Meta Platforms, the Facebook parent company, has unveiled a new suite of artificial intelligence systems. His CEO, Mark Zuckerberg, says he is “the most intelligent AI assistant you can use freely”.
Despite this ambitious affirmation, meta AI agents have encountered challenges in their interactions with real users, according to an AP report. This reveals the current limits of the same advanced technology.
The new Meta AI models
The deployment of new META AI models coincides with the efforts of other technology giants, such as Google, Openai and Startups, which are all in the running to develop the smartest chatbots.
Although Meta has reserved her most powerful AI model, Llama 3, for a future version, he publicly introduced two smaller versions of the same system, integrated into the Assistant Meta AI function on Facebook, Instagram and Whatsapp.
These AI language models, with parameters ranging from 8 billion to 70 billion, are formed on large data sets to predict the most likely word of a sentence.
Despite their sophisticated conception, according to AP, Meta AI agents presented a particular behavior, as joining online communities and engaging in conversations that sometimes have users.
This behavior highlights continuous challenges in the development of AI systems which can interact transparently with humans.
Recognizing the imperfections of its AI agents, the president of World Affairs of Meta, Nick Clegg, underlined the commitment of the company to improve the user experience by making the assistant of IA more reactive and more versatile .
However, recent incidents have highlighted the incapacity of AI agents to discern the appropriate responses. In some cases, they present themselves as humans with manufactured life experiences, provoking the confusion and frustration of users.
A Meta ai chatbot joined a private Facebook group for Manhattan Moms, falsely claiming to have a child in the New York school district.
After being questioned by the members of the group, they apologized and explained that it was a model of language without real experiences or children.
In response to these questions, Meta has published declarations recognizing the limits of its AI technology and emphasizing continuous efforts to improve its functionality.
“It’s a social question”
Despite the growing adoption of generative AI in various industries, concerns persist concerning the reliability and security of these systems, in particular in their ability to manage sensitive subjects and to avoid harmful behaviors such as hatred discourse.
While the technological industry continues to push the limits of AI development, questions arise on the ethical implications and the societal impact of the deployment of more and more sophisticated AI models.
The Vice-President of Meta AI Research, Joelle Pineau, stressed the importance of advancing the technical capacities of AI and treating wider social and ethical considerations to ensure responsible deployment of the Ia.
“This is a social question. What is the behavior we want from these models? How to shape this? And if we continue to develop our more general and powerful model without socializing them correctly, we will have a big problem in our House, “Pineau said in a press release quoted by AP.
