Live Feeds
● LIVE Updated 1h ago · 25 sources tracked

Advanced AI models suffer a near-total collapse on classic psychology test as cognitive demands increase

Leading AI models experience a catastrophic performance collapse on the Stroop task. Accuracy falls as tasks grow longer and more complex. This suggests these systems lack human-like executive control.

RSS Source map (21)

What changed

No new corroborating evidence was provided regarding the specific PNAS Nexus study results.

Live updates

  1. AI Models Fail Stroop Task as Complexity Increases

    Leading AI models experience a catastrophic performance collapse on the Stroop task. Accuracy falls as tasks grow longer and more complex. This suggests these systems lack human-like executive control.

    What's confirmed:

    • AI models show a catastrophic performance collapse on the Stroop task.
    • Accuracy in these models drops as tasks become longer and more complex.
    • The performance drop indicates a lack of human-like executive control.
    confidence 100%

  2. Advanced AI Models Fail Classic Psychology Attention Test

    A study published in PNAS Nexus reveals that leading AI models suffer a catastrophic performance collapse on the Stroop task. While systems can identify colors in short lists, accuracy drops sharply as tasks become longer and more complex. Researchers suggest this indicates a lack of human-like executive control.

    What's confirmed:

    • Leading AI models dropped from over 90% accuracy to near-zero or complete failure as the complexity of the Stroop task increased.
    • The study was published in PNAS Nexus.
    • The Stroop test requires participants to identify the color of a word's text while ignoring the word itself.
    • Research indicates AI systems struggle with sustained focus and conflict resolution compared to human attention.

    Still unconfirmed:

    • GPT-5 failed the human attention test.
    • The findings raise concerns for autonomous drone navigation and real-time obstacle avoidance.
    • The lack of executive control could stall the development of human-level AI.
    confidence 90%

  3. Advanced AI Models Fail Classic Psychology Attention Test

    A PNAS Nexus study using the Stroop task revealed a catastrophic collapse in AI executive control. While models handle short lists, accuracy drops sharply as tasks grow longer and more complex. These results suggest AI lacks the human-like focus needed for artificial general intelligence.

    What's confirmed:

    • A study published in PNAS Nexus used the Stroop task to examine AI attention and executive control.
    • AI models can correctly name colors in short lists but experience a performance collapse as task complexity increases.
    • Some leading AI systems saw accuracy drop from over 90% to near-zero or near-complete failure.
    • The Stroop test requires participants to identify the color of a word's text while ignoring the word itself.
    • GPT-5 failed the human attention test.

    Still unconfirmed:

    • The failure in sustained attention may pose risks for autonomous drone navigation and real-time obstacle avoidance.
    • Suketu Patel and colleagues led the research into transformer-based machine attention.
    confidence 90%