FACT CHECK | After Bato and Baste’s AI shares: How to identify AI-generated content online

Davao City Mayor Sebastian Duterte and Senator Ronald dela Rosa shared a viral post about a supposed interview of youth sentiment on Vice President Sara Duterte’s impeachment proceedings. 

However, the post was FAKE, and supporters of both politicians even tried to advise them that this was created using AI. 

We are fact-checking this and a flood of posts to help netizens determine AI content amid several viral content widely shared online.

AI = artificial intelligence

AI stands for artificial intelligence. Over the years, AI has been applied in several formats, ranging from the automation of text drafts to improving the processing speeds of tasks that would have taken hours if these were done manually, such as manually sorting data across applications. 

It is now possible to draft essays and arrange complex data in seconds, and this is only a few of AI’s applications. 

How does AI create photos and videos? 

Nowadays, AI “learns” how to create by matching instructions, usually text-based, with its database of files. With a massive collection of these images and other raw materials, AI programs can then draft a file or photo or video based on these instructions.

As AI processes all of these things constantly, it improves its output and approaches images as close as it can to reality. This can explain why images and videos created by AI are not so easily identifiable if they appear on someone’s social media reels so randomly. 

One of AI’s applications is making images come to life. An example of this is a mobile app or filter that allows one to animate photos of two individuals standing side by side so that they move towards each other and hug. 

There are other applications. 

In recent years, it’s become relatively easy to create AI videos. Online videos are increasingly becoming difficult to detect, especially for those who do not know better. 

Guide to detecting AI posts

We are creating this guide to help MindaNews readers detect viral AI posts, with some examples and how to avoid mistakenly sharing AI posts. 

This week alone, AI videos and photos have made rounds online, ranging from jokes of laborers supposedly vlogging from the construction site of the ancient pyramids to photos of popular individuals like Manny Pacquiao and to more serious claims like man on the street interviews.

One example is a photo of actor Bembol Roco designed to supposedly be wearing a Bumblebee costume (Bumblebee is a character in the Transformers franchise). It’s a joke posted by an individual user, created from an AI tool. As of 7 p.m., June 15, the post has received 337 shares, 68 comments, and 1.6K reactions. 

In one of the comments, Mega Star Sharon Cuneta is portrayed as “Mega-Tron,” another robot from the Transformers. 

There’s another example of Philippine national heroes such as Jose Rizal, Apolinario Mabini, Tandang Sora, Lapu Lapu, among others, endorsing brands. 

These are obviously AI-generated, but do not get the same attention as hyperrealism of some AI posts. 

Sharks, whales and barnacles

To those familiar with the technology, it is easy to identify the content as generated by AI. 

This week, MindaNews spotted a few posts of marine animals supposedly being rescued from barnacles and other pests. These were AI videos. Thousands shared the said posts and more have commented, with some not questioning its veracity.

Some netizens have repurposed the content to ride on the virality without explaining whether the posts are true or not. 

In one example, undersea divers “brush” and hose down an immobile shark supposedly covered in moss underwater.  In another example, it’s a video of a group of individuals supposedly rescuing what appears to be a whale from being fully covered in barnacles. 

The AY GRABE post

On Saturday afternoon, June 14, Davao City’s incumbent Mayor Sebastian “Baste” Duterte shared a video of a supposed interview of three young individuals originally posted by creator AY GRABE. The topic was whether they agree with the impeachment of Baste’s sister, Vice President Sara Duterte. 

We transcribed the video as follows: 

Media: Excuse me, pwede ba kitang matanong?
Interview 1?: Yes, po.
M: Pabor ka ba sa impeachment ni VP Sara–?
I1: No.
M: Bakit?
I1: Because it is obviously politically motivated.

Interview 2?: I mean, maraming may confidential funds sa gobyerno, iba nga billions pa. But why single out the VP? They want justice, but their justice is selective.

Interview3?: If these politicians are really serious in cleaning up the government, then investigate all, at isama na nila mga sarili nila.

The original post was posted last Saturday. As of Sunday, 9:15 p.m., the original post has received at least 62K shares, 17.2K comments, and at least 111K reactions. 

Mayor Baste shared the post and said: “And the liberals say they have Gen Z support. Come on.” 

It is not clear whether Baste is trolling his audience; his posts at that Facebook Page commonly features him making fun of news personalities and political enemies even as he is serving as Davao City’s mayor. 

His profile picture is not a photograph of himself but of Presidential son Sandro Marcos.

There were even comments in his post warning the incumbent Mayor that the post is AI. 

Baste’s post has received at least 1K shares, 91 comments, and 9.7K reactions as of 11 p.m., June 16.

On the other hand, his family’s ally, Senator Ronald dela Rosa, appeared to be misinformed about the same post. 

Sharing the same content the next morning at 7:49 a.m., dela Rosa said “Mabuti pa ang mga bata nakakaintindi sa mga pangyayari. Makinig kayo mga yellow at mga komunista!” 

Dela Rosa’s post has since received 3.2k shares, 7K comments, and at least 39K reactions. The post is still up, as of 11 p.m., June 16. 

Dela Rosa has since posted an unapologetic defense: 

How to detect AI posts, according to a tech content page

Amid a swarm of shares, comments, and defenses of these posts, Quezxon City-based Facebook Page EXE Tech Solutions offered advice on how to detect AI in viral content. 

EXE Tech gives the advice in a post titled: “EVEN SENATORS CAN BE FOOLED BY AI ”

EXE Tech correctly points out several indicators of AI content, as well as the indicators that fact-checkers look for in viral disinformative posts. For example, the post was indeed posted by dela Rosa’s official Facebook account (with blue check mark). 

Several elements in the “man on the street video” do indeed show symptoms of sloppy AI work: the logo on the interviewee’s polo shirt is garbled and made up, there is no discernible text anywhere in the video, the vehicles are not consistently rendered. 

EXE Tech encircled these items and noticed the things we did. The EXE Tech post has since received 10K shares, 799 comments, 22k reactions, as of 8:45 p.m., June 15.

MindaNews noticed several things, too. 

In the video, it wasn’t clear whether there were three interviewees (or two). The second and third interviewees sound alike. However, what’s clear is that the three (or two) individuals have different hair styles. There are two discernible voices among the interviewees. At least two of the interviewees looked too much alike and sounded like each other.

As users of AI tools for audio, we also determined that the audio sounds are too garbled (and too AI-processed). 

How to detect AI posts, our advice

1. Count to eight
   • AI videos generated by generators such as Veo 3 (a Google product) typically show eight seconds of clips at a time. The time limit per clip (not the entire video) is a credible indicator that the content is AI. The eight-second limit is what the Google Product is capable of doing so far. 
2. The laws of physics
   • In videos like the one posted by AY GRABE, it was easy to determine that it was an AI post because some areas of the post seemed too two-dimensional despite being a 3D rendering. We also noticed some details such as different hair styles, skin tones, nose shapes, and voices. 
3. Look closer
   • At the risk of increasing the post’s views, examine as much detail as you can see in any of these suspected AI posts. You’ll see that some textures randomly change for no reason, especially background textures. A study from Hong Kong University and the Chinese University in Hong Kong developed tools that could identify, through mathematical modeling, how textures were inconsistent across different frames (see puppy video screenshot below). In that video, the eyes moved illogically around the puppy’s head. The researchers’ tools also detected inconsistencies in how the grass looked. 

4. Too smooth, too shiny
   • AI photos or videos typically look hyperrealistic: That is, to say, the skin is too smooth, the colors are too saturated, faces look more inanimate than real, and there always seems to be something off (but you can’t usually explain it). If an image looks too much like a 3D animated movie like, say, Toy Story, it’s probably AI. AI photos and videos typically cannot correctly render blemishes on faces and cannot typically render realistic images. 

5. Look again
   • At the lower right part of the post that was shared by Duterte and dela Rosa was a watermark. Veo. This watermark is a strong indicator that the post was AI from the beginning. While not all of the AI renderers have this feature, this one could have stopped others from being misinformed about this so-called interview. 

A Veo watermark is on the lower right of the post.

According to AY GRABE’s Transparency Tab, the Facebook page has been around since 2022. The page typically posts pro-Duterte content. Several of its other content were also generated using AI. The YouTube page has likewise been around since 2022. 

The science so far

According to the Hong Kong study, “Recent advancements in diffusion-based video generation have showcased remarkable results, yet the gap between synthetic and real-world videos remains under-explored.” The study said that their research used tools to analyze appearance, motion, and geometry. 

A Cornell University study, meanwhile, says that there could come a time when it would no longer be enough for human instinct to detect AI. Tools must be developed to assist in detecting these renderings that misinform others. 

Here’s an online video on how to identify AI videos rendered by VEO3. 

As with all our other reports, MindaNews welcomes fact check leads or suggestions from the public. 

MindaNews is a verified signatory to the Code of Principles of the International Fact-Checking Network. (Yas D. Ocampo / MindaNews)