Midjourney AI image generator

Transparency and accountability 🙈

Midjourney is seen to suffer from several important transparency and accountability limitations: 

• Algorithmic decision-making. Details of Midjourney's underlying model are not publicly disclosed, making it difficult to understand how the AI generates images and what biases or limitations might be present. 

• Data sources. Midjourney does not provide information about or access to the data used to train the model, hindering independent research and analysis. 

• Copyright. Artists can request their work to be removed from Midjourney if they believe their copyright has been infringed. However, Midjourney only introduced the feature after artists complained, and then made it challenging for artists to remove their work by burying the information in the app terms of service.

• Complaints and appeals. Midjourney has faced criticism for its perceived sluggishness in addressing ethical dilemmas associated with deepfakes and other misuse of AI-generated content. 

Risks and harms 🛑

Midjourney stands accused of posing multiple serious risks and causing a variety of harms, including:

• Bias/discrimination. The training data for Midjourney reflects existing societal biases, which can result in the generation of images that perpetuate stereotypes or exclude certain demographics. This not only impacts the quality and inclusivity of generated content but also raises ethical concerns about representation in media. 

• IP/copyright. Users of Midjourney do not retain ownership rights over the images they generate, raising concerns about copyright infringement, especially since the AI was trained on a huge dataset that includes copyrighted material. Users may unknowingly create images that violate copyright laws, leading to legal repercussion. The ethical implications of using an AI trained on artists' works without permission further exacerbates this issue. 

• Privacy. Images created using Midjourney are publicly accessible, even if generated in private settings like Discord, meaning users can inadvertently expose sensitive or proprietary work, as anyone can search for prompts or usernames to access a user's entire image history.

• Offensive content. Midjourney can be misused to generate harmful content, such as depictions of violence or explicit material. 

• Misinformation. Midjourney can be easily manipulated to create misleading or harmful images, including those that propagate racist or conspiratorial themes. This capability allows users to generate images that may violate the platform's guidelines, contributing to the spread of misinformation online.

• Loss of creative skills. There is a risk of users becoming overly reliant on Midjourney for creative tasks, potentially stifling their own creativity and skills. 

• Employment. The more that users become reliant on Midjourney, the more likely it will result in the loss of creative jobs.

• Liability. Recent changes to Midjourney's Terms of Service indicate that users are responsible for any harm resulting from their use of the platform. This shift places the onus on individuals and businesses to navigate legal complexities associated with AI-generated content without support from the provider.

Incidents and issues 🔥

• November 2024. Man's use of AI to "resurrect" grandmother stirs controversy

• November 2024. Midjourney refuses to create images of Black African doctors treating white kids

• July 2024. Images of Australian children are used to train major AI models

• March 2024. Study: Top AI image generators 'easily' produce misleading election photos

• February 2024. Peer-reviewed journal publishes AI-generated rat penis

• December 2023. Midjourney v6 reproduces copyright-protected film images

• November 2023. Text-to-image AI models tricked into generating violent, nude images

• November 2023. Generating an AI image consumes as much energy as charging a smartphone

• September 2023. Synthetic Tiananmen tank man degrades Google Search

• July 2023. AI-generated Barbies reinforce racist stereotyping

• July 2023. AI image generators accept 85% of election manipulation prompts

• July 2023. Research on the visualisation of journalistic roles indicated Midjourney-generated images are biased

• May 2023. Amnesty uses fake AI images to promote Colombia national strike

• April 2024. Reddit warns AI companies not to misuse its data

• March 2023. Deepfake image of Pope Francis clad in a Belanciaga puffer jacket goes viral

• March 2023. Deepfake Donald Trump 'arrest' photos go viral

• February 2023. Vermeer 'Girl with a Pearl Earring' AI facsimile causes controversy

• February 2023. Midjourney attempted to block porn by banning words about the human reproductive system

• December 2022. The San Francisco Ballet was accused of IP theft after it had used a Midjourney-generated image to promote its annual Nutcracker production

• December 2022. Ammaar Reeshi drew the ire of artists and illustrators for creating childrens book Alice and Sparkle over a single weekend using Midjourney

• October 2022. US rules Zarya of the Dawn AI images are not copyrightable

• September 2022. Artist's private medical image used to train LAION-5B dataset

• August 2022. Artists lamented Jason Allen's first place at a Colorado fine art competition for accelerating the death of creative jobs

• August 2022. Atlantic journalist Charlie Warzel's use of a Midjourney-created image of InfoWars provocateur Alex Jones prompted people to accuse the magazine of eliminating the need for illustrators and cutting costs

• February 2022. Database of 16,000+ artists used to train Midjourney

Legal, regulatory 👩🏼‍⚖️

• Andersen et al v. Stability AI Ltd. et al (pdf)

Research, advocacy 🧮

• Stanford HAI (2023). 2023 AI Index Report - 3.2 AI Incidents (pdf)

• Thomas R.J., Thomson T.J. (2023). What Does a Journalist Look like? Visualizing Journalistic Roles through AI

• Wolfe R., Yang Y., Howe B., Aylin Caliskan A. (2022). Contrastive Language-Vision AI Models Pretrained on Web-Scraped Multimodal Data Exhibit Sexual Objectification Bias (pdf)

News, commentary, analysis 🗞️

• https://www.eweek.com/artificial-intelligence/midjourney-review/

• https://www.techradar.com/pro/software-services/midjourney-ai-image-generator-review

• https://www.sciencefocus.com/future-technology/midjourney