[21] I. Ben Ajenaghughrure, S. Da, C. Sousa, and D. Lamas, “Measuring Trust 
with Psychophysiological Signals : A Systematic Mapping Study of 
Approaches Used,” Multimodal Technol. Interact., vol. 4, no. 63, pp. 1– 
29, 2020, doi: 10.3390/mti4030063. 
[22] A. M. O. Connor, R. A. Judges, K. Lee, and A. D. Evans, “Examining 
honesty – humility and cheating behaviors across younger and older 
adults,” Int. J. Behav. Dev., vol. 46, no. 2, pp. 112–117, 2022, doi: 
10.1177/01650254211039022. 
[23] F. Diana, L. Cañamero, R. Hortensius, and M. E. Kret, “Merging sociality 
and robotics through an evolutionary perspective,” Sci. Robot., vol. 9, no. 
eadk6664, 2024, doi: 10.1126/scirobotics.adk6664. 
HRI Pioneer HRI 2025, March 4-6, 2025, Melbourne, Australia
1840
Authorized licensed use limited to: Universiteit Leiden. Downloaded on June 25,2025 at 15:10:17 UTC from IEEE Xplore.  Restrictions apply. 


Can robots trust us?: Neuropsychophysiological insights into honesty towards artificial agents

Fasya, E.L.

Bos, E. van den

Heylen, D.K.J.

Kret, M.E.

Eye contact during virtual storytelling: enhancing social evaluation through pupil mimicry

Piitulainen, Milla

Weaving Through: Digital Cowpaths and Placing the Web

Condro Lukitosari, Sonya

Identifying Sounds: Comparing interpretations of Sonic Histories of East Javanese Ludruk Folk Theater Performances

Krogull, A.

Rutten, G.J.

Multilingual writing practices as code choices: Dutch alongside French in private family letters

Smits, I.B.

Craig, C.

Kopylova, O.

Muzan ya na: the trauma of war and traditional Japanese poetry

Seventh Framework Programme (FP7)101056868Industrial Ecolog

Napiontek, J.

Fishman, T.

Pichler, P.-P.

Heintz, J.

Weisz, H.

Live (a) little: GHG emissions from residential building materials for all 400 counties and cities of Germany until 2050

Cremers, Y.E.M.

Ongerechtvaardigde verrijking ten gevolge van natrekking

Bruning, M.R.

Asdonk, S. van der

Smeets, D.J.H.

Boven, H.J. van

Buisman, R.S.M.

Lenglet, M.J.E.

Alink, L.R.A.

Terugplaatsing na gedwongen uithuisplaatsing: een blik op de uitvoeringspraktijk

Not a member yet