This article contributes to the debate around the abilities of large language models such as GPT-3, dealing with: firstly, evaluating how well GPT does in the Turing Test, secondly the limits of such models, especially their tendency to generate falsehoods, and thirdly the social consequences of the problems these models have with truth-telling. We start by formalising the recently proposed notion of reversible questions, which Floridi & Chiriatti (2020) propose allow one to ‘identify the nature of the source of their answers’, as a probabilistic measure based on Item Response Theory from psychometrics. Following a critical assessment of the methodology which led previous scholars to dismiss GPT’s abilities, we argue against claims that GPT-3 completely lacks semantic ability. Using ideas of compression, priming, distributional semantics and semantic webs we offer our own theory of the limits of large language models like GPT-3, and argue that GPT can competently engage in various semantic tasks. The real reason GPT’s answers seem senseless being that truth-telling is not amongst them. We claim that these kinds of models cannot be forced into producing only true continuation, but rather to maximise their objective function they strategize to be plausible instead of truthful. This, we moreover claim, can hijack our intuitive capacity to evaluate the accuracy of its outputs. Finally, we show how this analysis predicts that a widespread adoption of language generators as tools for writing could result in permanent pollution of our informational ecosystem with massive amounts of very plausible but often untrue texts.

The emotional properties of words, such as valence and arousal, influence the way we perceive and process verbal stimuli. Recently, subjective significance was found to be an additional factor describing the activational aspect of emotional reactions, which is vital for the cognitive consequences of emotional stimuli processing. Subjective significance represents the form of mental activation specific to reflective mind processing. The Lexical Decision Task (LDT) is a paradigm allowing the investigation of the involuntary processing of meaning and differentiating this processing from the formal processing of the perceptual features of words. In this study, we wanted to search for the consequences of valence, arousal, and subjective significance for the involuntary processing of verbal stimuli meaning indexed by both behavioral measures (reaction latencies) and electrophysiological measures (Event-Related Potentials: ERPs). We expected subjective significance, as the reflective form of activation, to shorten response latencies in LDT. We also expected subjective significance to modulate the amplitude of the ERP FN400 component, reducing the negative-going deflection of the potential. We expected valence to shape the LPC component amplitude, differentiating between negative and positive valences, since the LPC indexes the meaning processing. Indeed, the results confirmed our expectations and showed that subjective significance is a factor independent from the arousal and valence that shapes the involuntary processing of verbal stimuli, especially the detection of a link between stimulus and meaning indexed by the FN400. Moreover, we found that the LPC amplitude was differentiated by valence level.

The role of emotional factors in maintaining cognitive control is one of the most intriguing issues in understanding emotion-cognition interactions. In the current experiment, we assessed the role of emotional factors (valence, arousal, and subjective significance) in perceptual and conceptual inhibition processes. We operationalised both processes with the classical cognitive paradigms, i.e., the flanker task and the emotional Stroop task merged into a single experimental procedure. The procedure was based on the presentation of emotional words displayed in four different font colours flanked by the same emotional word printed with the same or different font colour. We expected to find distinct effects of both types of interference: earlier for perceptual and later for emotional interference. We also predicted an increased arousal level to disturb inhibitory control effectiveness, while increasing the subjective significance level should improve this process. As we used orthogonal manipulations of emotional factors, our study allowed us for the first time to assess interactions within emotional factors and between types of interference. We found on the behavioural level the main effects of flanker congruency as well as effects of emotionality. On the electrophysiological level, we found effects for EPN, P2, and N450 components of ERPs. The exploratory analysis revealed that effects due to perceptual interference appeared earlier than the effects of emotional interference, but they lasted for an extended period of processing, causing perceptual and emotional interference to partially overlap. Finally, in terms of emotional interference, we showed the effect of subjective significance: the reduction of interference cost in N450 for highly subjective significant stimuli. This study is the first one allowing for the investigation of two different types of interference in a single experiment, and provides insight into the role of emotion in cognitive control.

Warmth and competence are fundamental dimensions of social cognition. This also applies to the interpretation of ambiguous symbolic stimuli in terms of their relation to warmth or competence. The affective state of an individual may affect the way people interpret the neutral stimuli in the environment. As previous findings have shown, it is possible to alter the perception of neutral social stimuli in terms of warmth vs. competence by eliciting an incidental affect with the use of emotion-laden words. In the current experiment, we expected the valence and origin of an affective state, factors ascribing emotionally laden words, to be able to switch the interpretation of the neutral objects. We have shown in behavioural results that negative valence and reflective origins promote the interpretation of unknown objects in terms of competence rather than warmth. Furthermore, electrophysiological-response-locked analyses revealed differences specific to negative valence while making the decision in the ambiguous task and while executing it. The results of the current experiment show that the usage of warmth and competence in social cognition is susceptible to affective state manipulation. In addition, the results are coherent with the evolutionary perspective on social cognition (valence effects) as well as with predictions of the dual mind model of emotion (origin effects).

Vaccine acceptance is a key factor in achieving high immunization coverage and reducing the death toll of COVID-19 pandemic. Analyzing data from Europe and Americas we demonstrated that vaccine hesitancy strongly correlates with underreporting of COVID-19 deaths and cases. This correlation cannot be explained by the differences in economic indexes: GDP and Gini coefficient (measure of income inequalities). There is no correlation of vaccination percentage and Gini coefficient and the correlation with GDP is decreasing in time. The most striking is the comparison of Eastern European and South American countries; the latter group of countries shows significantly higher vaccination percentage while having a lower or comparable GDP and higher Gini coefficient. The analysis suggests that timely and reliable information about the COVID-19 cases and the associated deaths plays a key role in achieving population-wide vaccine acceptance.