STOPPER - An Executive Function Protocol for AI Assistants

AI language models excel at pattern matching but lack executive function - the regulatory mechanisms preventing impulsive, maladaptive behavior. Like humans with prefrontal cortex dysfunction, AI assistants exhibit “computational impulsivity”: rushing to solutions without diagnosis, repeating failed approaches, and exhibiting loop blindness. We propose STOPPER, a 7-step executive function protocol (Slow down, Think, Observe, Plan, Prepare, Execute, Read) addressing these deficits through external regulation.

After developing STOPPER independently, we discovered structural similarity to Dialectical Behavior Therapy’s STOP protocol (Linehan, 1993) - a clinical intervention for human emotional impulsivity. This convergent evolution suggests a hypothesis: executive function requirements may be universal across cognitive substrates. We identify four trigger types (Initial Prompt, Error Detection, Periodic Check, Uncertainty) and propose critical timing windows (0-10s pattern dominance, 10-30s intervention window, 30+s informed decision). Preliminary observations from manual application suggest potential effectiveness, though rigorous empirical validation remains future work. We present illustrative examples, discuss computational homology between human and AI executive function needs, and outline future research directions including automated implementation. This work bridges clinical psychology and AI safety, proposing that decades of DBT clinical evidence may inform AI reliability research if the computational homology hypothesis holds.