Neurophysiological Characteristics of Adaptive Reasoning for Creative Problem-Solving Strategy

TL;DR

This study probes the neural basis of adaptive reasoning under dynamic rule switching by combining a WCST-BCI task with EEG in humans and contrasting it with a multimodal large language model (MLLM). It identifies coordinated oscillatory patterns—early delta/theta linked to exploratory rule inference and later occipital alpha signaling stabilized attention after rule identification—and characterizes ERP components associated with feedback processing (FRN and P300) during adaptive switching. Humans exhibit genuine adaptive reasoning with context-sensitive rule abstraction, whereas the MLLM shows only short-term, feedback-driven adjustments without hierarchical rule formation, highlighting a gap between biological and artificial context-sensitive reasoning. The findings suggest that brain-inspired AI should incorporate oscillatory coordination across processing stages to achieve true adaptive reasoning in changing environments.

Abstract

Adaptive reasoning enables humans to flexibly adjust inference strategies when environmental rules or contexts change, yet its underlying neural dynamics remain unclear. This study investigated the neurophysiological mechanisms of adaptive reasoning using a card-sorting paradigm combined with electroencephalography and compared human performance with that of a multimodal large language model. Stimulus- and feedback-locked analyses revealed coordinated delta-theta-alpha dynamics: early delta-theta activity reflected exploratory monitoring and rule inference, whereas occipital alpha engagement indicated confirmatory stabilization of attention after successful rule identification. In contrast, the multimodal large language model exhibited only short-term feedback-driven adjustments without hierarchical rule abstraction or genuine adaptive reasoning. These findings identify the neural signatures of human adaptive reasoning and highlight the need for brain-inspired artificial intelligence that incorporates oscillatory feedback coordination for true context-sensitive adaptation.

Figures (4)

• Figure 1: Overview of the WCST-BCI paradigm. (a) Each session consisted of six rule blocks, where four rules (color, shape, number, and border color) were presented in a random order during the first four blocks, followed by two additional blocks in which the rules were randomly selected. The hidden rule automatically switched after ten consecutive correct responses, and short breaks were provided between sessions to reduce fatigue. (b) Within each run, a single trial began with a fixation cross, followed by the presentation of four key cards (“Keys on”) and one stimulus card. Participants selected the matching card according to the current hidden rule, after which feedback (“Correct” or “Incorrect”) was presented on the screen.
• Figure 2: Stimulus-locked band-wise topographic $t$-maps for $\Delta$CONF--SEARCH; red box and yellow dots indicate significant clusters ($p<0.1$).
• Figure 3: Feedback-locked band-wise topographic $t$-maps for $\Delta$COR--INC; red boxes and yellow dots indicate significant clusters ($p<0.1$).
• Figure 4: Grand-averaged ERPs at the Cz channel from a representative participant. (a) Stimulus-locked responses for CONF and SEARCH conditions. (b) Feedback-locked responses for COR and INC conditions.