How can Google identify AI-generated content?
When we first heard that ChatGPT is able to write compelling essays, engaging descriptions, profound poems and correct codes, the general public split into two groups:
- Those who wondered how to take advantage of this potential;
- Those who became terrified for the future of their profession.
We’ll tackle both groups in this article.
Identifying AI-generated text isn’t a new challenge for Google. In fact, it’s been a subject of research for years now.
Not only that, but even the content that deploys anti-detection algorithms can be detected, as well as the content translated from another language.
One of the most common methods for that is by using BERT algorithms to replace words with synonyms. However, the content generated with GPT-3 (and 3.5) is somewhat more sophisticated compared to previous versions and alternatives, so new ways of identifying this content need to be developed.
The ideas are already coming from the OpenAI team in the form of “invisible watermarking”, stating that it should become an industry standard.
Namely, this form of cryptographic watermarking is meant to help in detecting AI-generated content. But for everyone to adopt it, it needs to be effective, cheap, not require heavy computing and not affect the output quality.
At the time being, it’s not there yet. But eventually, it can help in preventing plagiarism and spamming.
Why is this so important for Google?
Because AI-generated content goes against Google’s guidelines and it’s perceived as spam. More precisely, AI-generated content that lacks value is considered “spammy”.
However, there’s one more thing in the grey area and that is – if you use a rewriting tool after you used ChatGPT, the final result cannot be detected. We’ll just leave this here.
Could ChatGPT replace Content Writers?
Now that we’ve answered the most important question regarding the detection of AI-generated content, let’s try to answer why do so many professionals feel threatened by this technology.
For Serbian National Broadcaster, PhD Dragiša Mišković, research associate at the Institute for Artificial Intelligence and leader of the Human-Computer Interface research group, stated the following:
“This technology is the essence of future products to be applied in the interaction between humans and machines. In the same way, we can perceive this as a toy with an aim to garner our attention, to overshadow us and for us to become addictive to it and ready to invest a certain amount of money for products that rely on this technology. Nevertheless, this can also be perceived as a danger for our profession.“
We’ve seen both glorification and skepticism surrounding GPT technology in the previous days. But the feeling of AI being our competitor in the workplace is legitimate, Mišković explains:
“We should reiterate that professionally speaking, this is a tangible competition. We are competing with artificial intelligence. We have to realize that we will not win if we all spent all our time dealing with Al. We must all be professionals in our own field of expertise. Al could never be able to replace a good physician, professor, playwright, or director. However, Al will be able to replace poor quality. And very soon,” he concluded.
Well, that’s good news for anyone who takes his job seriously. For those who are used to slacking, they may want to reconsider their choices.
Does ChatGPT in education offer new plagiarism possibilities?
If you indulged in any online discussion about ChatGPT, you’ve probably come across frustrated teachers who now have yet another problem on their plate. Plagiarism checkers they’ve used so far don’t seem to work with this sophisticated chatbot, and it seems that students are getting away with it.
This problem will be resolved once the invisible watermarking gets widely implemented, but what are they to do until then?
Or more specifically, is GPT-3 a threat to the global education system as well?
PhD Dragiša Mišković believes otherwise.
“If we look at the things that way, the very Internet negatively affected school education. Calculators negatively affected school education. The most important thing is that school education should prepare people for current and future times and to keep pace with everything that is going on across the globe. This issue is a bit more complex and we, as a society, should change our attitude towards education. If our aim is to get a satisfying grade, it does not matter whether we copied, found online or a chatbot generated the content in question. These are only different techniques,” he elaborated.
But rather than feeling desperate, general advice to teachers is to accept that this technology is our future and try to explain to students when it’s okay to use it, but also what are the consequences of relying on it too much.
