FEATURE
Visual Literacy in an AI-Driven World
by Lesley S. J. Farmer
Images constitute an increasing percentage of digital content, and that content is being disseminated globally almost instantaneously. These visual messages are generated by an increasingly broad spectrum of creators for a variety of purposes, not all of which are accurate or ethical. Particularly because people tend to believe visual information more than textual information these days, they might not realize how misleading some of those images can be. As a result, they may make bad decisions, and their actions may result in dire consequences such as political violence. AI, with its fabricated images, exacerbates this issue. More than ever, people need to be visually literate.
For their part, librarians and information professionals strive to provide literacy and intellectual access to information in all recorded formats. This mission expanded exponentially with the advent of the internet. Each new format brings challenges, but librarians are able to discern ways to evaluate the associated content and use the materials meaningfully and responsibly. These skills are needed more than ever to deal with AI, particularly new generative AI (GenAI) tools that can produce realistic text and images. At the same time, these tools—and society’s heightened interest in them—can be leveraged to increase information literacy instruction and practice.
In terms of AI, the majority of discussions have focused on AI-generated text; AI-generated images have been addressed less frequently. On the creators’ end, visual literacy about AI-generated images tends to concentrate on copyright issues. On the consumer end, the main focus has been to determine if the image is AI-generated. However, these issues encompass just a small aspect of visual literacy.
At a time when visual images constitute a growing part of information, librarians can leverage AI-generated images to foster and support visual literacy, which is a subset of information literacy. In the process, not only can generic visual literacy skills be instilled, but those unique characteristics of AI-generated images can be highlighted to deepen visual literacy understanding. This understanding can lead to informed visual creation via AI.
The rest of this article discusses visual literacy in light of AI-generated images and suggests strategies for empowering users to apply visual literacy to consume and produce such images in a knowledgeable and responsible way.
VISUAL LITERACY BACKGROUND
Fundamentally, visual literacy consists of the ability to locate, evaluate, understand, use, communicate, and generate visual images effectively and responsibly. In 2022, the ACRL developed a framework for visual literacy in higher education that provides knowledge, skills, and attitudinal indicators of visual literacy within the changing visual information landscape (ala.org/acrl/sites/ala.org.acrl/files/content/standards/Framework_Companion_Visual_Literacy.pdf).
Visual images use a set of visual elements—line, shape, form, space, color, value, and texture—that create a composition drawing upon the artistic principles of balance, symmetry, proportion, rhythm, movement, emphasis, harmony, and unity. Creators leverage these elements and principles to convey meaning. For instance, dark colors usually convey seriousness, while pastel colors convey innocence or newness. Significant objects are placed centrally or are larger than less important ones. Horizontal lines and symmetrical compositions typically evoke calmness or status quo, in contrast to diagonal lines and disproportionate compositions. These qualities are often unconsciously sensed by the viewer, but visually literate viewers can appreciate these decisions more analytically. They can discern the use of visual elements more critically, particularly when the creators are using them to manipulate the viewer. Think propaganda visual messages
Culture also impacts the use of these elements and principles. Colors often have culturally defined connotations. For instance, Vietnamese brides traditionally wear red, while Westerners tend to wear white. But in medieval times, green was a favored bridal gown color. Yellow can connote royalty, joy, evil, or cowardice, depending on the culture. Likewise, animals symbolize different meanings in different cultures: Owls may be wise or evil; snakes may be killers or healers; foxes may be wise or wily. Particularly as images are seen globally, viewers need to understand culturally defined visual symbolism in order to interpret those images accurately. Likewise, different cultures apply artistic principles differently. Here is another example: In some cultures, distance is indicated by placing the object higher up in the composition; in other cultures, relative distance is indicated by showing far away figures as smaller, although they can be on the same horizontal line.
UNDERSTANDING VISUAL IMAGE MESSAGES
Most viewers concentrate on trying to better understand the meaning the visual image conveys and what its message might be. Here are guiding steps to use in the context of visual literacy:
- Observe. What do you see? What do you notice first? What people and things are shown? How are they arranged? What is the setting? What is happening?
- Reflect. What words come to mind when you see the image? What is the visual point of view? What do you think the message is, or does it have one? What did
you learn by looking at this image?
- Style. How was the image made? How do the visual elements and principles evoke the message? How
does the medium—drawing, watercolor, oil paint, collage, photograph, sculpture—impact or “shape”
the message? If text accompanies the image, how does that affect your analysis?
- Reaction. After analyzing the image, how do you react to it? If it moves you to act, what would that action be?
Visual literacy also involves examining the context of the image. In that respect, visual literacy overlaps media literacy. The Center for Media Literacy (medialit.org) identified five constructs that viewers should consider when processing a visual message:
- Who created the visual message? Is that creator authoritative or reputable?
- Why was the visual message communicated? Is the creator trying to influence the viewer?
- What values and points of view are represented or omitted?
- How do the visual medium and techniques attract and engage the viewer? Does it evoke strong emotions? Does the technique feel manipulative?
- How might the message be interpreted by different viewers?
These questions are very relevant to AI-generated images, and they point out the complexity of authority.
APPLYING VISUAL LITERACY
TO AI-GENERATED IMAGES
All of the visual literacy aspects apply when viewing AI-generated images. Additional dimensions also need consideration for full analysis.
First of all, the process by which the images are created is a key factor. AI image generators collect millions of existing images, apply metadata or “tags” to each image, and create huge image datasets. Humans are the ones gathering and tagging those images, so the quality and representation of those images may be suspect. For instance, people of color are underrepresented, and those image that are collected often do not represent the spectrum of those groups. Stereotypes are likely to emerge (objectified Latinas, gangsta Black males). The adage “garbage in, garbage out” applies. As a result, AI image generators can perpetuate racial biases and, at the least, generate inaccurate and misleading images. An unfortunate reality is that the original source and provenance of the images are seldom identified, which brings up the importance of attribution and intellectual property: This is another visual literacy skill that resides under the information literacy “umbrella.”
Visual literacy in the realm of AI-generated images needs to include this knowledge in order to recognize these biases. That same skill is also needed by the person who prompts the AI image generator to “create” the desired image. In some cases, the prompters intentionally want to mislead the targeted viewers or influence them in a certain direction. In other cases, the prompters may not have the visual literacy knowledge to recognize the biases embedded in the image. In the latter situation, the prompters may find themselves making an embarrassing or cultural faux pas.
It should be noted that someone wanting to have an image created may be considered more a specifier than an original creator. That person provides the textual prompt for the AI algorithm to match the metadata associated with possible images and then synthesize it to generate the closest overall image match. This prompt engineering (the process of structuring text to describe a task for AI tools to perform) largely resembles the employing of good keyword searching strategies, a skill that is easily transferred to general information literacy.
DETECTING POSSIBLE AI FAILURES
While the AI tool tries to reconcile the different images that reflect the prompt, it does not always succeed. Human visual acuity is especially sharp when it comes to analyzing faces because it is a biological survival strategy. (Is that person dangerous? Is that person healthy?) Thus, details such as eyes, hair texture, skin condition, and facial gestures are scrutinized closely for reliability. When the hair texture is too even, when the face looks too symmetrical, when the skin looks “plastic,” or when the light reflection differs between two eyes, the viewer can usually detect AI-generated images. Hands are another giveaway. If they look distorted or blocky, or if there are too many or too few fingers, the image may be AI-generated. An Aug. 30, 2023, article by Chandra Steele, “How to Detect AI-Created Images,” in PC Magazine, provides some good tips to use (pcmag.com/how-to/how-to-detect-ai-created-images).
The notion of the uncanny valley, coined by robotics professor Masahiro Mori in the 1970s, describes the almost visceral feeling of eeriness or unease people feel when seeing a humanoid-like robot. This is also a hint that an image might be AI-generated. When a person’s image looks too perfect or slightly “off” to be real, that’s likely an AI fabrication. (It is also why engineers avoid making robots look too human.) Creating deepfakes is now possible using AI image generators, which further challenges “uncanny valley” limitations. (You can see examples in Daniel Victor’s March 10, 2021, New York Times’ article: nytimes.com/2021/03/10/technology/ancestor-deepfake-tom-cruise.html.)
Scientific knowledge also helps to spot AI-generated images, as well as images that have been manually edited and manipulated using digital image editing tools such as Photoshop and Maya. Contradictory light sources, impossible feats of gravity (holding a cow as if it weighed 5 lbs.), absurd biological displacements (sharks in swimming pools), and implausible weather patterns (multiple parallel tornado spouts) all provide evidence of some kind of image manipulation. It could stem from the AI drawing upon fake images or inaccurately synthesizing multiple images. The BBC’s June 9, 2020, article by Tiffany Wen on identifying fake images provides some good guidelines that can be applied to AI-generated images (“The Hidden Signs That Can Reveal a Fake Photo”; bbc.com/future/article/20170629-the-hidden-signs-that-can-reveal-if-a-photo-is-fake). In any case, by understanding the process of AI generation and manipulation, viewers gain a new visual literacy skill that is medium-specific.
AI image generators are already impacting how visual artists work. To a degree, these artists may feel conflicted. On the one hand, they sometimes find that prompting an AI tool to generate a defined image can jump-start their imagination, just as an image source book or an online image search can spark an idea. On the other hand, artists sometimes find their own published work being used as source materials for the same AI tool, without giving attribution, let alone remuneration, for the use of that image. Artists sometimes try to trace back the image to its original dissemination, which is a good practice for all AI users. Several online tools serve this function with more or less success. In a Nov. 9, 2023, article Ben Beck suggests several reverse-image searching tools (“The Top 7 Reverse Image Search Tools and How to Use Them”; clearvoice.com/resources/reverse-image-search-tools).
Because visual literacy also addresses the ethics of images, recognition of this issue and reflection upon its impact on artistic creating constitute an important part of visual literacy education. Jörg M. Colberg’s March 27, 2023, thoughts on AI images, in Conscientious Photography magazine, address what he calls “very convincing nonsense” in the context of visual literacy (“Thoughts on AI Images: Art, Very Convincing Nonsense, and Visual Literacy”; cphmag.com/thoughts-ai).
VISUAL LITERACY EDUCATION
Society—business, education, and legislation—is still grappling with ways to deal with AI, though less so with images than with text. Visual literacy has already been undervalued within educational curricula and more so within the context of AI. Nevertheless, the populace is busy experimenting with these AI image-generator tools. So as information professionals, librarians need to get ahead of the game and provide physical and intellectual access to AI image generators in terms of the critical consumption of AI-generated images as well as to use these tools to generate their own images. More generically, librarians can leverage interest in AI to promote visual literacy.
What, then, are ways to leverage AI-generated images to foster visual literacy? Here are several starting strategies.
- Research how algorithms and GenAI image generators are developed.
- Track the evolution of AI image generators.
- Critique AI-generated images.
- Trace AI-generated images and their citations (which
are often bogus).
- Test the same AI prompt to see if different images emerge.
- Examine AI-generated images in terms of bias.
- Show how transforming information from test to image impacts the information, exemplifying transliteracy.
- Discuss the legal and ethical concerns that AI art generates, including analyzing case studies where AI-generated image use led to cheating.
- Bring in visual artists to talk about their use of AI
image generators.
GENERATE YOUR OWN IMAGES
One effective way to gain visual literacy skills through AI is to generate those AI images. Such practice enables users to see how AI image generators work, how prompt engineering works, and how images convey accurate or biased information. In a Tech and Learning article, Diana Restifo identifies several free AI image generators for users to test (“Best Free AI Generators”; techlearning.com/news/best-free-ai-image-generators-for-teachers).
Some ways to leverage visual literacy to generate AI images include employing color theory, exploring typography, leveraging visual composition, taking advantage of symbols and symbolism, considering target audiences, being culturally sensitive, and complying with intellectual property rights.
AI-generated images and AI image generators offer a unique opportunity to rethink information literacy and to align it with today’s technology. Indeed, this generation of students has a once-in-a-lifetime opportunity to experience a new set of technology tools from the ground up. How exciting!
|
