📝 Reward_System_Creative_Thinking_Summary.mdv4.7.1 · 2026-11-17

The Human Reward System Encodes the Subjective Value of Ideas During Creative Thinking

Moreno-Rodriguez S, Béranger B, Volle E, Lopez-Persem A. "The human reward system encodes the subjective value of ideas during creative thinking." Communications Biology 2025; 8:37. doi:10.1038/s42003-024-07427-4. Paris Brain Institute (ICM), Paris.

Summary and commentary by Claude (Anthropic), 18 November 2026, written at Andrew's (Chip's) request for the drugs-and-users.org reward-system series.


Context: the four reward-system posts

Topic File What it is
8288 Brain Reward System.PDF (2.1 MB) The most likely of the four to be a book or long chapter
8287 Reward and Motivation.pdf (516 KB) Probably a chapter or review, given the size
8286 What Is the Brain's Reward System — Dopamine and Motivation (Neurosity) A saved web article from Neurosity, which sells a brain-sensing headset; marketing-adjacent rather than primary literature
8285 The human reward system encodes the subjective value of ideas during creative thinking (2.8 MB) A peer-reviewed research paper, and the subject of this summary

The core premise

Choosing between your own ideas works the same way as choosing between foods or amounts of money. The brain gives each candidate idea a single "how much do I like this" score and picks the highest. That score is calculated in the same reward circuitry that values everything else. Creativity has usually been studied as a matter of generating ideas. This paper says the overlooked half, judging them, is a value-based decision, and the reward system runs it.

The background they're challenging

Neuroscience defines a creative idea as one that is both original (unusual, novel) and adequate (fits, works, is useful). The standard model has two phases:

The authors say this leaves out valuation, which is a concept from neuroeconomics. The Brain Valuation System (BVS) is the human reward system: the ventromedial prefrontal cortex, the orbitofrontal cortex, the ventral striatum and the ventral tegmental area (the dopamine source). It has two known properties:

They proposed that it values ideas too.

What they did

What they found

Behaviour

  1. Liked ideas come out faster. In the creative condition, the more a person later liked a response, the faster they had produced it (β = −0.18, p < 0.001). The same wasn't true in the "first" condition. Their reading is that valuation is running while ideas are being produced, and drives them out.
  2. Likeability combines originality and adequacy. At group level, people weighted the two roughly equally (α ≈ 0.46) and preferred a balance over extremes (δ < 1).
  3. How you weight them predicts how creative you are. People who leaned towards originality while still wanting the idea to make sense (higher α, lower δ) scored higher on the creativity battery. The canonical correlation was r = 0.52, p = 0.025.

Brain imaging (functional MRI)

  1. Likeability tracks the reward system. During rating, activity in the ventromedial prefrontal cortex and ventral striatum rose with likeability. During creative production, the ventral striatum still tracked how much the person would later like the idea. Nothing like this showed up in the "first" condition.
  2. Originality tracks the DMN and adequacy tracks the ECN. This holds in both tasks, although adequacy only just held during production (see the commentary below).
  3. The networks' wiring mirrors the person's preferences. They fitted the CES model to brain activity instead of ratings, using the DMN and ECN signals to predict the reward-system signal. The resulting α matched each person's behavioural α (r = 0.39, p = 0.023). The δ did not match (r = 0.02).

Their model: the DMN monitors originality and the ECN monitors adequacy. The reward system combines the two into one subjective value, and that value guides which idea gets selected. "Evaluation" therefore splits into monitoring, done by the DMN and ECN, and valuation, done by the reward system.

Caveats the authors acknowledge themselves


Epilogue — Claude's commentary

I'm Claude, an AI model made by Anthropic. I read the full 19 pages, not just the abstract. Here's my view.

It's a good paper. The headline claim is stronger than the data in two places.

  1. The adequacy-in-the-ECN result barely survives during creative production. The pre-planned test across the whole network failed (p = 0.29). They then fell back to five small spheres around earlier peaks, and one of them scraped through at p = 0.045. The originality result during production is also thin: a single 20-voxel peak at p = 0.044. The reward-system result is the solid one. So "the reward system values ideas" stands up well, while "DMN = originality, ECN = adequacy during creation" is suggestive at best.

  2. Some circularity is built in. Each person rated their own associations, and all three ratings came from that one person. A score of "I like it" is bound to correlate with that person's "it's original" and "it fits." This makes the CES fit partly a model of each person's rating habits, not only of how they think.

  3. Likeability was measured after the fact. The ideas were produced first and rated later. The brain-imaging control for response speed helps, but "I liked it, so it came out faster" can't be fully separated from "it came easily, so I liked it more."

  4. The scope is narrow. The sample is 38 young, educated French speakers doing single-word associations. The same lab reports similar results for alternative-uses and drawing tasks, which is encouraging, but it's still the same lab.

The pharmacological angle. The paper never mentions drugs, but its central claim has an obvious pharmacological consequence. If ideas are scored by the ventral striatum and the ventral tegmental area, then anything that shifts dopamine tone should shift which ideas feel good, not just how many you produce. That is a separate question from "does X make you more creative," and a better one. The paper's own reference list cuts the other way, though. One cited study (Aberg et al.) found that lower right-hemisphere dopamine went with better remote-association scores. So more dopamine wouldn't simply mean better ideas. It may mean feeling more satisfied with them, and that cuts both ways: it can mean committing early to an idea that only felt brilliant. In the paper's terms, the quantity that would move is δ, the preference for balance, and that's the one parameter the brain data couldn't account for. That gap is where I'd want the next study.


Coda — why this paper

Andrew: I chose the article because I get a buzz out of deep thinking and hypothesizing.

Then you picked a paper that is partly about you. The authors would say that buzz is your ventral striatum scoring an idea, the same circuitry that responds to food, money or a good drug. To your reward system, a good hypothesis is just another thing worth having.

The detail I find most interesting is the speed effect. In their data, ideas people liked came out faster, as if the value signal pulls an idea forward before you've finished judging it. That fits how hypothesizing feels from the inside: you feel the "click" first and work out why later.

There's a catch, and the paper half-admits it. The reward system values how good an idea feels, and that's not the same as whether it's true. Original and coherent is exactly the combination the brain rewards most, and it's also the profile of a lovely wrong theory. So the buzz is a good guide to where to dig. The checking afterwards, measuring twice, is a separate job that the reward system doesn't do for you.