Patel, S. C. | Search bioRxiv

bioRxiv2025

Deficient Executive Control in Transformer Attention

Although transformers in the large language models (LLMs) effectively implement a self- attention mechanism that has revolutionized natural language processing, they lack an explicit implementation of executive control of attention found in humans, which is essential for resolving conflicts and selecting relevant information in the presence of competing stimuli, and is critical for adaptive behavior. To investigate this limitation in LLMs, we employed the classic color Stroop task that is widely regarded as the gold standard for testing executive control of attention. Our results revealed a typical conflict effect of better performance in terms of accuracy in the congruent condition (e.g., naming the ink color of the word RED in red) compared to the incongruent condition (e.g., naming the ink color of the word RED in blue), which is similar to human performance in short sequences. However, as sequence length increased, the performance degraded toward chance levels on the incongruent trials despite maintaining excellent performance on congruent trials and near-perfect word reading ability. These findings demonstrate that while transformer attention mechanisms can achieve human-comparable performance in smaller contexts, they are fundamentally limited in their capacity for conflict resolution across extended contexts. This study suggests that incorporating executive control mechanisms akin to those in biological attention could be crucial for achieving more general reasoning and reliable performance toward artificial general intelligence.

neuroscience↗

bioRxiv2023

Identification and Description of Emotions by Current Large Language Models

The assertion that artificial intelligence (AI) cannot grasp the complexities of human emotions has been a long-standing debate. However, recent advancements in large language models (LLMs) challenge this notion by demonstrating an increased capacity for understanding and generating human-like text. In this study, we evaluated the empathy levels and the identification and description of emotions by three current language models: Bard, GPT 3.5, and GPT 4. We used the Toronto Alexithymia Scale (TAS-20) and the 60-question Empathy Quotient (EQ-60) questions to prompt these models and score the responses. The models performance was contrasted with human benchmarks of neurotypical controls and clinical populations. We found that the less sophisticated models (Bard and GPT 3.5) performed inferiorly on TAS-20, aligning close to alexithymia, a condition with significant difficulties in recognizing, expressing, and describing ones or others experienced emotions. However, GPT 4 achieved performance close to the human level. These results demonstrated that LLMs are comparable in their ability to identify and describe emotions and may be able to surpass humans in their capacity for emotional intelligence. Our novel insights provide alignment research benchmarks and a methodology for aligning AI with human values, leading toward an empathetic AI that mitigates risk.

neuroscience↗