New research reveals that artificial intelligence models not only mimic human perception and language, but also share a fundamental brain-like mechanism for reward valuation. Scientists at EPFL and MIT have discovered that Vision-Language Models possess specialized units that anticipate rewards, analogous to the human nucleus accumbens (NAc)—a key brain region involved in motivation. When these units were artificially disrupted, the models showed a striking shift toward low-effort, low-reward choices, mirroring symptoms of anhedonia in depression.
The Research
Led by Melika Honarmand, Samin Mahdipour Aghabagher, and Martin Schrimpf, the team tested a state-of-the-art Vision-Language Model on effort-based decision tasks adapted from clinical psychology. They first identified "reward-anticipatory units" by analyzing neural activity patterns that correlated with upcoming reward magnitude. These units were functionally analogous to the NAc—they fired strongly when the model expected a high reward. The researchers then selectively perturbed these units and observed a dramatic change: the model became more likely to choose low-effort, low-reward options, even when high-reward opportunities were available. Crucially, when reward-based choice was removed, the model performed at baseline, confirming the deficit was specific to reward valuation, not general capability. The induced behavior aligned with two clinical scales: the Dimensional Anhedonia Rating Scale (DARS) and the Motivation and Pleasure Scale—Self-Report (MAP-SR).
Why It Matters
These findings offer a concrete computational model of anhedonia—a core symptom of depression that involves reduced motivation and pleasure. By replicating human reward deficits in AI, researchers can now test causal mechanisms that are hard to isolate in the brain. For individuals, this research underscores that motivation isn't just a “willpower” issue; it's rooted in specific neural (or computational) circuits that can be understood and potentially trained. Understanding these circuits can lead to better cognitive training exercises that target reward anticipation, helping people sustain effort toward long-term goals.
What You Can Do
You can strengthen your brain's reward anticipation by setting clear, achievable milestones and celebrating small wins. Practice visualizing the positive outcomes of your efforts before starting a task. Games that require short-term effort for delayed rewards—like puzzle solving or strategy games—may also train your reward system. The iqgenio platform offers brain training levels that incorporate reward-based learning to boost motivation.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
