When a member of the government says ” the French… “we guess that the following words are “and the French”. Easy prediction. But what if we could predict the rest of speech, one word at a time, from the language patterns found in a huge amount of text? That would be extraordinary and would say a lot about the inventiveness of our policies. The ChatGPT program was trained in generate text from the processing of an ocean of digital texts.
Launched by OpenAI, a private research laboratory in California, ChatGPT arouses fascination and concern. Accessible to the public, millions of us have tried it. The gifted machine respond on any subject. The influence of Marcel Proust on American literature; the comparative advantages of different political regimes; what to give a 10-year-old child; the role of protein in the diet. ChatGPT can also code, translate, write; imitate the style of the Bible, rapper Eminem or Marguerite Duras. His answers are quite accurate, sometimes funny, written in the style of a human conversation.
How will teachers now assess homework? How many of us will lose our jobs? Does the machine broadcast, without discerning them, false, biased, dangerous information? Can it become a propaganda tool? We are warned: the system is not perfect. But what place now for human intelligence in the face of this success of artificial intelligence?
These questions lead to other questions. Since when do we believe that speaking, writing, thinking consist to generate text ? ChatGPT’s answers are a series of words mechanically generated from previous texts whose original authors are, moreover, never cited. An undetectable statistical hack. Is this what we ask of our students? To be hypermnesic parrots capable of an anonymous synthesis?
If so, that’s worrisome. If a machine can produce, like us, school assignments, speeches, tweets, songs, poems or emails, it seems to me that this says more about our own relationship to language, about the poverty of our creations, about our acquired automatisms, than about the evolution of the machine. She could retort to us, like a child, if we reproached her for imitating us: “You started it. »
A certain relationship to knowledge
The human activity of thinking remains hardly comparable to the prediction of a string of words based solely on the ingestion of previous productions. It seems to me, on the contrary, to derive as much from oblivion as from memory, more from emptiness than from fullness. It is not a sum of knowledge that characterizes a embodied intelligence, but a certain relation to knowledge. A way of apprehending what we don’t know, what is buried, what is not yet. Thought works in the hollow between words and what they may mean. It does not generate answers, but other questions.
In this, far from restoring language according to known patterns, which on the contrary it tries to avoid, it offers it a future. embodied, it emanates from a living person, anchored in a real experience, endowed with a singular sensitivity. Thinking is not aimed so much at answering a question as at answering the meaning we will give to it. to be responsible So. This is why a thought bears the name of the person who expressed it.
The broad language models, like ChatGPT, constitute a considerable technological advance which will free us from mechanical tasks provided that we manage to use it without depending on it. These machines owe their existence to centuries of patience, thought and creativity. All that remains is for us to rejoice that we no longer have to speak, write and think like machines.