A photo that claims to show an explosion near the Pentagon has raised concerns about AI’s ability to produce false information.
A fake image of what appeared to be a massive explosion near the Pentagon on Monday was shared on social media, sparking a temporary drop in the stock market.
Within minutes, a wave of social media accounts, including some verified accounts, shared fake photos, adding to the confusion.
Authorities have since confirmed that no such incident occurred.
Social Media Detectives Including Nick Waters Belling CatAn online news verification group, quickly pointed out some notable issues with images such as:
- That there were no other direct eyewitnesses to corroborate the event, especially in a congested area like the Pentagon. “This is why it’s so hard (I’d argue virtually impossible) to create a credible faux pas of events like this,” Waters tweeted.
- That the building itself looks noticeably different from the Pentagon. This he can easily see by comparing the two images using a tool such as Google Street View.
- Other details, such as odd-looking floating lampposts and black poles sticking out of the sidewalk, also revealed that the image was not what it seemed. Artificial intelligence still has difficulty recreating locations without introducing random artifacts.
I’m pretty sure this photo that claims to show an “explosion near the pentagon” was generated by AI.
Check out the frontage of the building and how the fence blends into the crowd fence. No other images, videos or people have posted as direct witnesses. pic.twitter.com/t1YKQabuNL
— Nick Waters (@N_Waters89) May 22, 2023
How to distinguish AI-generated images from fake images
There are many generative AI tools such as Midjourney, Dall-e 2 and Stable Diffusion that can create lifelike images with little effort.
These tools are trained by looking at large amounts of real images, but fill in gaps with their own interpretations when training data is missing. As a result, people have extra limbs and objects that can deform to fit their surroundings.
When viewing images online that purport to show breaking news, you should be aware of the following:
- News does not happen in isolation – In the event of a large-scale explosion or large-scale event, you can expect a flood of reports from the scene from many different people and many different angles.
- Who is uploading content – Let’s take a look at the post history of a user account. Do their locations match the location of the event? See who they are following and who is following them. Is it possible to contact or talk to them?
- Use open source intelligence tools – Reverse image search tools such as Google Images and TinEye allow you to upload an image and identify when and where it was first used. There are some other tools you can use, like watching live public transport camera footage to make sure an event is happening.
- Analyze the image and its surroundings – Look for clues such as nearby landmarks, road signs, and even weather conditions in the imagery to help identify when and where the event occurred.
- hands, eyes, posture – Pay special attention to the eyes, hands, and general posture when working with images of people. His AI-generated videos that mimic humans, known as deepfakes, are prone to blinking problems because most training his datasets do not include faces with closed eyes. Hands that don’t grip objects properly or limbs that look unnaturally twisted can also help spot fakes.
For more information on news verification and open source intelligence investigations, Al Jazeera’s Media Institute has published several guidebooks in multiple languages, which can be downloaded below.
- find the truth in the fake [PDF]
- News Validation – A Practical Guide [PDF]
- open source research [PDF]
#thread🧵
1/ Spotlighting the dangers of deepfakes: What happened yesterday when a blue check fake Twitter account claiming to be Bloomberg released a shocking “Breaking News”.— Al Jazeera Journalism Review (@AJR_English) May 23, 2023
