1 Introduction

Artificial intelligence is built into a growing number of products, with which humans interact on a daily basis. Such products are language assistants in smartphones, self-driving cars and social robots. The AI revolution is also discussed in other areas such as sales platforms (Ravindar et al. 2022). In particular with the release of the generative artificial intelligence ChatGPT (Lund & Wang 2023), artificial intelligence represents a hotly debated topic in the mass media (The Economist 2022) being accompanied also by fears that artificial intelligence will have negative impacts on the world for instance resulting in job loss (Chelliah 2017). Others are more optimistic about the chances that will arise from artificial intelligence, for instance, aiding humans by taking over repetitious tasks and some researchers think about the benefits of “human–machine coexistence” (Hamid et al. 2017). For discussions around the automation term in the context of artificial intelligence research see a recent work (Montag et al. 2023). As artificial intelligence will likely also become a driver of the economy (Aghion et al. 2017), societies with more optimistic views might more profit from AI as early adopters of this technology. See also recent work in the area of medical disciplines (Pinto dos Santos et al. 2019).

Until now, only a few studies have investigated which factors might influence individual differences in attitudes towards AI. A recent study not only presented a short screening tool to assess positive and negative attitudes towards AI (called ATAI or ATAI framework), but also observed that females reported on average lower positive attitudes towards AI than males—in particular in the German sample described, and less so in the Chinese sample (Sindermann et al. 2021). However, gender differences in negative attitudes were less visible in this work. Pronounced differences could be also observed for culture effects in the same work, with Chinese participants reporting on average higher positive attitudes towards AI than German participants (accepting AI dimension).Differences between negative attitudes towards AI were less pronounced between Chinese and German study participants (fearing AI dimension). The investigation of the personality variables fit also within a recently proposed framework, which has been put forward to understand how attitudes towards AI might be shaped. The so-called IMPACT model proposes that an Interplay of Modality, Person, Area, Country/Culture and Transparency variables are relevant to understand how attitudes toward AI form (Montag et al., 2024a; Montag et al., 2024b). Personality belongs to the P-variable. Of interest for the present study are observations that personality seems to play a role to understand individual differences in attitudes towards AI, but effect sizes are small (Sindermann et al. 2022): consistently it was observed that the only Big Five personality trait linked to negative attitudes towards AI was neuroticism. Here, higher scores of neuroticism were linked to higher negative attitudes towards AI in both Germany and China (Sindermann et al. 2022). Neuroticism represents a super-dimension comprising many facets such as being anxious and tending to be more in a depressed state (Zhang 2020). To better understand what facets of neuroticism might be driving negative attitudes towards AI, we conducted the present study not focusing on the Big Five personality traits to understand individual differences in attitudes towards AI, but instead on an evolutionary approach from personality psychology: Whereas the Big Five have been derived from a lexical approach (Montag and Elhai 2019; Piedmont and Aycock 2007), Jaak Panksepp carved out seven primary emotional systems by means of electrical stimulation studies, lesion and pharmacological studies being homologously conserved across the mammalian brain (Panksepp 1998, 2011). These primary emotional systems represent tools for survival and are called SEEKING, CARE, PLAY, LUST (positive emotions) and FEAR, ANGER, SADNESS (negative emotions). According to Panksepp’s Affective Neuroscience Theory (ANT), these primary emotional systems are anchored in subcortical regions of the brain (Davis and Montag 2019). Individual differences in the neural circuits underlying these primary emotional systems in terms of structure and function underly individual differences in human personality. Information about the exact brain regions involved according to Pankseppian Affective Neuroscience Theory can be found in older works (Montag and Panksepp 2016; Panksepp 2011).

Of interest, for the present study, both the Big Five concept and primary emotional systems can be brought together. A meta-analysis provided support for the idea that high SEEKING might be the bottom-up driver of high Openness to Experience, high PLAY the bottom-up driver of high Extraversion, high CARE / low ANGER the bottom-up driver of Agreeableness and for the present study of most interest high FEAR, SADNESS, and ANGER the bottom up driver of neuroticism (Marengo et al. 2021). Bottom-up driver means that anatomically speaking the primary emotional systems—according to ANT being subcortically located—influence the lexically derived Big Five of Personality from evolutionary ancient brain regions (for further discussions see Montag and Davis ( 2018)). The emotion FEAR is of evolutionary significance because activation of the FEAR circuitry helps humans to get out of a danger zone. SADNESS is activated by separation distress and finally ANGER activity can be observed in situations of territorial conflicts in the mammalian kingdom and also frustrations, to name a few (Montag and Panksepp 2017; Panksepp 1998). To better understand the link between negative attitudes towards AI and neuroticism, we applied Panksepp’s Affective Neuroscience Theory. Due to the link between FEAR, SADNESS, ANGER and neuroticism (Davis et al. 2003; Montag and Panksepp 2017), we expected these three primary emotional systems to be positively associated with negative attitudes towards AI. Against the small effect sizes (neuroticism-negative attitudes towards AI/fearing AI) observed in the recent work, we also expected only small associations in the present work. Correlations are presented in the present work for all primary emotional systems and ATAI dimensions. Correlations going beyond our hypothesis should be seen as exploratory.

2 Methods

2.1 Participants

A total of 402 participants could be recruited via the website https://www.anps-research.com for the present study. We only included participants who stated either to be a native English speaker or, otherwise, competent in English. n = 17 participants stated to be non-native speakers and not to be competent and were therefore excluded from the analysis. We asked participants to only fill in the survey when being at least eighteen years old. 16 participants were younger than that age and were therefore also excluded. This resulted in n = 369 participants. Further, we checked for monotonous answer behavior regarding the ANPS and the ATAI scale (separately). One can debate if data cleaning on five items of the ATAI is meaningful, in particular with those who entered five items a neutral response (which is imaginable to stay completely neutral on the AI topic). For reasons of uniformity in our strategy, we decided to exclude those participants, too (this explains the light dip in the middle of the “normal distribution” of the ATAI depicted in Fig. 1). Correlations do not change meaningfully though with including and excluding those participants. Following our cleaning strategy, in total 18 participants showed monotonous answer behavior on either the ANPS and/or the ATAI and were excluded. The final sample consisted of 351 participants (n = 142 males and n = 209 females; mean-age: 35.1 (SD = 13.3)). Education level was as follows (highest degree received): less than high school (0.6%), high school graduate (9.4%), some college (16.5%), bachelor’s degree (30.8%), master’s degree (31.1%), doctoral degree (11.7%). Numbers do not add up to 100% due to rounding errors. Both the larger data set (n = 369) and the here presented smaller data set (n = 351) can be found at the Open Science Framework for further analysis of interested scientists to independently check on differences in correlations between the larger and the smaller data set: https://osf.io/9zf28/.

Fig. 1
figure 1

Distribution of scores of the positive and negative attitude towards artificial intelligence scales (please note that on the x-axis total sum scores are presented; histograms were produced with the SPSS package)

Full size image

All participants provided informed e-consent and agreed that the data could be shared in an anonymous way with the scientific community to foster open science practices. Please note that IRB approval was not deemed necessary, because data collection was completely anonymous, and participants could stop filling in the survey at all times (before finally submitting their data). As an incentive, all participants were provided with insights on their primary emotional systems scores at the end of the survey.

2.2 Questionnaires

Participants filled in the English version of the Affective Neuroscience Personality Scales (ANPS 3.1) which consists of 112 items (Montag et al. 2021). From these 112 items, each 14 items assess one primary emotional system. The only primary emotional system not assessed is the LUST system because answering items on one’s sexual activity might in particular trigger biased responses. Each item was answered on a six Likert scale ranging from strongly disagree (1) to strongly agree (6). Please note that some items need to be recorded before building the scales. Internal consistencies of the primary emotional systems were excellent with SEEKING (α = 0.79), PLAY (α = 0.85), CARE (α = 0.82) and FEAR (α = 0.92), SADNESS (α = 0.83) and ANGER (α = 0.87). In the ANPS also a spirituality dimension (assessed with 12 items) can be found, which was included against the background of spirituality being a relevant variable in the treatment of addiction. The internal consistency of the spirituality scale was also excellent (α = 0.93).

Aside from the ANPS 3.1. all participants filled in the English version of the Attitudes towards Artificial Intelligence Scale (ATAI) (Sindermann et al. 2021). This scale consists of five items with two items assessing positive attitudes towards AI (accepting AI) and three items assessing negative attitudes towards AI (fearing AI). In the present work, the ATAI items were answered with a five-point Likert scale ranging from strongly disagree (1) to strongly agree (5). Internal consistencies for the ATAI were as follows: negative attitudes towards AI (α = 0.62) and positive attitudes towards AI (α = .62). Please note that the ATAI consists of few items. Therefore, also in light of the external validity provided by Sindermann et al. (2021), the internal consistencies were deemed to be adequate.

2.3 Statistical analyses

Both the ANPS scales and ATAI dimensions were visually inspected and resembled rather normal distributions (see also Fig. 1). Jamovi 2.4.8.0 was used to produce descriptive statistics and run t-tests. Descriptive statistics are presented for all relevant variables for the total sample and male/females subsamples. We decided to run parametric tests. In a first simple analysis, two-tailed Pearson correlations are presented. These were also produced with the Jamovi package. In the next step, we also investigated in a stepwise regression model what represents the best predictor variables for negative attitudes towards AI. This was done with SPSS 29.0.1.0 (same was true for the education analysis and production of histograms). We decided upon an atheoretical stepwise approach to understand what variables are most critical in our data set to understand negative attitudes towards AI. No regression model is presented for predicting positive attitudes towards AI because primary emotional systems played a negligible role here.

3 Results

In Table 1 descriptive statistics for all relevant psychological variables are presented. Please note that not only descriptive statistics for the total sample but also for the male and female subsamples are presented (see Table 2). Females scored significantly higher on the CARE, FEAR and SADNESS of the ANPS. Females were associated with more negative attitudes towards AI compared to males. Females were also associated with lower positive attitudes towards AI compared to males. The ATAI differences failed to be significant though (see Table 3). Of interest, education levels were not significantly associated with negative attitudes towards AI (F(5,345) = 0.658, p = 0.655, eta2 = 0.009) nor with positive attitudes towards AI (F(5,345) = 1.188, p = 0.314, eta2 = 0.017). Correlations for the associations between psychological variables of interest are presented in Table 4.

Table 1 Descriptive statistics for all relevant psychological variables
Full size table
Table 2 Descriptive statistics for all relevant psychological variables for the male and female subsamples
Full size table
Table 3 Independent T-Test analysis regarding gender and the scales of interest
Full size table
Table 4 Correlations between primary emotional systems and attitudes towards artificial intelligence (n = 351); further correlations are reported with spirituality and also between both ATAI dimensions
Full size table

Based on the correlational analysis in Table 5 we also provide a deeper analysis of the negative primary emotional systems and ATAI scores on the item level.

Table 5 Correlations between primary emotional systems and single items of the negative attitude towards artificial intelligence scale
Full size table

A stepwise regression model was conducted with negative attitudes towards artificial intelligence representing the dependent variable. Independent variables were the negative primary emotional systems (FEAR, SADNESS, ANGER) and the spirituality dimension all entered into one block. The regression model explaining most variance (5.5%) included the SADNESS and spirituality variables only (second proposed regression model: F(2,348) = 10,038, p < 0.001, R2 = 0.055; adjusted R2 = 0.049). SADNESS alone explained 3.7% of the variance (first regression model: F(1,349) = 13,507, p < 0.001; R2 = 0.037; adjusted R2 = 0.035), adding spirituality led to an increment of 1.8% explained variance.

4 Discussion

Recent research showed that neuroticism is positively associated with negative attitudes towards artificial intelligence (Sindermann et al. 2022). As affective neuroscience theory suggests that the primary emotional systems FEAR, SADNESS and ANGER are neuroanatomically bottom-up drivers of neuroticism (Marengo et al. 2021), the present work aimed to shed light on associations between these primary emotional systems and negative attitudes towards artificial intelligence.

In line with our expectations, positive correlations appeared between all negative primary emotional systems and negative attitudes towards artificial intelligence. The highest association with negative attitudes towards AI was observed for the SADNESS system, which was also backed up by our stepwise regression model. According to Jaak Panksepp the SADNESS system is triggered by separation distress (Davis and Montag 2023; Watt 2023; Watt and Panksepp 2009), which from an evolutionary point of view, represents a dangerous state for a human, because young children separated from caregivers are especially vulnerable, but in general our species is stronger in groups than alone. Eliciting psychic pain (being sad) when facing a loss of a significant person might result in searching for social support (being taken CARE of) to reduce the activity of the SADNESS system as well as contributing to social group cohesion. From this perspective, one might also think of designing AI with having in mind that if AI fulfilled human social needs, it might reduce negative attitudes towards this technology.

Interestingly a higher active SADNESS system (in terms of an emotional personality trait) was robustly associated with all single items assessing negative attitudes towards artificial intelligence including fearing AI, believing that artificial intelligence will destroy human mankind and AI being a driver of job loss. Our data suggest that SAD persons might be more drawn into the mentioned scenarios/items, with the highest correlation found on item level, namely, “AI destroying human mankind”. Our data speak for the idea that SADNESS underlying a neurotic personality trait might be the facet of neuroticism being most strongly linked to negative attitudes towards AI. Our data could also mean that being in a stable relationship with persons might buffer against negative attitudes towards AI.

Beyond this investigation, we also observed an interesting finding, namely that spirituality was significantly associated with both negative and positive attitudes towards artificial intelligence. In detail, higher spirituality went along with more negative and less positive views on artificial intelligence. Persons scoring high on spirituality report among others to be “touched by the beauty of creation” and spirituality “being a primary source of inner peace” (see page 165: Montag et al. 2021). Being a spiritual person thereby goes along with being more skeptical towards AI. This result was not hypothesized and therefore it is just a starting point to be further explored. Please note that the finding in the present work would also not hold for adjustments for many correlations such as Bonferroni corrections.

The present study comes with several limitations. First of all, this study is of cross-sectional nature, and we cannot disentangle cause and effect. Second, this study is based on self-report and therefore could be influenced by a lack of introspective abilities, etc. This is also important to stress because the neuroscientific assessment of activity in the brain circuitries underlying primary emotional systems would make an important add to the present investigation. In this realm, it needs also to be mentioned that further unraveling of the brain mechanisms underlying cognitive functions might stimulate progress in AI developments and perhaps AI democratization (Bain and McCay 2023). This is also true when one considers the brain’s efficiency in terms of energy needed to solve tasks compared to current AI systems (Fuller 2019). A third limitation: the study sample is not representative and included people from different countries. We did not assess country or culture, just being a native English speaker and being proficient in the English language, when not being a native speaker. This said, the data quality of the present work looks good because gender differences in the ANPS are as often reported in the literature (Özkarar-Gradwohl and Turnbull 2021): For instance, we also observed higher CARE and SADNESS scores in females compared to males, which fits with observations in other samples. Internal consistencies were mostly in a respectable area. Further, we observed again the trend that males had more positive views towards AI than females (Sindermann et al. 2021).

Further limitations: Recent work observed some cultural effects on the ATAI (Sindermann et al. 2021, 2022)—this could not be investigated in the present sample. Finally, the here observed effect sizes and explained variance of the regression model are in the very small area showing that other variables beyond personality are likely of higher importance to understand individual differences in attitudes towards artificial intelligence. Regarding the administered attitudes towards AI scale, we mention that the ATAI is only a rough screener for attitudes towards AI, and future studies also need to administer more fine-grained tools to grasp more facets of attitudes towards artificial intelligence (see a twenty item measure on general attitudes towards AI: Schepman and Rodway 2020), which might lead to different personality-attitudes towards AI associations; see exemplary recent works using not the ATAI presenting further associations (Kaya et al. 2022; Park and Woo 2022).

In sum, the present work is to our knowledge the first to use Panksepp’s primary emotional systems and Affective Neuroscience Theory to shed light on attitudes towards artificial intelligence. This work shows that such an evolutionary framework might help to understand a bit better why some persons have more negative or positive views on artificial intelligence. Going beyond our work, more integration of neuroscientific research could further stimulate AI developments (Bain and McCay 2023), but ethical aspects should be strongly considered when knowledge from neuroscience and AI is applied, e.g. as outlined in a recent paper on diagnostics of criminal recidivism in teenagers (Muñoz and Marinaro 2022).