Comparative summary of major computational models of addiction, showing their key mechanisms, behavioral focus, empirical grounding, and limitations.
| Model | Key mechanism | Main brain systems | Behavioral focus | Strengths | Limitations | Empirical support |
|---|---|---|---|---|---|---|
| TDRL/MF-RL | Dopaminergic reward-prediction error learning; cached stimulus-response values drive behavior. | VTA, ventral/dorsal striatum, OFC. | Habit formation, persistence, relapse. | Mechanistically precise; explains compulsion and habit strength. | Neglects craving, flexible thinking, emotion, recovery, intra-individual variability; dopamine-centric. | Rodent neurophysiology & RL behavior; ventral striatum (Keiflin and Janak [36] 2015); OFC & DLS, habit shift (Lucantonio et al. [37] 2014); Rodent RL (Groman et al. [38] 2019). |
| Dual-system (MB + MF) | Arbitration between goal-directed (MB) and habitual (MF) control. | Lateral PFC, OFC, ACC, striatum. | Transition from voluntary to compulsive use. | Captures flexibility vs. habit; links to CBT/CM. | Limited account of craving, emotion, and spontaneous recovery. | fMRI arbitration (Lee et al. [45] 2014); rodent modeling (Lesaint et al. [46] 2014); behavioral devaluation (Byrne et al. [47] 2019). |
| Incentive-sensitization theory (IST) | Sensitized mesolimbic dopamine amplifies cue-triggered “wanting” independent of pleasure. | Nucleus accumbens, VTA, ventral pallidum, PFC. | Craving, cue-reactivity, relapse. | Distinguishes “wanting” vs. “liking”; explains persistent cue-driven craving. | Lacks mechanisms for recovery, cognition, and symbolic meaning; dopamine-centric. | VP electrophysiology (Tindell et al. [56] 2004); DAT variation (Singer et al. [57] 2016); Review (Berridge and Robinson [55] 2016). |
| iRISA | Exaggerated cue salience + impaired executive inhibition. | Insula/ACC (salience), DLPFC/IFG (control), striatum, DMN. | Cue-reactivity, craving under weak control, and relapse. | Integrates cognition and motivation; strong imaging evidence; recovery correlates. | Limited modeling of belief/identity change or natural recovery. | Neuroimaging meta-analyses (Zilverstand et al. [23] 2018; Ceceli et al. [61] 2025; Zilverstand and Goldstein [60] 2020). |
| Bayesian/Active inference | Maladaptive high-precision priors bias inference about drug relief; craving as predictive belief. | Cortical predictive-coding circuits; dopamine/serotonin systems. | Cue reactivity, craving, relapse, and recovery as belief updating. | Unifies cognition, emotion, and context; models craving without cues. Accounts for natural recovery and identity change. | Abstract; limited circuit-level data; early empirical base. | Computational & imaging (Gu and Filbey [17] 2017; Harlé et al. [67] 2019; Kulkarni et al. [14] 2023). |
ACC: anterior cingulate cortex; CBT: cognitive behavioral therapy; CM: contingency management; DAT: dopamine transporter; DLPFC: dorsolateral prefrontal cortex; fMRI: functional magnetic resonance imaging; iRISA: impaired response inhibition and salience attribution; MB: model-based; MF: model-free; OFC: orbitofrontal cortex; RL: reinforcement learning; TDRL: temporal-difference RL; VTA: ventral tegmental area.