New study challenges how researchers measure executive function in AI and human cognition
Researchers found that basic cognitive processes — not specialized executive abilities — explain most similarities in how people inhibit impulses, shift tasks, and update information. The finding could reshape how companies and institutions design cognitive assessments, employee training programs, and AI systems that attempt to replicate human decision-making.
Originaltitel: Lower level cognitive processes explain most of the unity of executive functions
<p>Executive functions (EFs) are a set of cognitive abilities that regulate behavior to reach goals retained temporarily in mind. One of the most widely accepted theoretical accounts of EFs is the EF unity and diversity framework: a pattern of intercorrelation (unity) among three domains of EF (inhibition, shifting, updating) that is not near perfect, indicating separability (diversity) of the EFs. However, the structural validity of this framework remains debated, particularly regarding the difficulties in dissociating executive abilities from lower level cognitive processes (LLPs). EF cost measures, derived by subtracting performance in nonexecutive control conditions from executive task performance, have been proposed to isolate executive functioning, although this has faced increasing criticism, and cost scores are typically available only for inhibiting and shifting tasks, not updating. To address this issue, we reanalyzed four data sets (two adolescent and two adult samples, each with >180 participants) using structural equation modeling. An LLP latent factor representing shared variance from control conditions (task conditions with minimal executive demands) was regressed onto three intercorrelated EF factors, each capturing shared variance from the executive conditions without relying on cost scores. The results supported the three-intercorrelated EF factor model when only EF factors were included. In the model that incorporated the LLP factor, LLP strongly predicted all three EF domains and altered their intercorrelations, rendering most nonsignificant. We conclude that the EFs’ intercorrelation patterns are substantially influenced by LLP, as the “unity” of EFs no longer holds once LLP is accounted for in all three EF latent domains</p>