The Library will be closed on Monday, May 27, 2024, in observance of Memorial Day.

Images, Misinformation, and Sherlock Holmes

Andrea Borchert, Librarian, Koreatown Media Lab,
Illustration collage of images relating to post

Images, Misinformation, and Sherlock Holmes: misinformation is designed to mislead, but with a couple of tools, you too can have the confidence of a fictional detective.

sir Arthur Conan Doyle
Arthur Conan Doyle Portrait from the Library of Congress Database

As days get shorter and Winter grinds into gear, take a moment to consider misinformation and social media, just in time for the Holidays! Information, especially photographs, can spread quickly online, and there is no method that is 100% effective at distinguishing information from misinformation. Even Sir Arthur Conan Doyle, medical doctor and author of the Sherlock Holmes Mysteries, has fallen for misinformation. Doyle believed two girls had taken photographs of fairies. He believed this so much he published a book on it, The Coming of The Fairies. But they hadn't photographed fairies. They’d cut out pictures of fairies and photographed themselves with the cutouts. They managed to mislead the author of some of the greatest detective stories ever, and they didn’t even have a green screen! (We do have green screens, in the Octavia and Koreatown Media Labs, but you can trust us). Elsie Wright and Frances Griffiths aren’t the only people to use camera tricks to photograph themselves with fairies. The TESSA database has several fairy pictures of pageant queens posing with fairies.

Elsie Wright
The "Cottingley Fairies" with Frances Griffiths. Image from page 49 of The Coming of the Fairies, from The Gutenberg Project
Frances Griffith
Trick photo showing Rose Queen Dorothy Edwards, and a miniature image of Dorothy Edwards, dressed as a fairy with wings and a wand, standing atop a rose held by one of the court princesses, [1933]. Eyre Powell Chamber of Commerce Collection

In an effort to be less susceptible to fairies, more like Sherlock Holmes when looking at information, and less like Arthur Conan Doyle, there are tools and methods you can use to check images online for reliability. When looking at images presented by news sources, you can always consider the reliability of that news source; do they cite their sources, are other news companies talking about the topic, what are the fact checkers saying, and when were these articles published? But, with social media, images can spread quickly without being verified by the news media. There are two basic types of image misinformation on social media: images that are taken out of context and altered images.

big black cat on Lankersheim Blvd
This is an altered image from 1955, found in the TESSA database. To be taken out of context, we could publish it as a photo taken this year, in San Pedro.
Cut-out photograph of black cat is composited onto photograph of street background, [1955]. Valley Times Collection
google gif
Gif made from a screenshot of a Google Search bar with an arrow pointing at the image search

One of the first things to do to check on images you have questions about is a reverse image search. With a reverse image search, you upload an image or enter a link to an image, and the search engine looks for other examples of that image. There are lots of tools to do this, including Google Lens. To do a reverse image search in Google, in the Google Search bar, click on the little camera icon to the right to do an image search. If we do a Google Lens search for that photo from the library’s TESSA Database of a black cat towering over Hollywood, we’ll get multiple results. One of the first results is a link to Calisphere that leads us to the image’s information metadata and links back to the TESSA database. (Check out Calisphere.org. It's a super cool and reliable database for California-specific historical information! We can call databases "super cool" we work in a library.)

Screenshot of Google Lens
Screenshot of Google Lens

Anytime you find yourself searching Google Images, find the 3 vertical dots to the right of the results, this is the Kabab. Clicking on the Kabab reveals more options that help to research images. Let’s say there’s been a rain, and someone posts that they saw a shark on the freeway. Search for "freeway shark," click on the image you are researching, and find the 3 dots, the kabab, on the upper right of the image.

Gif of image search results with a red arrow pointing towards the Kabab
Gif of image search results with a red arrow pointing towards the Kabab
Screenshot of the About this image tool from Google
Screenshot of the About this image tool from Google

Clicking on the kabab makes more tools appear. One of the tools is About This Image. If you click on About this image, it will give you more information about when the image was posted. For example, the version of the picture of the shark on the Freeway that we clicked on is at least 5 years old. It is, thus, not from the recent storm.

Screenshot of the results from clicking on About this image while doing an image search for “Hurricane Shark”
Screenshot of the results from clicking on About this image while doing an image search for “Hurricane Shark”

Google’s About This Image tool helps users find out more about an image, but About This Image is not 100% accurate. There are many other reverse image search tools researchers can check. TinEye.com, for example, is a well-regarded search tool that will sort results by Newest, Oldest, Biggest, and Most Changed. Yandex is a Russian-based search engine that makes it easy to sort results by type of file.

Reverse image searches can be tremendously important, especially when you want to figure out when an image started to become popular. But if you want to figure out how a photo was made and if it was altered, there are other tools to turn to AI, for example. AI, Artificial Intelligence, can generate new, photorealistic images quickly. AI generally works by gathering information from thousands of images and stitching them together based on its algorithms and the prompts it receives. There is a tool, AIorNot.com, that purports to tell the difference between AI-generated images and photographs. It is not 100% effective either.

Ai or Not’s judgment of Théâtre D’opéra Spatial, an AI-generated work that won the 2022  Colorado State Fair's annual fine art competition.
Ai or Not’s Judgment of Théâtre D’opéra Spatial, an AI-generated work that won the 2022 Colorado State Fair's annual fine art competition

A reverse image search may also help uncover evidence that an image is AI. In a reverse image search on TinEye.com, the first few results mention that Théâtre D’opéra Spatial is an AI image that won an art contest. However, that level of interest may not be available for every potential AI image.

If you are interested in putting on a deerstalker hat and looking closely at an image to figure out if it's AI, there are methods you can use. Just like AI or Not, they are not 100% effective. The better you can see an image, the more likely you will be able to spot signs of AI-generated details. So, if you are trying to figure out if an image is AI by looking closely at it, look for the largest version of the image you can find and zoom in.

Image made in Adobe Firefox using the prompt “A human wearing glasses and earrings, facing front. Notice that the background is blurry. The collar is different on different sides of the figure’s body. The earrings don’t match, and the necklace doesn’t connect.
Image made in Adobe Firefox using the prompt “A human wearing glasses and earrings, facing front. Notice that the background is blurry. The collar is different on different sides of the figure’s body. The earrings don’t match, and the necklace doesn’t connect

AI-generated images tend to have blurry backgrounds. AI-generated images of human faces have trouble with symmetry. Eyes, earrings, and clothing are likely to be mismatched. Shirt collars, for example, may be different on different sides of a person’s neck. Details may melt into and out of existence in odd ways. Glasses may not connect with both sides of a head or may have different frames on the right and left sides of the person’s head. Hands and teeth may have the wrong number of parts, fit oddly into the space, or be out of proportion with other aspects of the image. AI-generated images may include watermarks, although those can be cropped out. It is likely that the flaws in AI-generated images will improve over time, but it is also likely that the methods viewers use to distinguish AI images from photographs will improve too. The game is afoot?

To really feel like Sherlock Holmes, take a look at photo metadata! Metadata is information about information, in this case, information about a photograph. If you take a photo with your phone, your phone will automatically record information about the photo, like the type of phone it was taken on, the time it was taken, and, potentially, the location where the shot was taken. This information is automatically embedded in the photograph and is viewable with the right tools. There are many tools that help make metadata visible, for example, Metadata 2Go.com. You can also open files on your computer to find their metadata.

Photograph of Pio Pico taken from a cell phone
Photograph of Pio Pico taken from a cell phone
Screenshot of the metadata found on the photograph of Pio Pico Library
Screenshot of the metadata found on the photograph of Pio Pico Library

The amount of metadata attached to images varies. Images on Instagram, for example, have been stripped of metadata to save bandwidth. Metadata can also be edited. So photographic metadata cannot always answer all the questions an image detective has. One piece of information that is supposedly included in photo metadata, but is usually unavailable, is the location where a photo was taken. Geolocating is a process of looking through images closely to figure out where they might have been taken. Some of this is very complicated. Experts might look at a photo’s metadata to figure out the time it was taken, then use the angle of shadows in the photo to figure out where an image originated. But some geolocating is easier.

Black Lives Matter protest in Downtown Los Angeles during the COVID-19 pandemic from the TESSA database
Black Lives Matter protest in Downtown Los Angeles during the COVID-19 pandemic from the TESSA database

Looking at this image from TESSA, for example, we can see that the signs and writing on the shirts are in English. Which is a good sign that the location is in an English-speaking country. There’s a sign for a Hospital, so this location is near a hospital. There’s a street sign that says Flower and another that can’t be made out but looks like a low-number street, perhaps 5th St. There are multiple One Way only signs. The building in the back has an unusual, rounded shape and an orange tower with a clock in front of it. So with a little playing around with the image and Google Maps, you can look for a Hospital, find Good Samaritan Hospital, and then find Flower and 9th. If you look closely, you can see the shapes of the same buildings in the background.

Detail from Black Lives Matter protest in Downtown Los Angeles showing the Hospital Sign
Detail from Black Lives Matter protest in Downtown Los Angeles during the COVID-19 pandemic showing the Hospital Sign
Flower Street sign
Detail from Black Lives Matter protest in Downtown Los Angeles during the COVID-19 pandemic showing the Flower Street sign
orange clock tower and interesting roof line
Detail from Black Lives Matter protest in Downtown Los Angeles during the COVID-19 pandemic showing an orange clock tower and interesting roof line
Screenshot from Google Maps of the corner of Flower and 9th
Screenshot from Google Maps of the corner of Flower and 9th

This is a laborious process, and it’s easy to go down the wrong path. But when it works, geolocating can make you shout "Elementary!*" in celebration. Technology changes quickly. It can be frustrating to keep track of what information you can trust. But even as the tools used to create images change, you can always trust the Library to do its best to point you toward reliable sources.

Greenscreens in the Octavia and Koreatown Media Labs are available for use.
Greenscreens in the Octavia and Koreatown Media Labs are available for use
A step in the process of Green Screening fairies into the library
A step in the process of Green Screening fairies into the library
A step in the process of Green Screening fairies into the library
Fairies in the Library!...wink

You can trust us. After all, the fairies do.

*Sherlock Holmes never said this in the books. But don’t let facts get in the way of a celebratory exclamation.


 

 

 

Top