The no-thought machine
Remember when I was writing on the “matching machine”, an impressively accurate prediction of what Machine Learning would look like, by André Maurois, dating back from 1950?
Well, in an ideal world, matching machines would have become thought machines, even creative ones, for what is creative thought, if not the ability to articulate together concepts and ideas?
Turned out that Machine Learning preferred exploring the art of paraphrasing, which is also a matching one.
I did the experiment with Chugzi.com, on the English translation of Maurois’ text.
Here is the result:
Original text (André Maurois, translated by me):
— This is our time! he said. How can you expect this world to work properly when a cultured Frenchman ignores the name and the work of one of the greatest intellectual organisers of this time? My dear Aurelle, Dr. Vannevar Bush is the man who made possible, by his method, the use of atomic energy. Now he has written a revolutionary article on the superman’s working processes.
Chugzi’s AI-powered paraphrasing:
We live in the age of great ideas, indeed. We live in the age of great ideas. He said: It is our time! How can you expect the world to work properly when a brilliant Frenchman ignores the great work of one of the most brilliant minds of the 20th century? I will not mention anyone else, but Dr. Vannevar Bush is the man who made possible, by his method, the use of atomic energy. And now he has written a revolutionary article on the superman’s working processes.
I am not surprised. It appears that this AI merely looked for synonyms, and has a poor stylistic sense. It remains very interesting in two ways, though.
First, as a tool for inspiration. The repetition of “We live in the age of great ideas”, for instance, is not genius in itself. But having it scattered around a longer text gives a voice to a character. Someone who insists heavily on whatever he utters becomes recognizable, it suggests a nevrosis or insecurities that may prompt the reader’s empathy.
Second, the confrontation of human writing with artificial rephrasing reveals the process, both of human creativity and of artificial shortcomings. The milestones of the discourse (“the world to work properly”, “atomic energy”…) did not change, highlighting what is essential to a text’ understanding, and what can be re-arranged, to make thought clearer, more fluent, and more elegant.
Machine learning does what our brains know how to do, but cannot explain why: recognizing faces, differenciating muffins from puppies, findind similarities. For long, human brains were the most powerful tools on earth for critical mind, they excelled at differenciating what mattered in a speech, and what was accessory to it, to persuade better. Rhetorics and Sophistics were born from this ability. Now, it seems that AIs are becoming increasingly good at mimicking it.
However, let’s remind students that with every technological progress comes its counter-weapon. Paraphrasing is made easier by AI, but anti-fraud systems are also improving. Most universities have been using anti-plagiarism technologies for years now, that work the same way, with contrasting intentions, than the paraphrasing AIs.
