“Welcome technological advances, but understand the limits of this type of system”

Lhe general public has discovered and been fascinated by the power of ChatGPT, this conversational agent developed by OpenAI, which has recently been made available free of charge on the Internet.

ChatGPT is an artificial intelligence (AI) tool for testing the performance of the newest language generation models, called Transformers. ChatGPT can be used for many tasks: summarizing a text, answering questions, creating or correcting a computer program, doing an assignment, or even simulating an interaction with a deceased person.

While everyone plays with and proudly recounts “mind blowing” experiences, very few are concerned with the ethical aspects of these black boxes. Of course, we must salute the technological advance and the remarkable performance, but we must also understand the capabilities and limits of this type of system.

Read also: Article reserved for our subscribers ChatGPT, the software capable of writing confusing little texts

What is ChatGPT up to? Texts, which he digested using algorithms… GPT-3 (Generative Pre-trained Transformer 3) is the language model of ChatGPT, a 175 billion parameter neural network architecture that uses pre-trained linguistic representations from huge databases (Wikipedia, web pages) to produce texts that imitate human language in several languages. By construction, the model only recorded information on the words and parts of words and their context of appearance.

Flooding the world with “fake news”

ChatGPT is therefore trained to give, from a textual input, the following sequence of the most probable words – which is not really an answer to a question! The model is obviously flexible enough not to always produce the same text. ChatGPT also uses reinforcement learning through human trainers who have ranked several alternative answers. The model is thus optimized from human interventions.

ChatGPT is a socio-technical system that weighs on a large number of ethical values. He can imagine stories that are harmless or corrosive: we have seen stories of rabbits… or the design of weapons.

The most direct danger of these systems is the possibility of flooding the world with billions of “fake news”, since ChatGPT is capable of text inventions of all kinds if played with. It could then be fed with this new artificial data collected from the Internet.

Read also: Article reserved for our subscribers Beyond artificial intelligence, the ChatGPT chatbot owes its speaking skills to humans

We would arrive at an incredible digestion of opinions, knowledge, bibliographies perhaps false by machines, amplifying disinformation on the Internet. It is still easy to find ChatGPT at fault on very specific questions, on which it is unable to reason and to disentangle the true from the false. But it is necessary to learn to question the responses of the system. Banning ChatGPT at school might not be the best solution; learning to use it is surely a key to accelerating our understanding of these systems and keeping our free will.

You have 50.33% of this article left to read. The following is for subscribers only.

Leave a Comment

Your email address will not be published. Required fields are marked *