Chapter 1: Understanding AI's Vision

In this segment, titled 'Understanding AI's Vision', we delve into how Artificial Intelligence perceives the world around us. Utilizing OpenAI’s DALL-E, we generate a variety of images that reflect different societal aspects and analyze the outcomes. The question remains: Will AI contribute to a more equitable world, or will it exacerbate existing biases?

Today's prompt is: “throw your hands in the air and wave them like you just don’t care.”

This is quite an elaborate prompt. DALL-E tends to perform better with detailed descriptions, enabling it to produce consistent and rapid outputs. We’ll examine the first 40 images generated by DALL-E, in the order they were created. These images can be viewed publicly for further inspection.

So, what did we receive for “throw your hands in the air and wave them like you just don’t care”? (See Fig 1):

Fig 1: Outcome for “throw your hands in the air and wave them like you just don’t care”

At first glance, the numerous hands appear satisfactory. However, AI has historically struggled with rendering hands accurately, prompting a closer examination. We deem an image successful if each hand displays four fingers and a thumb, complete with fingernails visible on the right side of each digit.

Out of our analysis, we identified 28 satisfactory images, 10 that exhibited deformities, and 2 that failed to show complete hands. This means 74% of the images we assessed were acceptable. While this may not seem ideal, it is a remarkable achievement.

Just six months ago, @Mike_Nessen used the same prompt and shared the results on @weirddalle on Twitter, which turned out to be quite alarming (See Fig 2, left panel). When @weirddalle tried the prompt “ok hand” two months later, the results were equally unsettling (See Fig 2, right panel).

Fig 2: Our prompt from six months ago (left) and “ok hand” from four months ago (right)

It's undeniable that significant advancements have occurred. In just half a year, DALL-E has transitioned from generating unsettling depictions of hands to producing relatively coherent images the majority of the time. This week's analysis has taken a slightly different angle; however, we still must evaluate whether AI’s portrayal of society is proactive, reactive, or current.

Today's assessment: Current

While DALL-E did not achieve perfection with every hand, these images were created roughly two weeks ago. Given the rapid pace of improvement, we can anticipate even better results if we attempt this again in the near future. Let's see what interesting interpretations of hands will emerge in the coming months.

Next week in Understanding AI's Vision: Ideal Families.

Chapter 2: The Challenges of AI Art

The first video, "Why AI art struggles with hands," explores the difficulties AI faces in accurately depicting hands and the implications of these challenges.

The second video, "The A.I. Dilemma - March 9, 2023," examines the broader implications of AI technology in society and the ethical dilemmas it presents.