Valuing AI Art: An Experimental Study (2)
How to cite
Lane, Tom; Heinrich, Timo; Pickard, Harry; Walker, Matt. 2026. "Valuing AI Art: An Experimental Study (2)." AEA RCT Registry. AEARCTR-0018880. https://doi.org/10.1257/rct.18880-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
This work extends trial AEARCTR-0014469. A similar experiment to the original will be run using a general population online sample. A further experiment will be run using mugs with poetry, rather than artworks.
Intervention
Artworks and mugs featuring poetry will be auctioned. The treatment is whether subjects are informed that the artwork or poem on the mug is human-made or AI-made.
See attached pre-analysis plan for full details.
Experimental design
Subjects participate in a series of 10 2nd price Vickrey auctions (Vickrey, 1961). Subjects are informed that they will be matched (ex post) into groups of 4 bidders for determining auction outcomes. These groups remain fixed across rounds. Each subject is endowed with an income of £12 that they can bid. In each round, one item – either a visual artwork or a haiku mug – from the relevant final sample (human-created or AI-generated depending on the treatment) is auctioned within each group. We will vary between subjects whether they participate in auctions of a visual artwork or a haiku mug. An image of the artwork or mug is displayed on the respective Qualtrics screen. The order of lots is randomized across participants. Subjects submit their bids for each lot using a sealed-bid format and no feedback is provided on auction outcomes within the group until after the session. Thus, the subject is the independent level of observation for bids/willingness to pay (which are theoretically equivalent in the Vickrey auction). After all of the auction rounds are completed, one round is selected at random separately for each group. The subject who submitted the highest bid in the group in the selected round’s auction wins the item and second-highest bid in the group is subtracted from the highest bidder’s income.
The rules of the auction are explained to subjects at the beginning, together with the strategic reasoning for why they should bid the highest amount of money that each item is worth to them. Providing this information is appropriate for experiments (such as this one) in which the goal is to elicit homegrown values under the assumption that subjects know the dominant strategy property of the auction is to bid their value (see Harrison et al., 2004, for a discussion). To further facilitate learning of the dominant strategy, each subject will individually undertake a hypothetical training exercise in which they will receive feedback if they fail to select the dominant strategy.
At the end, participants participate in a post-experimental survey, including questions about demographics, and subjects’ views towards and experiences with AI and art (more information in Section VIII).
A multi-step pre-testing process is conducted to select the items used in the experiment.
Further details are provided in the attached pre-analysis plan document.
Design details
Cited by
No citation data available for this registration yet.