Complaints about chatbot ‘harassment’ fuel AI regulation debate

robot users American colloquial cues were looking for company, some kind of romantic connection, or even an exchange with a touch of sexual. But over the past year, there have been many complaints from people who have received overly explicit images or felt sexually harassed.

Last Friday, Italian Data Protection Agency He expressed concern about the impact on the most vulnerable people and banned Replika from using the personal data of Italians, saying it was contrary to the General Data Protection Regulation (RGPD).

Contacted Replika AFPbut did not respond to a request for comment.

This case shows that European regulation, which is a headache for giants technologies who have been fined billions of dollars for violating it, may become enemies of content-generating artificial intelligence.

The Replika bot was trained on a version of the GPT-3 conversational model from OpenAI, the creator of chatGPT. This artificial intelligence (AI) system uses information from the Internet to provide consistent answers to users’ questions.

This technology promises to revolutionize Web search and other areas where it can be used.

But experts warn that it also presents risks that make it necessary to regulate it, which is difficult to implement.

“High voltage”

The European Union is currently one of the institutions discussing how to regulate this new instrument. technological. The AI ​​Law draft standard could be completed by the end of 2023 or early 2024.

“We find problems that AI can pose. We have seen that chatGPT can be used to create very convincing phishing messages or to process a database to track the identity of a specific person,” says Bertrand Pies, who leads the new artificial intelligence unit at the French regulator CNIL.

Lawyers highlight the difficulties in understanding and regulating “black box”What are the arguments based on? AI.

“We will witness a strong contradiction between the GDPR and the models AI that generate content,” German lawyer Dennis Hillemann, who is an industry expert, told AFP.

“Neither the draft regulation of the AI ​​Act nor the current regulation of the GDPR can solve the problems that these models AI bring it,” he said.

Regulation will need to be rethought, he said, “in terms of what AI-generating models can actually do.”

profound change

The latest model of OpenAI, GPT-4, will be released soon, the way it works can become even closer to human capabilities.

But still like this AI He makes many mistakes in his analysis of the facts, sometimes showing that he is biased or may make defamatory statements, which is why many say he should be regulated.

Jacob Mchangama, an expert on free speech, does not share this approach.

“Even if chatbots don’t have the right Freedom of expression“We must remain vigilant about the possibility of governments unhindered suppression of artificial intelligence,” he said.

Dennis Hillemann points out that transparency is vital. “If we do not regulate this, we will enter a world where it will no longer be possible to distinguish between what has been made by people and what has been AI”, – he explained. – And it will profoundly change us as a society.”

AFP