The art-making process for “AI artists” is determining what string of words would yield the greatest picture — not what colors to combine or brush strokes to attempt.
As the complexity of AI art generators grows, so does the ability to trick spectators.
A poll was devised to see how well Yale students could tell if the art was made by AI or by humans. Respondents were asked to score nine different artworks, which included both AI-generated and human-created pieces. On average, the 504 students polled could distinguish whether art was created by AI or by humans 54% of the time. This equates to around five right answers out of nine questions.
The command “two pigs peering up at the moon, watercolor painting” was entered into DALL-E 2, an AI picture generator, and the image above was created. This artwork was drawn by a person.
While the last AI artwork perplexed respondents, 86 percent correctly identified this artwork as AI-generated. The prompt “a fortune-telling shiba inu reading your destiny in a crystal ball digital art” was entered into DALL-E 2 to generate the artwork above. Notably, the left pupil of the shiba inu is absent, among other minor anomalies.
This AI-generated artwork, created with the phrase “An empty teacup, oil painting,” fooled 59 percent of respondents into thinking it was created by humans.