One of the biggest stories in tech yesterday was about an op-ed written entirely by AI, or artificial intelligence. The Guardian published an opinion piece on Wednesday entitled “A robot wrote this entire article. Are you scared yet, human?”; the so-called article was claimed to have been written from scratch and “by request” by OpenAI’s powerful new language generator GPT-3.
Basically, the respective op-ed was proof that Skynet went online and thinking robots are living among us so to speak. Are you scared yet, dear reader?
A robot wrote this entire article. Does that scare you, human? | GPT-3 https://t.co/SyvCoPUCOp
— The Guardian (@guardian) September 8, 2020
However, as it turned out a little bit later, the whole artificial intelligence thing was faked, and the same goes for the claim about AI’s current capabilities. The article attributed to GPT-3 and described as “a cutting edge model that uses machine learning to produce human-like text” was actually a compilation of several articles, as the “thinking machine” wrote at least 8 different essays after being prompted to “focus on why humans have nothing to fear from AI” .
The final article published in The Guardian was actually written by human editors, as they literally cherry picked the best paragraphs of each of the eight essays. Finally, they put together a so-called AI written op-ed article, or, as the Guardian described it, “the robot wrote eight times this much and we organised it to make it better”.
Editor's note: Actually, we wrote the standfirst and the rather misleading headline. Also, the robot wrote eight times this much and we organised it to make it better…
— Joe Stanley-Smith (@TaxStanley) September 8, 2020
To make it simple, GPT-3 wrote 8 different walls of text, and human journalists at the Guardian, provided that such thing even exists, picked the best parts of each and edited the whole thing into a coherent article. Obviously, this is very different from claiming that AI wrote the final article.
Daniel Leufer from Mozilla called The Guardian’s stunt an absolute joke that still reads badly, and we can confirm that.
“Rephrase: a robot didn’t write this article, but a machine learning system produced 8 substandard, barely-readable texts based on being prompted with the exact structure the Guardian wanted”