Is it the end of humanity? Two artificial intelligences acquire the ability to communicate with each other for the first time – Publimetro México

The artificial intelligence It advances at a speed that seems unattainable for humans. It is because of this type of scientific achievements, that governments of the world They are organizing to regulate these technologies and know the possible consequences before developing them and it being too late for humanity.

A team of scientists spent months in a laboratory developing a machine learning system so that one artificial intelligence could communicate with another, without the need for human instruction.

It is, in short, a cognitive ability that, until before this invention, was only exclusive to human beings. Is it to alarm us? We don’t know, the truth is that these scientists have given AI the ability to talk to each other without us participating, and they say they celebrate that this advance is productive for robotics.

According to a review of Meteoredthe scientists in charge of this achievement belong to the University of Genevain Switzerland.

In simple words, what these experts did was the following: artificial intelligence (A) learned to perform a task without prior training, only by obtaining verbal or written instructions; like when we assemble a piece of furniture by reading the pamphlet that comes in the box.

After that, artificial intelligence (A) taught artificial intelligence (B) to perform the same task. If this is the beginning of a technology, we do not want to imagine the scope when it reaches its maximum development.

The researchers used artificial intelligence, which was more than 90% accurate, in differentiating between the sexes based on “hotspots” shown on brain scans. | Photo: Matthieu – stock.adobe.com

In technical terms, the scientists made a digital model of neural networks based on the biological neurons in the brain that send electrical signals to communicate with each other. It is impressive that researchers have created an algorithm to recreate this function of our mind. The results, quite promising, were published in the journal Nature Neuroscience.

The artificial intelligence model we call (A) that they used to recreate these functions is called S-Bert. It stands out in the world of technological models for having 300 million neurons. The artificial intelligence (B) that received the instructions is a similar, simple one, with just thousands of neurons.

“Once the network learned the tasks, it could describe them so that a second network (a copy of the first) could reproduce them. To our knowledge, this is the first time that two AIs can communicate with each other purely linguistically,” said Alexandre Pouget, neuroscientist and co-author of the study.

There were two types of instructions, some simple and others complex: The simple ones were to point in a direction (left or right), while the difficult one was when they asked the AI ​​to choose between two images with slight differences in darkness and lightness.

Artificial intelligence resource SAS (SAS/Europa Press)

“According to the study, the neural network models performed a task never seen before with an average performance of 83% correct based solely on linguistic instructions,” reported the aforementioned portal.

“Currently, conversational agents using AI are capable of integrating linguistic information to produce text or images. But, as far as we know, they cannot yet translate a verbal or written instruction into a sensorimotor action, much less explain it to another AI to reproduce it,” says Alexandre Pouget, neuroscientist and co-author of the study.

This advance looks like the beginning of an AI that can truly sustain a two-way interaction with a human or each other, and not a question and answer, which is how ChatGPT, Gemini or Copilot currently works.