2026-10-14T20:25:05Z https://www.scidb.cn/oai

10.57760/sciencedb.40369 2026-09-13T20:11:29Z

2026-09-13 From Human-Directed to AI-Directed Social Perception: Behavioral and Neurophysiological Evidence for a Three-Dimensional Trait-Transfer Framework doi:10.57760/sciencedb.40369 en This dataset was generated from an experimental study examining the transfer of human social perception traits to the social perception of artificial intelligence (AI). The study used a human–AI trait judgment task in which participants judged whether Chinese trait phrases varying in trait dimension and valence were applicable to either humans or AI. The experiment adopted a 2 (target category: human, AI) × 3 (trait dimension: warmth, competence, morality) × 2 (valence: positive, negative) within-subject design. Sixty university students were initially recruited. After excluding participants with poor EEG data quality or insufficient valid trials, data from 58 participants were included in the final dataset. The experimental task was programmed and presented using E-Prime 3.0. EEG signals were recorded using a 32-channel Brain Products actiCHamp Plus EEG recording system at a sampling rate of 500 Hz. The main measures include trait attribution counts, response times, event-related potential (ERP) indices, and event-related oscillation (ERO) power indices.The dataset is provided as a processed and summarized Excel file and does not contain personally identifiable information. Behavioral data were processed by excluding non-response trials and trials with response times shorter than 200 ms or longer than 6000 ms. Trait attribution counts and mean response times were then calculated for each participant under each experimental condition. EEG data were preprocessed in MATLAB R2024b using EEGLAB 2025.0. Preprocessing steps included rereferencing, 0.1–30 Hz band-pass filtering, artifact trial rejection, independent component analysis (ICA), and artifact component identification and removal using ICLabel. ERP indices were time-locked to the onset of trait phrases and included the mean amplitudes of the N170 and late positive potential (LPP). ERO indices were obtained using short-time Fourier transform with a Hanning window, and alpha- and beta-band power changes relative to baseline were extracted for further analysis.The temporal information in this dataset mainly concerns millisecond-level experimental timing, including response times and the time windows used for ERP and ERO analyses. ERP epochs were extracted from 200 ms before to 1000 ms after trait phrase onset, and ERO epochs were extracted from 1000 ms before to 1000 ms after trait phrase onset. Spatial information is reflected in the EEG electrode locations, which followed the international 10–20 system. ERP and ERO measures were calculated from predefined electrodes or regions of interest. As this was a laboratory-based psychological experiment, the dataset does not involve geographic spatial information.The Excel file contains summarized behavioral and EEG indicators. Depending on the data structure, each row represents either one participant under one experimental condition in long format, or one participant in wide format. The main variables include participant ID, target category, trait dimension, valence, trait attribution count, response time, N170 mean amplitude, LPP mean amplitude, alpha-band power, and beta-band power. Target category includes human and AI; trait dimension includes warmth, competence, and morality; valence includes positive and negative. Trait attribution count is measured as the number of “applicable” responses. Response time is measured in milliseconds (ms). ERP mean amplitude is measured in microvolts (μV). ERO power is expressed as baseline-corrected power change in decibels (dB). If the dataset is organized in long format, it contains 58 participants × 12 experimental conditions, resulting in 696 condition-level records. If the dataset is organized in wide format, each participant corresponds to one row, resulting in 58 participant-level records.Invalid trials and excluded participants have been removed from the final dataset. The dataset does not include raw continuous EEG recordings. Because the data were obtained from behavioral responses and EEG recordings, potential sources of measurement error include individual response variability, ocular or muscle artifacts, equipment noise, and choices made during preprocessing. These potential errors were reduced through artifact rejection, ICA-based correction, valid-trial checks, and condition-balance checks. The data file is provided in standard Excel format (.xlsx) and can be opened and analyzed using Microsoft Excel, WPS Office, LibreOffice Calc, SPSS, R, or Python. Artificial Intelligence; Social Perception; Trait Attribution; Event-Related Potentials; Event-Related Oscillations li jin Meng Zhen Sun Yu Zhong Yiping PUBLIC https://creativecommons.org/licenses/by/4.0/ dataset Science Data Bank