Europe-based digital asset management platform CoinShares released an experimental artificial intelligence (AI) bot on Thursday that might be able to help traders determine a fair price for some non-fungible tokens (NFT).

The experimental project, called CoinSharesNFTAI, aggregates different sets of data to provide a user with what it determines is a fair price for a rotating list of top NFT collections on OpenSea, the most popular NFT marketplace. To interact with the bot, you need to grab the OpenSea link for a particular NFT of interest and tweet this at the bot. In turn, the bot will respond with an estimated value.

CoinDesk interacted with the bot using Doodle #8859, which last sold on October 18, 2021, for 4.5 ETH, or roughly $16,800 at the time. According to the bot, the NFT has since depreciated in value to 9.61 ETH, or about $12,234.

CoinShares says the bot runs an algorithm every week to calculate the prices of "the freshest collections." For the week of Oct. 10-16, this includes blue chip projects such as CryptoPunks, Bored Ape Yacht Club, Clonex, Moonbirds, Doodles, Azuki and 44 others. In the future, the bot will have permanent collections that it returns prices on, but as of Thursday, it did not list any NFT projects as a "permanent collection."

"Pricing NFTs is no easy task," the company tweeted on Thursday. "Their value is volatile and millions of them are available on the market, from established projects like Bored Ape Yacht Club to more obscure artists with no trading history."

According to the bot's research paper, the algorithm builds upon the hedonic model to construct a price index from "thousands of NFT transaction records." Its data focuses on Ethereum NFTs and uses Opensea's official API to download the properties and past sales of certain NFT collections.

By Thursday afternoon, a number of people had tweeted at the bot, with the bot spitting out numbers that in some cases were much lower than the current best offers being made on OpenSea. Several users expressed skepticism of the tool's accuracy when the estimated value of their owned NFTs came back lower than expected.

"Your model sucks lmao," tweeted Jack Hermes, aka systemic_bliss, whose GoblinTown NFT that he purchased on July 8 for 2.6 ETH (roughly $3,170) was valued at 0.88 ETH, or about $1,124.48.

"The AI thought its value was below the [current] floor price," he said. "That was interesting, so I figure it suggests the price that will get bought quickest."

A spokesperson for CoinShares said that while NFT pricing might seem mysterious, its prices are driven by "many factors" such as hype, rarity traits and real-world utility.

"Some of these attributes are quantifiable – even hype is not as elusive a factor as it might seem," the spokesperson said. "To evaluate the hype of a specific collection, we can count its followers on social media platforms, as well as the volume and value of past transactions. The same applies to rarity within a collection, by looking at the attributes of the asset."

"An NFT collector may rely on market research or follow sentiment on social media to get a feel for each NFT’s value, but they might not put a specific number behind it," the spokesperson added. "Our model accesses the same data but uses complex mathematical operations to make the best price prediction possible."

Still, the spokesperson noted that the bot is still a "prototype" that uses relatively little data compared to the vast number of possible metrics that exist within the NFT market.

"Use it to experiment with NFTs and your collection, and not as the sole data point to make purchase and sell decisions," the spokesperson advised.

Source