In controlled tests evaluating artificial intelligence performance alongside consumer willingness to pay, researchers found that people offered roughly double the price for human-made goods compared to identical items produced by machines. The findings, derived from systematic evaluations of leading large language models and two preregistered experiments on human valuation, indicate that while generative software can perform tasks cheaply, human involvement continues to command a substantial economic premium.
The study, conducted by researchers Ellen Langer of Harvard University and Philip Maymin of Fairfield University and published in Frontiers in Artificial Intelligence, explored whether machines exhibit mindfulness and whether people are willing to pay for it. The authors framed mindfulness according to Langer’s definition as the simple act of actively noticing new things, addressing two central propositions: whether modern models notice contextual details and whether human noticing retains monetary value.
To test whether artificial intelligence models notice critical context, the authors evaluated a common reasoning challenge that circulated online in 2026 involving whether to walk or drive 50 meters to a car wash. In the first systematic trial of the problem, most leading models recommended walking, effectively advising users to leave the vehicle at home. Across an archive of 4,240 responses evaluated on a nine-question battery, GPT-5.5 and Fable 5 passed all 90 test trials, whereas competing models successfully solved one question but failed subsequent ones.
Model performance proved sensitive to the surrounding software environment and prompting structure. In an exploratory comparison, three of four evaluated models performed worse when deployed inside an agent harness than when operated from a plain command line. Direct prompting interventions also showed mixed results: instructing five models that previously failed the car-wash question to notice produced only 9 successful passes in 400 attempts, while explicitly naming the underlying goal corrected some models entirely but left others unchanged.
To measure the economic value placed on human attention, the researchers ran two preregistered experiments involving 1,006 and 589 participants who evaluated six distinct products. When informed that a product was made by a human rather than artificial intelligence, participants roughly doubled the amount they reported being willing to pay, demonstrating an average price premium of 103 percent.
The premium attached to human labor showed nuanced variation based on how the process was portrayed. Describing a human worker as exercising active judgment increased willingness to pay by 20.9 percent, yet a human described as strictly following procedures without creative judgment still commanded 53.8 percent more than artificial intelligence. Participants also reported higher quality expectations for human-made items, leading the authors to conclude that human production retains independent commercial value without requiring machine failure.
In analyzing machine performance, the researchers concluded that mindfulness in artificial intelligence cannot be treated as a single, static trait. Instead, the authors describe mindfulness in computational systems as an operational map, where a model’s capacity to notice context varies according to the specific question asked, the phrasing of the prompt, and the surrounding software framework. Because models that resolved one test challenge frequently failed an adjacent task, and because agent harnesses degraded performance compared to direct command lines, the authors suggest that artificial intelligence evaluations must change their focus. Rather than assigning simple pass or fail grades, future benchmarks should systematically map which specific contextual details individual systems successfully register or ignore.
The investigation into prompt sensitivity also exposed sharp boundaries in how models process instructions intended to induce awareness. In experiments with models that had failed the car wash reasoning problem, the researchers applied guidance modeled on instructions known to reduce decision biases in human subjects. While these prompt structures encourage active noticing in people, they produced minimal effect in algorithmic architectures, yielding only nine passes across 400 attempts. In contrast, directly identifying the underlying goal corrected certain systems completely while leaving other models entirely unchanged. This variance demonstrates that noticing in large language models is not a generalized capability, but a behavior heavily dependent on specific prompt phrasing and external scaffolding.
Turning to consumer valuation, the findings show that mindfulness carries its own distinct economic price above an already substantial baseline premium for human involvement. The two preregistered experiments revealed that consumers value human effort even when creative discretion is absent. Although portraying a human maker as actively exercising judgment yielded an additional 20.9 percent increase in willingness to pay, participants still offered a 53.8 percent premium for human labor described as strictly following routine procedures without creative judgment. This demonstrates that while active mindfulness clearly enhances commercial appeal, human origin itself establishes a powerful baseline advantage that machines do not erode.
Part of this willingness to pay appears tied to quality expectations, as participants anticipated that goods made by people would be better than machine alternatives. When intelligence becomes cheap and widely accessible, the authors conclude that human production maintains durable value in its own right. Human work does not need artificial intelligence to fail to hold its economic standing, as consumers place intrinsic worth on human authorship while awarding an added premium for active noticing.
Subject of Research: Evaluation of machine mindfulness and the economic value of human production
Article Title: When intelligence is cheap, mindfulness is priceless: AI performance and the value of human production
Article References: Langer, E. J., & Maymin, P. Z. (2026). When intelligence is cheap, mindfulness is priceless: AI performance and the value of human production. Frontiers in Artificial Intelligence, 9, 1914761. https://doi.org/10.3389/frai.2026.1914761
Image Credits: AI Generated
DOI: 10.3389/frai.2026.1914761
Keywords: active noticing, human production, large language models, mindfulness, preregistration, willingness to pay, prompt sensitivity, GPT-5.5, Fable 5, agent harness, command line evaluation, human premium, creative judgment, procedural labor, product valuation, consumer expectations, car wash problem
News Source: Blake Davidson. (November 16, 2026). Consumers Pay Double for Human Made Products Compared to Artificial Intelligence. Scienmag.



