This is how the UJA system based on ChatGPT works to combat hate messages on the networks

Until now, hate speech on social networks has been fought with censorship. The platforms have developed systems for the recognition of this type of messages, to proceed with their elimination. However, it has been shown that this strategy is not the most appropriate, it is even counterproductive, because the censorship that haters receive generates a kind of solidarity with them. The University of Jaén (UJA) has developed a different and more effective artificial intelligence system based on ChatGPT.

María Teresa Martín and Luis Alfonso Ureña, principal investigators of MODERATES.

The tool to combat hate speech on social networks that the UJA has devised uses the same technique as social organizations. NGOs and associations have been using an alternative to censorship for some time that is more effective for them, and which has now also assumed the research group SINAI (Intelligent Information Access Systems) UJA.

What is the ChatGPT-based AI technique to combat hate messages used by this UJA group?

The SINAI group counters hate messages with a different technique: instead of removing the entries in question, offensive messages are countered with arguments that dismantle the attack to the group to which they are directed. This intelligent system promotes constructive communication through positive messages, but with the substantial difference that the counter narrative to combat hate speech is generated by a robot equipped with ChatGPT technology.

This tool, which is still in the experimental phase and is not installed on any social network, is part of the MODERATES project: Content Moderation on Social Networks, which has financing of almost 300,000 euros corresponding to the call for Ecological and Digital Transition of the Ministry of Science and Innovation. AND relies on natural language processing techniquesone of the lines of artificial intelligence that is advancing the most in recent years.

What more step is taken with this AI system to combat hate messages

“In this project we go a step beyond the detection of hate speech, for which we already have quite effective systems. In this project we not only detect hate speech, but also mitigate it with counter-narratives in a dynamic way»explains the researcher from the SINAI group, María Teresa Martín, who, together with Luis Alfonso Ureña, leads the MODERATES project.

This tool opens a new path in the moderation of social networks and forums, since it “until now there was nothing like it, because it was very difficult to generate language as if it were a human”. AND its application is multiplebecause it could be installed both in a social network and in a forum, or even in a digital communication medium, to automatically moderate the comments that users generate about the news.

“Generative AI models give us very good results and they work very well. In the counter-narrative, arguments are given in which it is shown that the hate speech is false, and all this is generated automatically», adds María Teresa Martín.

How this system has been arrived at to combat hate speech

To arrive at this new tool, the SINAI group of the University of Jaén has had to carry out a preliminary work of training of robots, so that they are able to recognize expressions, linguistic expressions, words… that denote hatred towards a group. A job, for which they have used textual corpora, lexicon resources, dictionaries, web pages which are processed to obtain hate speech.

In addition, the insults and hate speech collected are not only those that appear in the Dictionary of the Royal Academy of the Spanish Language, but also those used in everyday language, on social networks, on the street… Hence, they have used a set of expressions and expressions of language typical of the linguistic reality of the place where the system is going to be implemented, because it is not the same, for example, the Spanish spoken in Spain as the from Peru.

Where did you get the collection of insults used in this and other projects?

In order to have a rich and updated corpus of insults, this research group launched on Telegram the Fierce chatbot, still active. This channel on the social network has an artificial intelligence system that continually challenges the user to insult him, to raise the tone of the conversation and use his most uncouth resources to denigrate him. The objective of this chatbot was none other than to create a huge database with real insultsand not so much those that are collected in more formal dictionaries.

With this tool and the others used for the collection of material for the training of artificial intelligence systems for language recognition, it has been possible to a very high level of efficiency, close to 90%, according to the results obtained in a language recognition contest with artificial intelligence, in which this research group participated. “It is true that in these contests it is a very controlled environment, but it can be said that the systems have a very high capacity to detect hate speech.”

In what other areas has natural language processing technology been used?

The SINAI group has transferred its natural language processing technology to other scenarios such as the detection of suicide risk and gambling addictionbased on messages expressed on social networks.

These are projects in which artificial intelligence, linguistics and psychology are combined, in which the artificial intelligence system does its detection work, but the subsequent intervention is left in the hands of a team of expert psychologists.

Artificial intelligence and natural language processing systems have enormous potential in an environment such as social networks, with millions of messages daily, which due to their volume cannot be moderated by people. And works like those of this UJA research group are making this technology the best ally.

Deixe um comentário

O seu endereço de e-mail não será publicado. Campos obrigatórios são marcados com *