Can we spot fake public comments generated by ChatGPT(-3.5, -4)?: Japanese stylometric analysis expose emulation created by one-shot learning.

Public comments are an important opinion for civic when the government establishes rules. However, recent AI can easily generate large quantities of disinformation, including fake public comments. We attempted to distinguish between human public comments and ChatGPT-generated public comments (including ChatGPT emulated that of humans) using Japanese stylometric analysis. Study 1 conducted multidimensional scaling (MDS) to compare 500 texts of five classes: Human public comments, GPT-3.5 and GPT-4 generated public comments only by presenting the titles of human public comments (i.e., zero-shot learning, GPTzero), GPT-3.5 and GPT-4 emulated by presenting sentences of human public comments and instructing to emulate that (i.e., one-shot learning, GPTone). The MDS results showed that the Japanese stylometric features of the public comments were completely different from those of the GPTzero-generated texts. Moreover, GPTone-generated public comments were closer to those of humans than those generated by GPTzero. In Study 2, the performance levels of the random forest (RF) classifier for distinguishing three classes (human, GPTzero, and GPTone texts). RF classifiers showed the best precision for the human public comments of approximately 90%, and the best precision for the fake public comments generated by GPT (GPTzero and GPTone) was 99.5% by focusing on integrated next writing style features: phrase patterns, parts-of-speech (POS) bigram and trigram, and function words. Therefore, the current study concluded that we could discriminate between GPT-generated fake public comments and those written by humans at the present time.

Distinguishing ChatGPT(-3.5, -4)-generated and human-written papers through Japanese stylometric analysis.

In the first half of 2023, text-generative artificial intelligence (AI), including ChatGPT from OpenAI, has attracted considerable attention worldwide. In this study, first, we compared Japanese stylometric features of texts generated by ChatGPT, equipped with GPT-3.5 and GPT-4, and those written by humans. In this work, we performed multi-dimensional scaling (MDS) to confirm the distributions of 216 texts of three classes (72 academic papers written by 36 single authors, 72 texts generated by GPT-3.5, and 72 texts generated by GPT-4 on the basis of the titles of the aforementioned papers) focusing on the following stylometric features: (1) bigrams of parts-of-speech, (2) bigram of postpositional particle words, (3) positioning of commas, and (4) rate of function words. MDS revealed distinct distributions at each stylometric feature of GPT (3.5 and 4) and human. Although GPT-4 is more powerful than GPT-3.5 because it has more parameters, both GPT (3.5 and 4) distributions are overlapping. These results indicate that although the number of parameters may increase in the future, GPT-generated texts may not be close to that written by humans in terms of stylometric features. Second, we verified the classification performance of random forest (RF) classifier for two classes (GPT and human) focusing on Japanese stylometric features. This study revealed the high performance of RF in each stylometric feature: The RF classifier focusing on the rate of function words achieved 98.1% accuracy. Furthermore the RF classifier focusing on all stylometric features reached 100% in terms of all performance indexes (accuracy, recall, precision, and F1 score). This study concluded that at this stage we human discriminate ChatGPT from human limited to Japanese language.

Criminal mutilation homicides in Japan: Corpse dismemberment and disposal pattern types, and offender characteristics.

This study examined the offender characteristics of criminal mutilation homicides in Japan for offender profiling. The current sample (N = 108) showed that the most typical offender was a male, in his 30s (age that observed most often), one-person, and acquainted with the victim. This study highlighted that the cases of mutilation homicide differed in comparison with the regular homicide group in terms of the victim-offender (VO) relationship and the number of offenders. Furthermore, the multiple correspondence analysis classified mutilation homicides into four behavioral pattern types based on two dimensions (level of "efforts of corpse dismemberment" and "efforts of corpse disposal"). In the first type "high evidence destruction effort (n = 29)", having the greatest efforts to destroy evidence, all offenders were men. The rates of no VO relationship and serial murders were higher. In the second type "convenient disposal (n = 32)", likely to abandon victim body in home and trash, the rate of female offenders who assaulted family members was statistically significantly higher among the four types. In the third type "high geographic distance (n = 23)", traveling long distances for corpse disposal, the percentage of cases by co-offenders was likely to be higher than in the other types. Lastly, the offenders in "low evidence destruction effort (n = 24)" were likely to kill other acquaintances and have the least efforts for corpse dismemberment and disposal. From these results, the criminal profilers can estimate mutilation homicide offenders on the basis of efforts of corpse dismemberment and disposal.

External validity of Concealed Information Test experiment: Comparison of respiration, skin conductance, and heart rate between experimental and field card tests.

The Concealed Information Test (CIT) is a memory detection method based on an examinee's physiological responses to crime-relevant items (critical items) in comparison with crime-irrelevant items (noncritical items). I examined the external validity of CIT laboratory experiments by comparing respiratory speed (RS), skin conductance response (SCR), and heart rate (HR) between a laboratory experiment (n = 30) and a field test (n = 30) in a card test. A linear mixed models analysis showed that the critical items in both CIT contexts elicited CIT effects: RS repression, SCR increase, and HR deceleration. Additionally, the critical items in the field elicited a larger RS suppression than those in the laboratory; however, SCR and HR on the critical items showed no statistical differences between the laboratory and field CITs. Moreover, the tonic HRs in the field tests were higher than in the CIT experiments, but there were no significant correlations between the tonic HR and each CIT effect in RS, SCR, and HR. These results suggested that CIT detection efficiency was not affected by arousal levels and may imply that the CIT effects observed with RS, SCR, and HR in the CIT laboratory experiments has adequate external validity and can be generalized to the CIT field.

[Effects of recollection and familiarity on recognition memory in the Concealed Information Test].

The effects of recollection and familiarity on recognition memory in the Concealed Information Test (CIT) were investigated. In the learning phase, participants encoded 20 words that were presented in one of four frames on a personal computer screen. Next, in the recognition test, the participants were divided into "Remember judgment (Recollection)" or "Know judgment (Familiarity)" groups, based on their Remember/Know judgment when recognizing learned items. In the CIT phase, physiological responses to questions about learned (i.e., critical) and non-learned (i.e., non-critical) items were measured and recorded. The results indicated that there was a deceleration of respiration speed (RS), an increase in skin conductance response (SCR), and a drop in heart rate in responses to critical items for both groups. Furthermore, the effect sizes of RS and SCR were greater in the "Remember judgment" group compared to the "Know judgment" group. These results suggest that critical response patterns are generated by recollection and familiarity. However, the more vividly participants recognized critical items the larger were the magnitudes of RS and SCR response patterns.

[Effects of false memories on the Concealed Information Test].

The effects of false memories on polygraph examinations with the Concealed Information Test (CIT) were investigated by using the Deese-Roediger-McDermott (DRM) paradigm, which allows participants to evoke false memories. Physiological responses to questions consisting of learned, lure, and unlearned items were measured and recorded. The results indicated that responses to lure questions showed critical responses to questions about learned items. These responses included repression of respiration, an increase in electrodermal activity, and a drop in heart rate. These results suggest that critical response patterns are generated in the peripheral nervous system by true and false memories.

[Effect of fear on eyewitness' retrieval in recognition memory].

The effect of fear on the retrieval of eyewitness' recognition memory was experimentally investigated. Participants were presented with a thematic series of slides of either neutral or ambiguous stimuli. The ambiguous stimulus consisted of a red liquid flowing on a girl's head. Participants who saw the ambiguous stimulus interpreted the liquid as either red paint, or the girl's blood. Differences in the interpretation of the ambiguous stimulus affected the participants'recognition memory. Participants who interpreted the stimulus as blood displayed inferior memory compared to those who interpreted the stimulus as red paint or those who were shown the neutral stimulus. Moreover, when the participant's interpretation was changed from blood to red paint by an experimenter, the memory for central details was enhanced. These results suggest that fear interferes with retrieval. Participants' memory could be improved by reducing fear during the retrieval stage.