One of the biggest threats and disruptions to privacy and democracy in this digital age is deepfake technology. A ‘deepfake’ or synthetic media, is a video editing technology that manipulates and mimic a person’s facial expressions, mannerisms, voice, and inflections based on a large amount of data of other people to create a hyper-realistic video depicting them doing or saying things that never happened ( Westerlund, 2019).

The current consensus is that the average human’s ability in recognizing deepfakes is similar to the machines ( Vitak, 2022). However, the result seems to vary depending on their own confidence and belief in their cognitive abilities. Some studies suggest that some individual differences determine if a person is good at recognizing deepfakes or not ( Shahid et al., 2022). In this study, we will look at the relationship between personality traits and people’s self-reported efficacy in recognizing deepfakes.

The HEXACO personality model describes six facets of personality structures: Honesty-humility, Emotionality, Extraversion, Agreeableness, Conscientiousness, and Openness to experience ( Lee & Ashton, 2009; unpublished report; Zettler et al., 2020). This instrument model is selected because of three reasons: (1) the measurement covers a wider and more complex range of personality facets that go beyond the five-factor model ( Ashton & Lee, 2007); (2) the Honesty-Humility factor measures traits like sincerity, boastfulness, pretentiousness, and fair-mindedness that are associated with dishonest or inauthentic behaviors ( Ashton & Lee, 2008) relating to self-reported efficacy; and (3) the model provides flexibility in measuring contextually unique situations ( Oostrom et al., 2019; Pletzer et al., 2020). Advancements in information technology, including AI, socially intelligent robots, and other autonomous systems, will have a profound impact on human life, necessitating research in typical personality to understand and address individual differences in adapting to these new challenges ( Matthews et al., 2021), not to mention deepfakes. Multiple studies in various contexts have shown that personality traits influence an individual’s self-efficacy ( Lodewyk, 2018).

The Honesty-humility dimension reflects an individual’s fair-mindedness, modesty, and cooperation. A person with high Honesty-humility might not think they are good at recognizing deepfakes, regardless of their true ability while an individual with low Honesty-humility might be biased in their self-reported ability in recognizing a deepfake.

Emotionality reflects an individual’s degree of anxiousness, fearfulness, and sentimentality - the experience of anxiety in response to life’s stressors. To overcome this anxiety, the sense of being able to recognize deepfakes is important to reduce that anxiety. One way to become less anxious is to appreciate deepfakes as a “cultural technology” ( Cover, 2022) that contains artistic and creative values. People with high Emotionality may be more motivated to use deepfakes as an “antidote” from the pressures of everyday life, so they have higher self-reported efficacy to detect them, not to be avoided but as potential things to be used according to their interests (technology appropriation; see Prayoga & Abraham, 2017)

Extraversion reflects an individual’s degree of sociability. Individuals high in Extraversion might have higher self-efficacy due to their higher social esteem, boldness, and familiarity. Van der Zee et al. (2002) found that extroverts are friendly and less formal in their interactions with others. This is closely connected with emotion recognition (part of emotional intelligence) which affects the success of negotiations. By using the paradigm of the social construction of technology ( Kwok & Koh, 2021), humans are parties who “negotiate” with technology to better recognize the technology, including deepfakes, and can adapt it to not become victims of technology—or misappropriate technology for evil interests—but rather agents who utilize technology to improve humanity and prevent harm posed by technology (such as deepfakes).

An individual’s degree of cooperation, tolerance, flexibility, and patience is reflected in the Agreeableness dimension. More agreeable people are at a larger risk for security, and social engineers (like deepfake designers) specifically target Agreeableness attributes like benevolence and compliance.

Conscientiousness reflects precisions, cautiousness, and a degree of self-control. Individuals with higher Conscientiousness thread might have higher self-efficacy in recognizing deepfakes. This is in line with the hypothesis of Köbis et al. (2021) that increasing Conscientiousness will make people motivated to invest cognitive resources to detect deepfakes, thereby enhancing their capacity to recognize truth and decreasing their desire to spread false information.

Openness reflects the willingness to experience new things and is associated with lower risk aversion. Research by Uebelacker and Quiel (2014) shows that open people don’t create suitable coping mechanisms because they misjudge their vulnerability to being a target of social engineering (like deepfake designers).

This confirmatory study tested the hypotheses that the dimensions of HEXACO personality traits, i.e. (1) Honesty-humility, (2) Emotionality, (3) Extraversion, (4) Agreeableness, (5) Conscientiousness, and (6) Openness can predict self-reported efficacy in recognizing deepfakes.

There is only one data collection stage. There is no exposure in this study because the research was not an experimental study.

Demographically, some participants were residents of DKI Jakarta province ( N=90) which is the capital of Indonesia. In addition, other participants were residents of the Java Island (non-DKI Jakarta; N=86); Sumatera Island ( N=21); and the rest ( N=3) came from East Kalimantan, North Maluku, and West Nusa Tenggara provinces.

The psychometric properties and descriptive statistics of the variables are shown in Table 1. The results of this study indicate that the residuals of the multiple regression model are normally distributed ( Figure 1) and all HEXACO personality dimensions are negatively correlated with self-efficacy in recognizing deepfakes; except for Agreeableness, which positively correlated (see Table 2). However, the results of the regression analysis with F(6,199)=13,295, p=0.000, R ²=0.292, showed that only Honesty-humility and Agreeableness were able to predict the efficacy (see Table 3). No difference was found between women and men, t(198)=−0.120, p=0.904, Cohen’s d=0.018, SE Cohen’s d=0.154, in terms of self-efficacy.

Table 3 shows the unadjusted ( B) and adjusted ( β) estimates for each predictor of which the potential confounders are the personality traits dimensions other than the focused predictor.

Recognizing all deep fakes elements requires a certain level analytical capability and general intelligence ( Ahmed, 2021). We need to look not just at people’s cognitive abilities, but also at their belief in carrying out these abilities to recognize the information of deep fakes contextually. In other words, their self-efficacy.

This study found that, to a certain degree, individuals’ personality traits do affect their self-efficacy in terms of detecting deepfakes. Because self-efficacy expression depends on context-to-context, it is not surprising that some traits can predict it better than the others.

Personality trait of Honesty-humility had negative predictive correlation with self-efficacy in recognizing deepfakes, β=-0.255, t(193)=-2.491, p<0.05 ( Table 3). “Persons with very high scores on the Honesty-Humility scale avoid manipulating others for personal gain, feel little temptation to break rules, are uninterested in lavish wealth and luxuries, and feel no special entitlement to elevated social status” (Lee & Ashton, 2009, para 1). A person’s Honesty-humility trait do not want to engineer others but, ironically, this trait makes them vulnerable to being manipulated by others ( Ternovski et al., 2021), including deepfakes. It can drive higher errors for the trait in recognizing deepfakes, exposing weaknesses that could be exploited.

Thompson et al. (2016, p. 54) once stated, “Honesty-Humility may not only be less likely to exploit others, they may also be strongly opposed to being the target of exploitation.” Nevertheless, this present study shows that in the presence of deepfakes technology that has a high possibility to manipulate someone, Generation Z with trait Honesty-Humility feels helpless, so it is less functional in detecting deepfakes.

That is a notable discovery of this present study, and could be explained by the findings of Weger et al. (2022) that Honesty-Humility has a negative correlation with general ( r=-0.168, p=0.002) and specific ( r=-.0270, p<0.001) technology acceptance. This is reinforced by the findings of Sindermann et al. (2020) that Honesty-Humility has a negative correlation with all aspects of technology acceptance, namely perceived usefulness ( r=-0.25, p<0.001), perceived ease of use ( r=-0.16, p<0.001), intention to use ( r=-0.17, p<0.001), and predicted usage ( r=-0.18, p<0.001). In fact, someone with high technology affinity is able to perceive deepfakes less negatively ( Kleine, 2022). This is presumably because they feel they have knowledge and “master” deepfakes.

Therefore, to not fall for deepfakes, Generation Z with a high Honesty-Humility trait need to reduce their conservative attitude towards technology in order to detect potential harm and even utilize deepfakes effectively. Future studies can test this with an experimental design that involves measuring these two traits and people’s ability to detect malicious vs. non-malicious deepfakes videos.

Emotionality cannot predict self-efficacy in recognizing deepfakes, β=-0.029, t(193)=0.332, p>0.05 ( Table 3). Austin and Vahle (2016) found that Emotionality—a trait that is positively correlated with empathy and social engagement—can predict the dimensions of Enhance (providing support and reassurance as interpersonal emotion management strategies) and Divert (the practice of using humor and pleasure pursuits to lift the spirits of others) of the Managing the Emotion of Others Scale (MEOS). This means that the Emotionality dimension is also positively correlated with the emotional intelligence needed to recognize deepfakes. Yang et al. (2022) emphasized the pivotal role of emotional intelligence in improving artificial intelligence technology so that it becomes a useful deepfake in the context of clinical encounters. By knowing that deepfakes themselves are increasingly being prepared with elements of emotional intelligence, then recognizing deepfakes also requires a better one; and this intelligence can actually be found in people with higher Emotionality. However, individuals high in Emotionality might be less confident in their own ability to accurately recognize deepfakes, as they might consider more factors and doubt themselves more ( Thompson, 1998). With this uncertain direction, it is not surprising that no predictive power of Emotionality was found on self-efficacy.

Extraversion is a personality trait that cannot predict self-efficacy in recognizing deepfakes, β=0.003, t(193)=0.044, p>0.05 ( Table 3). Hosler et al. (2021) put forward that detecting deepfakes is actually recognizing unnatural displays of emotion in voices and faces. Emotion apparently plays a central role in recognizing deepfakes because emotion is a higher-level semantic construct—which is difficult to counterfeit up to now—that could offer hints for detection. In an unpublished report, Kill states that emotion recognition is an ability that is honed in someone with a high extraversion trait ( 2021). However, Extraversion is also found to be positively correlated with excitement-seeking and a lower preference for consistency ( Uebelacker & Quiel, 2014) - whereas “pairwise self-consistency learning” ( Zhao et al., 2021, p. 15023) is needed to recognize deepfakes. Therefore, the effects of Extraversion traits appear to cancel out of each other resulting in no predictive correlation with the self-efficacy.

Agreeableness trait can predict self-efficacy in recognizing deepfakes; however, not as hypothesized, the direction was found positive – not negative, β=0.309, t(193)=4.090, p<0.05 ( Table 3). People with high Agreeableness are eager to cooperate and reach a compromise with others (Lee & Ashton, 2009). One of the good “others” in the context of deepfake recognition or detection is the “wisdom of the crowds” ( Groh et al., 2022), which Surowiecki (2004) defines as “the collective intelligence that arises when our imperfect judgments are aggregated”. Agreeing with (or high Agreeableness to) the collective intelligence should reduce the chance of falsely recognizing deepfakes, including its algorithm attempts that present visual obstructions such as misalignment, partial occlusion, and inversion.

Agreeableness trait has a positive correlation with perception of forensic science ( Sarki & Mat Saat, 2020). Deepfakes detection can be seen as part of forensic science. People with high agreeableness are known for their cooperativeness; agreeableness is often referred to as safeguards against antisocial behavior ( Frias Armenta & Corral-Frías, 2021), including - in the context of this present study - deepfakes creation and distribution. They esteem innovative forensic methods in their environment and have a positive attitude toward it for the common good ( Sarki & Mat Saat, 2020).

People who are more agreeable tend to make more accurate decisions about whether to believe information, which reduces their vulnerability to victimization ( Cho et al., 2016). This is confirmed by the empirical findings of van Winsen (2020) that agreeable individuals exhibit more secure online behavior and have a lesser likelihood of becoming a victim of cybercrime.

This study found that Conscientiousness was not able to predict self-efficacy in recognizing deepfakes, β=-0.079, t(193)=-0.798, p>0.05 ( Table 3). Although deepfake recognition requires conscientious characteristics such as prudence and a sense of responsibility, Lawson and Kakkar’s (as cited in Sütterlin et al., 2022) research recently found that Conscientiousness is partially correlated with belief in conspiracy and conservatism - making it less efficacious in recognizing deepfakes.

This study found that Openness was not able to predict self-efficacy in recognizing deepfakes β=-0.028, t(193)=-0.312, p>0.05 ( Table 3). In an unpublished report, Jin (2020) found that values of Openness to change do not correlate with the perceived ethical implications of deepfakes ( e.g., “These videos can uncontrollably deceive and influence many people”, p. 24). In addition, contrast with the certain direction of the influence of Agreeableness and Honesty-humility on the self-efficacy; the direction of the Openness prediction is ambiguous. On the one hand, Openness is related to the low ability to recognize deepfakes. It is because Openness was found to be positively correlated with cognitive ability ( Curtis et al., 2015; Rammstedt et al., 2016), but cognitive abilities encourage more protective online behavior, indicated by more interest in discussing how people who use deepfakes manipulate their audiences - rather than developing ability to apply scepticism on the authenticity of videos ( Ahmed, 2021). On the other hand, there is a logic in favor of Openness as a buffer to prevent vulnerabilities from being manipulated by social engineering. For example, Eftimie et al. (2022) associated Openness with cognitive exploration tendencies which, based on their study, will stimulate responsible behavior including security best practices - which in the context of this study is deepfake recognition.

Based on the study findings, to avoid falling for deep fakes, there are two “optimal” personality traits that are worth exercising, i.e. Honesty-Humility and Agreeableness. First, the Honesty-Humility trait needs to be positioned strategically so that people with this trait can not easily be trapped or “absorbed” by the counterfeits from deepfakes technology, ie by reducing conventionalism ( Leone et al., 2012) towards technology, that is allegedly inherent in this trait. Second, agreeableness trait should be directed at various deepfake detection methods and technologies that are beneficial to community members.

A number of studies have shown that both general and technological self-efficacy are able to predict the actual ability associated with the use of the technology ( Alnoor et al., 2020; Raghuram et al., 2003; Tetri & Juujärvi, 2022). This is because the efficacies determine organizing actions, behavioral intention and strategies, and preparedness for change, as well as reducing emotional sensitivity which is a source of performance anxiety.

Of course, there is no denying the possibility of inflated or overestimated belief, or the Dunning-Kruger effect ( Koc et al., 2022), which in the context of this study means that people who have high self-efficacy in detecting deepfakes actually have low actual abilities. In their research on bullshit detection, Cavojová et al. (2022) explained that the overestimation is caused by metacognitive (un) awareness, i.e. “These highly overconfident people suffer from a double curse – not only they do not know, but they also do not know that they do not know ... [that] is the result of self-enhancement motivation” (p. 1, 2).

The limitation of this research is the use of non-probability sampling with limited generalizability. Nevertheless, this study has implication for the development of psychoinformatics - a branch of psychology that explains attitudes, competencies, and behavior in using information technology. Further research is suggested to implement random sampling and experimental methods to ensure a causal–not only predictive–relationship between personality traits and deepfakes detection self-efficacy.