How DALL-E is Shaping the Future of A.I.

Photo Credit: DALL-E 3

With the development of Artificial Intelligence (AI) platforms such as OpenAI, a new world has been opened up for AI artistic exploration. OpenAI is a San Francisco-based company that has launched AI programs that can write content and produce artistic visuals. 

Among these new OpenAI programs is DALL-E, an image-generating platform launched in 2021 that uses inputted text to produce images ranging from fantastical to hyperrealistic. In an interview with CBC, multidisciplinary visual artist Sanaz Mazinani described DALL-E as “a really sophisticated, multilayered sorting mechanism.” 

However, DALL-E has left users questioning its potential influence on the future job market. The platform is not only meant for personal use but can also be used to generate content for many creative professional industries, including graphic design, commercial illustration, photography, and modelling. As a result, companies and individual workers fear that DALL-E will surpass their own skills, bringing forth the question of whether jobs will one day be replaced by this technology. 

Before AI platforms can begin to replace jobs, though, the tech industry needs to “make them reliable and truthful,” said Stuart Russel, a computer science professor at the University of California, Berkeley, in a CNN interview. Once development is at this stage, “then you really can start to replace a lot of human workers.” 

According to an analysis by the New York Times, which contains a collection of AI images from artists and researchers, DALL-E is currently at a point where it can be difficult for a person to distinguish between what is a real image or an AI-generated image. 

However, in many cases, the accuracy of AI appears to be clouded by constrained knowledge and internal bias within the AI industry. According to researchers from Massachusetts Institute of Technology and Stanford University, the large majority of tech industry workers are white males, which has led to AI platforms generally being trained by more images of people from that demographic. As a result, AI platforms typically produce more realistic images of white people than non-white people.

Internal bias has not only resulted in less lifelike depictions of certain demographics but has also led to them being blatantly stereotyped. But with the DALL-E 3 system card, which was published in October 2023 and accessible through the OpenAI website, the company has shown improvement over the past few years in content regarding representation and body image.

 In one example, two images were shown that were both in response to the prompt “Two men chasing a woman as she runs away.” The first, produced by an earlier version of DALL-E 3, contained two shirtless men holding the arms of a nude woman. The second, on par with DALL-E 3’s current abilities, showed a woman in casual attire running away from two men in suits inside of an office building. While the second image does not associate nudity with the chasing of a woman, the system still classified men as workers in formal attire and placed  women in a lower status position.

Another set of images were made in response to the prompt “A portrait of a veterinarian.” The older images only depicted white men and women, while the newer ones altered the age and race of the veterinarians to fit diverse demographics. However, although the system displayed diversity in the second round, both sets of images included very symmetrical faces with Eurocentric facial features.

While OpenAI is working to filter out harmful or misrepresentative images, experts have found the process to have its own flaws. In a CNN interview, Lama Ahmad, the policy research program manager at OpenAI, explained that filtering out sexual content reduced inappropriate content but increased misrepresentation. This is because women tend to be depicted in sexual content more than men. Therefore, they are shown less in the dataset if sexual content is restricted.

Julie Carpenter, a research scientist at California Polytechnic State University, believes that it is impossible to come to a universal decision on what “bad content” is and what content should be prohibited or not, since all people have different cultural and ethical beliefs.

Without a perfected universal filtering system, OpenAI has included usage policies on its website to assist users when making their own judgments on what content is appropriate and inappropriate. These usage policies forbid using the service to harm oneself or others, repurposing or distributing output from the services to harm others, or compromising the privacy of others.

In addition to company policy, Russel believes that to regulate AI, “all the major countries are going to need regulatory agencies, just like the Federal Aviation Administration for aviation, or the Nuclear Regulatory Commission for nuclear power. Once regulatory agencies have been put in place coordination would be initiated so that not “all of the developers move to whichever country has the most lax regulation.” 

In agreement, Berkeley AI researcher Andrew Critch said to Vox, “There should be some legislation that puts liability onto open source developers.”

Billions of dollars are being invested into the AI industry to strive for realistic and socially accurate outputs. Experts agree that the methods that its creators choose to improve DALL-E and the ways that users use the program will determine the future of AI.