Audi and the Challenge of Cross-Media Plagiarism Detection

Over the weekend, the German carmaker Audi became the center of a massive plagiarism controversy as a new video campaign was accused of ripping off content from a prominent Chinese video blogger.

The ad, which was produced by the London-based advertising agency M&C Saatchi, featured prominent Hong Kong actor and musician Andy Lau Tak-wah as he delivered a monologue about Xiaoman, the eighth solar term and second solar term of summer on the traditional Chinese calendar.

However, the monologue was not his, nor was it written by Audi or M&C. Instead, it was published  in 2021 on Douyin, the Chinese equivalent of TikTok, by a prominent video blogger that uses the name Beida Mange. 

Mange, who has nearly 4 million followers, pointed out the copying in a video comparing the two works. This kicked off a firestorm of controversy in China, with Audi, M&C and Lau all apologizing for the copying. Audi has also removed the video.

Audi, for its part, blamed the plagiarism on a “lack of supervision and lax review.” While that may be true, it actually belies a unique challenge to cases such as this. Namely, there is no easy way to detect this kind of plagiarism, no matter how hard one works to spot it.

The Challenge of Cross-Media Plagiarism Detection

There’s not much doubt that plagiarism and copyright infringement detection has improved by leaps and bounds over the past 20 years or so.

Turnitin launched in 2000, Audible Magic began providing a similar service for audio files in 2002 and YouTube’s Content ID System debuted in 2007. All these systems are remarkable feats of technology and have helped shape our understanding of copying. 

However, there’s one key limitation that runs through each of them. They compare like works to other works of that type. Writing is compared against other writing, images against images, audio against other audio and video against video. 

While these systems are perfectly fine for detecting when a student copies an essay off Wikipedia or a user uploads a pirate movie to YouTube, sometimes issues cross media types.

Note: There is an exception here for audio works being used in video works. We see this regularly with copyright blocks for songs on YouTube.

For example, in the Audi case, the videos would not match at all. They are completely different videos. The issue is that the script from the Audi video was plagiarized from a video 

However, even if Audi had performed a plagiarism check on the script, it’s dubious if it would have detected the copying. Any such plagiarism check can only look at what is in its database, and it is unclear if the original video has been converted to text to make a comparison possible.

In short, while Audi and M&C may have been lax, that doesn’t necessarily mean that a more scrutiny could have prevented this issue.

A Repeating Problem

This is not a new problem. In July 2015, we looked at the case of The Dollop, which was accused of plagiarizing from the website Damn Interesting for its podcasts. Here, text articles on one site became a script from which an audio podcast was created. 

Similarly, in October 2021, we discussed the Cinemassacre Monster Madness plagiarism allegations. Here, a 2003 article was copied and pasted from to make a script for a review of the movie 28 Days Later. 

In both of these cases, a check of the script before recording would have likely detected the issue. The reason is that there is only one conversion, in both cases a text work was copied into a script for either audio or video.

With the Audi case, there were essentially two conversions. The original video was turned into a text work (a script), which was then turned into a new video. This makes detection much more difficult and is similar to translated plagiarism in that the technology is going to struggle to spot the similarities.

But while technology may struggle, people don’t. That’s why, in all of these cases, the copying was first detected by the public, who went on to point out the issue and create a controversy both on social media and the broader internet.

However, that’s also a point where it is too late. Any infringement has taken place and any public outcry is already well underway. If these issues were detected before publication, a major headache for everyone could have been prevented.

So, What Can We Do?

Preventing this issue is difficult. So much time has been spent developing tools to detect straightforward copying that simple acts such as converting a work into video can throw off the commonly used tools.

That said, there are a few things that users can and should do. 

1. Focus on Text – Text is the most common denominator. Video can be reduced to a script, a song to lyrics, etc. As technology gets better at transcribing audio and video content, this becomes a promising way to detect crossover plagiarism
2. Search Before Producing or Publishing – Even if the tools aren’t 100% perfect, finding out some issues is better than finding out none. 
3. Increase and Improve Citation – Many of these issues aren’t legal ones, but ethical ones. To that end, simple citation can go a long way to improving things. Even if a work is used inappropriately, showing a good faith attempt to cite can blunt much of the criticism one might expect.

To be clear, this is an area where technology is rapidly advancing. Improving the way we parse multimedia content in particular aids in searchability and accessibility, additional tools for detecting plagiarism is just an added benefit.

Ultimately, the main thing is to be aware of these particular issues and plan accordingly. This includes placing additional checks and having a process for denoting who was working on what and when. 

Tracking the production process closely, citing known sources and checking for potential issues may not be a 100% effective strategy, but it is a huge step forward.

Bottom Line

The various anti-copying tools we have available were designed to solve a very specific problem that existed on the internet over 20 years ago. They have proved very capable of that, but the internet has shifted.

Now, virtually every creator is dabbling in various media. Authors are posting to TikTok and YouTube, musicians are dabbling in visual arts, and everyone is picking up the pen to fill their social media accounts.

As creators do more and more crossover works, the risk of crossover plagiarism jumps significantly. Though the tools are getting better for detecting these issues, it’s important to understand the very real technology challenge here and that the current tools have limitations.

This is cold comfort to Audi and M&C. Whether their oversight was “lax” or not, there would have been very real challenges in detecting this copying because of the nature of it. It may not have been impossible, but it’s unclear how much more effort would have been necessary.

In the end, this will be an ongoing struggle. Creators will continue to try and diversify their works and, at the same time, technology will try to bring it to a base level for the purpose of searchability and accessibility. 

Plagiarism and copy detection is simply caught in the crossfire.