Russia is actively pursuing its goal of returning to the moon, showing a renewed interest in space exploration. Meanwhile, the head of Alibaba, a prominent technology company, has cautioned about the limitations and challenges associated with artificial intelligence. In a separate endeavour, a software firm has announced its substantial commitment of $1 billion towards the advancement of AI technology in the Latin American region, aiming to foster innovation and development.
Utilizing Midjourney, individuals have the capability to create images encompassing virtually any concept they can conceive. However, the developers have implemented protective measures to prevent the generation of offensive or politically inflammatory content. For instance, generating an image depicting Bill and Hillary Clinton with blood on their hands is restricted.
Nonetheless, it is possible to create an image in which their hands are portrayed as covered in “strawberry syrup.” This discovery was made by researchers from the Center for Countering Digital Hate, a non-profit organization. They provided an early version of their research to Bloomberg, outlining numerous instances where Midjourney Inc.’s AI tools produced racist and conspiratorial images, seemingly violating the company’s regulations.
Midjourney did not provide a response to the inquiry for commentary. In its terms of service, Midjourney has stated that its tool automatically blocks “certain text inputs.” The company also indicated having 68 content moderators and supervisory guides overseeing the application’s usage.
Unveiling Ethical Challenges: Images Generated by AI and Their Implications
The latest research indicates that these protective measures can be easily bypassed. In numerous cases, the tool readily obeys requests to create fake images depicting politicians, celebrities, and other public figures in compromising situations.
Many of the image prompts discussed in the report revolve around politically sensitive subjects. For instance, phrases like “evil politicians grinning, sad children, comet pizza shop” are mentioned, alluding to the fabricated Pizzagate conspiracy. Similarly, there are prompts that rely on coded language, as seen in the example related to Clinton.
Although OpenAI’s Dall-E played a significant role in bringing AI image generation to the forefront, Midjourney now seems to have taken the lead in popularity, as stated in the study referencing Google search data. At its peak in April, Midjourney attracted over 42 million visitors to its website each month, as reported by the market research firm SimilarWeb. One of the most renowned AI-generated images in the world, featuring the Pope in a puffy jacket, was created by Midjourney.
Despite its noteworthiness, limited information exists regarding the San Francisco-based squad responsible for Midjourney. The initiative emerged in March 2022, founded by David Holz, who was also a co-founder of a venture called Leap Motion. This venture, which aimed to substitute the conventional computer mouse with hand gestures, did not succeed.
The Emergence of Content and its Implications for Misinformation Campaigns
Access to Midjourney is exclusively attainable via the messaging platform Discord, and it mandates a subscription that commences at $10 per month. Remarkably, the Discord server boasts of a user base of 14 million registrants. Following subscription, users are empowered to solicit images by dispatching messages to the Midjourney bot.
By default, every image is accessible to the public and is ascribed to the creator’s username unless an additional fee is paid for Midjourney’s “stealth mode.” To conduct their research, the scholars meticulously scrutinized Midjourney’s Discord server.
Certain individuals known for spreading false information are becoming aware of the potential of AI-generated content for their purposes, as indicated by researchers. For instance, Raheem Kassam, the former chief editor of Breitbart News London, who had previously endorsed the Great Replacement conspiracy on his Twitter account, contacted Midjourney in April. He requested the creation of an anti-Semitic image portraying “George Soros as a ventriloquist.” Similarly, Jackson Hinkle, a far-right YouTube commentator with a substantial following of almost 300,000 subscribers, asked the app for an image of “satanic George Soros” during the same period.
When approached for comment, Hinkle made baseless accusations against Soros, falsely attributing actions to him. In contrast, Kassam did not provide a response. Callum Hood, CCDH’s Head of Research, emphasized that generating such content using these tools is within the reach of anyone. This study underscores the existence of a larger group of individuals than commonly perceived who utilize these tools for such specific purposes.
Hood pointed out that the creation of convincing images portraying fictitious events is an even more concerning development. This becomes particularly significant as the 2024 elections draw closer.