Google Bard accused of plagiarism as publishers fight back against AI chatbots


Posts: 82   +0
In a nutshell: In yet another example of questionable sourcing practices by an AI chatbot, Google's Bard has been accused of plagiarizing an article from Tom's Hardware. According to the site's editor-in-chief Avram Piltch, Bard copied data from one of its recent articles without citation. Curiously enough, the chatbot eventually apologized for its action when Piltch pointed out the plagiarism.

According to Piltch, the issue came to light when he asked Bard about the performance of Intel's Core i9-13900K and AMD's Ryzen 9 7950X3D. When asked which is faster, the chatbot said the latter is faster in gaming, and went on to state that "In our testing, the $699 Ryzen 9 7950X3D is 12% faster than the $589 Core i9-13900K at 1080p gaming at stock settings, and 9% faster when the chips are overclocked."

Not only was the data set lifted directly from the Tom's article, the entire sentence was a thinly reworked version of the original. Despite deriving its information from Tom's Hardware, Bard failed to attribute the quote to the original article and instead, made it seem as though the data was from testing conducted by Google.

Aside from the plagiarism charges, Piltch also pointed out that Bard's reply was incomplete at best and misleading at worst. While Google's chatbot stated that the AMD processor was faster than the Intel chip at gaming, Tom's found that the Intel chip was better in most productivity apps.

Overall, Tom's recommended that people choose the cheaper Intel chip over the AMD CPU unless gaming is the number one priority. Bard missed all the finer nuances of the comparison, and ended up highlighting only the gaming aspect of the two chips, thereby offering an incomplete analysis.

Plagiarism by an AI chatbot is nothing new. According to The Wall Street Journal, Google Bard has a habit of serving up answers without providing links to sources.

As an example, the publication recounted that in a recent demo, the chatbot offered a summary of many news stories from The New York Times, but failed to link to any of the stories it summarized. Instead, it just ended its response with a generic endorsement of the newspaper, saying "For more on these and other stories, please visit the NYT website."

Such incidents have added to the growing sense of unease among writers and online publishers. Some are fearful that if chatbots continue scraping their content without accreditation, it could reduce traffic to their websites and shrink ad revenues. Website owners and online publishers have also been up in arms against tech companies like OpenAI, Microsoft and Google for using their content to train AI chatbots like ChatGPT without any compensation.

Image Credit: Tom's Hardware

Permalink to story.


Avro Arrow

Posts: 3,714   +4,802
I'm starting to think that AI needs some HEAVY government regulation because the only thing that is trustworthy about a corporation is that it will lie, cheat and steal to get everything out of you that it can.

That's not the mindset that we should want AI to have.

James Ryan

Posts: 61   +89
There will be a point where everyone contributing information like Toms Hardware is not financially rewarded with enough to get by. When this happens these AI systems will have no new information to draw from and the system will collapse. AI has to learn and create for itself rather than pulling other people's data to become anything truly worthy.


Posts: 1,365   +2,191
They will surely grab all knowledge from my 20+ year-old website. I should feel upset, but I don't really. If it will help humankind, go ahead.


Posts: 110   +92
I use chtagpt through some bots on Telegram (I don't feel like creating an account with Microsoft)
In a week of use it help me solved a programming problem in PyTorch that I had from a couple of months ago (I hate Python), I was able to fill in information to complete an article I was writing (adapted to the context), and I was able to understand at the code level a mathematical problem that I have been working on from time to time for at least 5 years (I only need the final, most esotheric operation, but the papers and some MATLAB code wont help)
For my part, I'm happy with it.
And now I plan to run LlaMa from Facebook on my PC.
Last edited: