Skip to Main Content

Generative AI at UVA

This guide features links and information about generative AI, including ethical use, citations, considerations for use, and more.

Images, Video, Music, and more...

Generative AI isn't limited to just text. It can produce images, videos, music, and other forms of media. From deepfake videos, which we will include in this category, to AI-generated artwork and music, the use and misuse of this technology in media is growing. In this guide, we’ll explore the types of generative non-text AI, examples of tools you can try, concerns and challenges, recent news, and resources to learn more.

Types and Example Tools:

Deepfakes: Imagine a video of a celebrity or politician saying something they never really did. Deepfakes are tools that can edit videos to make it look like someone is doing or saying something they haven't. Think of it like Photoshop but for videos. 

Text to Image: If you've ever wished for a tool that could turn your words into photographs, illustrations, or digital paintings, this is it. Describe an image in words, and these tools try to create an image that matches. 

Generative Video: Write a short description, or upload a photo, and generative video tools will aim to produce a video clip based on your description or image. 

3D/Animation: Describe an object or a character, or upload a photo, and these tools will craft it into a 3D model. Or make a 3D model move based on your movements. 

Text to Music / Song Composition: Describe a musical composition in words, such as a mood or a theme, and these tools will turn your descriptions into music. 

Voice reproduction (text to speech, speech to speech): Enter some text, or your own recording, and these tools attempt to impersonate someone else. 

Vocal and Instrument Impersonation: These can accurately replicate specific voices or the sound of musical instruments. 

Image upscaling / Increase image resolution tools

Other: there are many other exciting tools out there. 

Many of these categories of tools are now strongly integrated into major software suites (Microsoft Copilot integrated into MS Office, Canva AI suite, Adobe, Google Gemini)

How Generative AI Works with Images:

 

[Image from NYTimes Article: Pavich, Frank. “Opinion | This Film Does Not Exist.” The New York Times, 13 Jan. 2023, https://www.nytimes.com/interactive/2023/01/13/opinion/jodorowsky-dune-ai-tron.html. Accessed 22 Aug. 2023.] 

Concerns and Challenges:

As fascinating as generative AI and deepfake technologies are, they come with a set of concerns. For the average person, these concerns could affect everything from how we consume news to how we trust online content. Here's a closer look at some of the challenges posed by these technologies and potential harm that we all could face: 

  • How Fast This Is All Changing: The sheer speed at which generative AI progresses is daunting. This guide is a start, and will be updated as is possible, but staying up to date is a challenge. 

  • Hyperrealism and Believability: Today’s generative AI produces media indistinguishable from reality. Deepfake videos and synthesized voices can convincingly mimic real individuals, complicating our ability to trust online content. Digital literacy now demands advanced skills in evaluating authenticity. Skepticism and verification strategies must continually evolve alongside technology.

  • Erosion of Trust and the “Reality Crisis”: A proliferation of synthetic media has led to what some call a “reality crisis,” where audiences increasingly doubt legitimate videos, news reports, and even historical documentation. Authenticating content and maintaining public trust has become significantly more difficult, raising concerns about societal polarization and disengagement.

  • Potential for Misinformation and Disinformation: The ability of AI to create compelling media might lead to widespread misinformation (spreading false information without intent) and disinformation (deliberately spreading false information), which can spread like wildfire.  

  • Global Political Concerns: The potential to create fabricated videos or audio clips of world leaders can influence public opinion or spark international incidents.  

  • Fake News: A direct offshoot of misinformation, AI-generated news, if unchecked, could lead to the propagation of biased or utterly false narratives. With the ability to generate believable fake news stories, people might be misled into forming opinions or making decisions based on falsehoods. 

  • Potential for Personal Harm: Deepfakes can be used maliciously to impersonate individuals, leading to personal and professional damages. Personal videos or images could be manipulated to portray individuals in situations they were never in, leading to cyberbullying or reputational damage. 

  • Deepfakes, Misuse, and Criminal Conduct: Deepfakes could be used for blackmail, fraud, or to create false evidence. For instance, students could be wrongly accused of misconduct through fabricated videos. 

  • Criminal Justice Concerns: In legal settings, the reliability of evidence could be questioned if deepfakes or generated photographs become prevalent. The sophistication of fakes raises significant concerns in the justice system, especially when AI-generated content is presented as evidence. 

  • Vocal Reproduction and Scams: Imagine getting a call from a 'family member' asking for emergency funds or personal details, only it's not them but an AI-generated voice. AI's ability to reproduce voices can be weaponized for scams, defaming individuals, or impersonating influential personalities. 

  • Copyright and legal considerations – see tab entitled: “Copyright and Intellectual Property” 

  • Algorithmic Bias and Representation Issues: AI generative tools may perpetuate biases based on training data, resulting in unequal representation or harmful stereotypes, especially regarding marginalized communities. Critical assessment, transparency in datasets, and active measures to diversify AI training materials are now essential to ethical AI deployment.

  • Environmental Impact: The environmental footprint associated with training and running generative AI models is increasingly scrutinized. Popular AI tools require immense computational resources, raising sustainability concerns.

  • Global Regulatory Variance:Regulations on AI vary significantly worldwide, complicating international use and collaboration. Users and creators must stay informed of international laws and anticipate how differing regulations impact the global sharing and use of AI-generated content.

In the News:

Examples of Generated Images and Media:

Resources/To Learn More: