Rethinking model-based and model-free influences on mental effort and striatal prediction errors

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• A prevailing assumption posits that low-effort model-free learning is automatic, while complex model-based learning is effortful and selectively employed.
• This cost-benefit view suggests a trade-off influencing strategy selection in decision-making and learning.

Purpose of the Study:

• To challenge the standard assumption regarding the automaticity of model-free learning and the effort-based arbitration between learning strategies.
• To re-evaluate the role of mental effort in the selection between model-free and model-based learning mechanisms.

Main Methods:

• Re-analysis of neuroimaging data investigating reward prediction errors in the ventral striatum.
• Experimental manipulation of task instructions to influence model-based behavior and assessment of associated mental effort.

Main Results:

• Flaws in previous analyses of ventral striatum data suggest no evidence for automatic model-free prediction errors.
• Increased task instruction clarity promoting model-based behavior was associated with reduced, not increased, mental effort.
• Findings contradict the cost-benefit arbitration model for strategy selection.

Conclusions:

• Model-free learning may not be an automatic process as widely assumed in neuroscience.
• Humans can optimize mental effort by employing a singular model-based strategy, rather than arbitrating between multiple strategies.
• These results necessitate a re-evaluation of fundamental assumptions in influential theories of learning and decision-making.