Using Artificial Intelligence in Archival Practices

The main goal of this portion of the project was the creation of metadata (descriptions, subjects, locations, and alt text) by using a generative AI program. The reason for doing this was to see if we could speed up process of description creation in the archival process. We digitized a total of 70 pictures from the Special Collections Archive at the University of Mary Washington to give to the AI Although the program did not exactly create the descriptions as prompted, it was still able to generate the metadata for locations and subjects, as well as alt text for each image. While this is not an infallible way for description creation, and does not fully replace a human, it did make this step of the process quicker and helped to streamline the process.

A sign hanging on a Mary Washington building that states “Ease on down the Road with MWC.”
Courtesy of the University of Mary Washington Special Collections, Box 31 – Photo 029

What Should I Be Looking Out For?

The Prompts that Were Used

The only thing that the program was asked was to create a description of the photo, specifically with the prompt “Can you give me a description of this photo?” along with an attachment of each respective photo. Any additional information that was generated was done because the AI decided to include it on its own. It is worth noting that a new chat was created every 10 photos, as to not overwhelm the program. The descriptions that were created were closer to alt text, which is a more visual-based description that is used for those that are visually impaired and use screen readers. The longer bullet-pointed sections are better for metadata usage, as they are much more comprehensive.

The Labels

The name of each picture is comprised of two numbers: the “Box” number and the “Photo” number. The “Box” number refers to the box in which the photo was taken from in the Special Collections Archive at the University of Mary Washington. There are two of them; 31 and 32. All of the photos in these two boxes were undigitized and personally scanned by our team for the purpose of this project. The photo number was given to each picture in order to distinguish them from each other as well as count how many total photographs were in each box. There were 36 photos taken from Box 31 and 34 from Box 32.

Information in the Text

Pay attention to the text! Any information that is inaccurate will be highlighted. Not all of the descriptions are perfect. In fact, there were even some instances where the program provided little to no detail at all. If these descriptions were to actually be used in real metadata, they would need to be heavily edited.

Visual Details

Look at each of the photos and compare them to their respective descriptions. Pay close attention to what information is accurate versus what is not.

Successes

Failures

Facial Recognition

One very clear ability that the AI was able to accomplish was recognizing specific people. It was able to make connections between several photographs that were of the same event and recognize the same person who appears across several pictures, without being asked.

Lack of Details

The majority of the problems that we had with the AI were due to it not generating a description, only giving the information on a specific subject, or in a severe case giving no information when prompted. This could be fixed by further prompting or changes in the original prompt given, however to keep this experiment controlled and accurate there was no further prompts or changes to the original text.

Organizing and Expanding on Information

Although it was only prompted to create descriptions for each photograph, it generated both a description (or alt text) as well as more details about the subjects or setting. The program even tried to provide names for the exact location (such as the name of a field or dorm hall) — some of these were accurate, while others were not.

Incorrect Information

When the program did not know specific details about a certain event or subject, it appeared to just provide false information based on a guess. For example, in Photo 003 from Box 31, the AI said that it was taken after the major snowstorm that happened in 1983, but these photos are not related to that event at all. It seems as though it assumed this due to the fact that the people were wearing puffy jackets. Although this seemed to be primarily caused by the quality and the light sources in the photos. To ensure that these failures do not happen next time, the main issues would be reviewed and potentially redone by the archivist or person trying to create the descriptions or metadata.

See a Sample of the Descriptions and Metadata Below!

The following photographs were chosen to demonstrate a few of the various successful descriptions that the AI was able to generate, such as being able to identify specific locations or tracking the same person throughout multiple photographs. Other pictures were chosen for this sample to show that the AI is not perfect, meaning it will make mistakes like falsely identifying the time period, labeling the school as UMW rather than MWC, or just not generating a description at all. Some additional photos were also chosen purely for aesthetic purposes. The full collection is available to be explored and can be found at the bottom of the page.

Box 31 – Photo 001

Photo Description / Alt Text
The image is a classic square-format color photograph featuring five individuals—two women and three men—posing together on a grassy campus area. Based on the filename and the attire, this likely dates back to the early 1980s, capturing a moment of campus life or a departmental gathering.

Standard Metadata

The Subjects: The group is dressed in quintessential late-70s to early-80s outdoor wear. Notably, one individual is wearing a U.S. Army field jacket, while others are in quilted vests and tan utility jackets, suggesting a cool but sunny day.
The Setting: In the background, you can see a large red brick building with white columns, a signature architectural style of the UMW campus. The bare trees and lighting suggest either late fall or early spring.

Box 31 – Photo 003

Photo Description / Alt Text
This photograph offers a candid glimpse into student life at the University of Mary Washington during the early 1980s, likely captured during the recovery efforts following the major snowstorm of February 1983.

Standard Metadata

The image features five students—four women and one man—working together to secure or dismantle a large, heavy-duty canvas tent. The students are dressed in typical early 80s winter attire, including:

Puffy “members only” style jackets and parkas.
High-waisted denim jeans.
The student on the far right is wearing a bright orange reflective safety vest over a blue jacket, suggesting a coordinated campus effort.

Box 31 – Photo 005

Photo Description / Alt Text
This is a large group portrait featuring approximately 16 young adults—likely Resident Assistants (RAs) or student leaders—posing in front of a large, military-style canvas tent. The scale of the tent and the number of participants suggest a significant campus-wide collaboration or a specialized training day.

Standard Metadata

The Group: The mix of students includes several individuals seen in the previous photos. Notably, the young man in the U.S. Army field jacket is visible in the back row, and the woman in the green and navy blue puffer jacket is kneeling front and center.
Identification Markers: Several students are wearing bright red-and-white striped vests. These may have been “staff” vests used to identify student leaders during the event, or they could be related to a specific safety demonstration (like CPR or first aid training) often hosted by Reslife.

Box 31 – Photo 006

Photo Description / Alt Text
This is a medium group shot of nine individuals posing outdoors. The lighting is bright and direct, typical of a crisp, sunny day. The group is positioned in front of the large military-style tent seen in photo 005, with a prominent red brick campus building visible in the background.

Standard Metadata

This image is particularly interesting because it captures the primary “characters” we’ve seen throughout this archival set:

The Center Figures: The woman in the green and navy blue puffer jacket and the young man in the U.S. Army field jacket are positioned at the heart of the group.
The Bookends: On the far right is the man in the heavy navy blue puffer coat, and on the far left (second from the edge) is the woman in the blue and grey vest, both of whom were featured in the very first photo of this series (001).

Box 31 – Photo 010

Photo Description / Alt Text
The image features three individuals—one man and two women—sitting together at a table draped in a white cloth, likely in front of a campus building.

Standard Metadata

The Subjects: On the left, a young man in a plaid button-down shirt smiles for the camera.
- In the center, a young woman in a floral print dress with a bright pink waist sash has her arms around her colleagues.
- On the right, a woman in a red-and-white collared shirt and glasses completes the group.
- All three are wearing blue circular buttons that appear to read “WE’RE GLAD YOU’RE HERE,” a traditional welcoming sentiment for new students and families.
The Setting: They are positioned in front of a red brick building with large white columns, which is a hallmark of the UMW campus architecture. The presence of a cardboard box filled with folders or files on the table suggests they are processing registrations, distributing keys, or handing out orientation materials.
The Fashion: The hairstyles and clothing—particularly the center woman’s sash and the woman on the right’s large-framed glasses—are very consistent with the early-to-mid 1980s.

Box 31 – Photo 011

Photo Description / Alt Text
The image features three young women standing close together in front of a prominent brick building with large white columns and multiple sets of double doors.

Standard Metadata

The Subjects: On the left, a woman wears a dark, pinstriped dress with puffed sleeves. She has a name tag and a blue circular button that reads, “WE’RE GLAD YOU’RE HERE.” * In the center, a woman in a light pink, high-collared dress smiles playfully at the camera, also wearing the same blue “welcome” button.
- On the right, a woman in a horizontal and vertical striped dress is captured mid-laugh, leaning slightly toward the others.
The Setting: They are positioned on a green-carpeted area or walkway. In the background, a man in a white short-sleeved shirt and tie stands near the building’s entrance. Above the group, a partial sign is visible that appears to say “HERE!”

Box 31 – Photo 017

Photo Description / Alt Text
The photo features a group of people gathered on wide concrete steps, characterized by the same large white columns and red brickwork typical of the Mary Washington architecture.

Standard Metadata

The Subjects: In the center-left, a woman in a bright red dress stands next to a wooden pole topped with colorful balloons (yellow, blue, and red). A white sign on the pole reads “Brah-Z” in bold black lettering.
- Behind her, another woman in a brown plaid skirt and a white-collared blouse looks toward the camera with a smile.
- Several other individuals are seated on the steps in the background, dressed in casual summer attire like striped pants, shorts, and t-shirts. One person in the far background is wearing the same blue “welcome” button seen in the first photo of this set.
The Signage: The “Brah-Z” sign is a particularly interesting detail. In the context of Mary Washington’s history, this likely refers to a specific student group, a residential floor designation, or a playful team name for orientation week.
The Atmosphere: The balloons and the organized seating suggest a designated “meeting point” for incoming students to find their groups or leaders during move-in festivities.

Box 31 – Photo 022

Photo Description / Alt Text
This photograph depicts a lively and whimsical spring scene on a college campus, likely from the early-to-mid 1980s. It captures a blend of student recreation and a lighthearted campus tradition or event.

Standard Metadata

The Foreground: The most striking element is a car parked on the grass, completely transformed to look like a giant white mouse. It features large pink-and-white ears attached to the sides, oversized eyes on the windshield, and a dark “nose” on the hood.
The Subjects: To the left of the car, two young men are active on the lawn. One is in mid-motion playing with a soccer ball, dressed in a white polo and athletic shorts with striped tube socks. The other stands nearby in a red t-shirt and yellow shorts, watching the activity.
The Setting: The scene takes place on a large, grassy quad. In the background, a stately three-story red-brick building stands with white columns and a prominent balcony, characteristic of neoclassical campus architecture. To the left, large trees are beginning to bud, suggesting early spring.

Box 31 – Photo 024

Photo Description / Alt Text
This photograph captures a classic 1980s campus fundraiser or spring festival event, characterized by humor and student involvement.

Standard Metadata

The Main Event: Two young men are participating in a “pie-throwing” booth. They are seated behind a large wooden board with their faces peeking through cutouts. Both are covered in whipped cream, smiling broadly at the camera.
The Setup: A long folding table in front of them is decorated with blue and white crepe paper streamers. On the table sits a red-and-white cooler, brown paper bags, and a stray paper plate.
The Signage: A large hand-drawn sign is taped to the front of the table. It reads “WESTMORELAND’S PIE THROWING 50¢” in colorful block lettering, identifying the event as being hosted by Westmoreland Hall residents.
The Background: The scene is set on a grassy quad surrounded by large, mature trees. In the distance, other students can be seen walking near a traditional multi-story brick residence hall.

Box 31 – Photo 031

Photo Description / Alt Text
This is a vintage landscape photograph, likely from the early 1980s based on the film grain and color palette, depicting a sunrise or sunset over a wide body of water.

Standard Metadata

The program did not generate any additional information.

Box 32 – Photo 003

Photo Description / Alt Text
This ninth photo from Box 32 provides the “Watermelon” half of the “Crab Feasts & Watermelon !!” theme. It features two young men posing with large watermelons in front of the same iconic white columns seen in the previous registration photos.

Standard Metadata

The Duo: Two young men are standing side-by-side, each holding massive, oblong watermelons.
- The man on the left, with blonde hair and a tanned complexion, is wearing a green t-shirt that reads “RYKEN PHYSICAL EDUCATION” and light-wash blue jeans. He is looking off-camera to his right with a broad smile.
- The man on the right has dark hair and is wearing a grey and navy blue long-sleeved sweatshirt and dark blue jeans. He is looking directly at the camera with a cheerful expression, struggling slightly to hold two watermelons at once.
The Setting: They are standing on the stone porch of a building, framed by the deep grooves of two massive classical columns. The lush green trees of a campus quad are visible in the background, along with a brick walkway on the far right.

Box 32 – Photo 031

Photo Description / Alt Text
This final photograph offers a beautiful elevated perspective, likely taken from further up the ridge or from a higher floor of a nearby building. It serves as an establishing shot that ties the entire “Grill on the Hill” event together geographically.

Standard Metadata

In the immediate foreground, the focus is on a green-surfaced outdoor basketball court.

The Game: A casual game is in progress, with a few students in t-shirts and shorts active on the court.
The Perimeter: A chain-link fence separates the court from the surrounding grassy areas. Several students are leaning against the fence or sitting on the edge, watching the game or the larger festival beyond.
The Path: To the left, a brick-paved walkway leads down toward the main event. A steady stream of students is moving along this path, creating a sense of movement and flow toward the festivities.

Box 32 – Photo 033

Photo Description / Alt Text
This final photograph, captures a high-energy “success” moment at the grilling station, taken just seconds after the previous shot. It perfectly encapsulates the camaraderie of the student volunteers running the event.

Standard Metadata

The central focus of the image is a celebratory high-five exchanged across the smoky grills.

The Interaction: A student in a yellow button-down shirt and blue shorts (who was partially obscured by smoke in the last photo) raises his hand to meet the hand of the cook in the striped polo. They are celebrating a perfectly flipped burger or perhaps just the high-energy “flow” of the service.
The Team: The young man in the white “Alexandria Lacrosse” t-shirt on the left looks on with a wide grin, pointing toward the grills, while the student in the white graphic tee on the right continues to manage his station with a Miller Lite in hand.
The Atmosphere: The flames are even more visible here, licking up through the center grill, adding a sense of heat and urgency to the scene. The ground remains covered in white paper patty dividers, a testament to the volume of burgers they’ve already served.

Final Thoughts

Overall, this experiment is just the beginning of how AI can make the archival process easier. All of these descriptions were generated off of only one prompt, so if this were to be used in actuality, AI must continue to be prompted to get the most accurate results. Whenever possible, human intelligence should be used in combination with the AI to ensure the results are more accurate and in the proper context. Sometimes the AI assumes pieces of information based on the provided photos or other contextual elements (such as locations, events, people, jobs, buildings, etc.) which will need to be fact-checked or reviewed by someone who is knowledgeable on the subject . Prompting specifically for metadata and giving it specific standards (such as Dublin Core) will also generate better results.

Want to See More?

Click Through the Full Collection Below