More than 58,000 shares, more than 1,100 comments… The article entitled “A robot wrote this article in full. Are you afraid, human? ” published last Tuesday by the Guardian stood out. And for good reason: it was written by an artificial intelligence system.
The British daily has opened its columns to a new type of editor, GPT-3. It is a system developed by the Californian company OpenAI, founded in particular by Elon Musk. What does the robot write? Sentences intended to reassure us. “To begin with, I have no intention of eradicating humans. In fact, I have no interest in harming you in any way. Annihilating humanity seems to me to be a rather pointless effort, ”says GPT-3. The machine writes a little further that it would sacrifice “with pleasure my existence for the good of humanity.” […] Some might say that I wish to become all-powerful. […] Why would I want it? To be all-powerful is not an interesting goal. I don’t care if I am or not, I don’t see it as a goal. In addition, it is quite tiring. Believe me, being almighty gets me nowhere. ”
The whole is quite well written and argued. But the machine by far did not write this text on its own. The Guardian asked GPT-3 to write a short text, 500 words, on the theme “Why humans have nothing to fear from machines”. Most importantly, the published text is not exactly what the virtual editor created: it is a compilation of eight different texts written by GPT-3. They are indeed “human” journalists from the Guardian who selected paragraphs from these texts to compose the final editorial.
The machine thus needed a significant helping hand from humans. This is therefore a relatively modest light on the capacity of artificial intelligence systems to compose texts. We already read them for sports reports or in connection with financial results. And we’re just starting to watch them for news. In terms of pure creation, GPT-3 and others are far not (yet?) At the level of humans.