Mark Zuckerberg promises a revolutionary technology that will be “close to human intelligence”

It seems that the train of artificial intelligence is moving faster than we can imagine. After the “Chat GPT” program became one of the revolutions of this technology, the “Meta” company talked about a new model that “approaches the intelligence of the human mind.”

Meta said on Tuesday that it would provide researchers with access to a new “human-like” artificial intelligence model that it said could analyze and complete images with a higher degree of accuracy than existing models.

“The current artificial intelligence and machine learning are useless,” said Yan Lucon, scientific director of artificial intelligence at Meta, the company that owns Facebook, in a statement to reporters, adding that “humans have logic while machines lack it.”

Lecon was speaking on the sidelines of an event in Paris as Meta showcased a new AI model called Image Joint Embedding Predictive Architecture / I-JEPA.

What is this new technology?

Lukon aims to develop machines that can mimic human learning by acquiring internal models of the world, enabling faster learning, better problem-solving, and adaptability in new situations.

Artificial intelligence researchers have tried to devise learning algorithms that capture basic common sense information about the world and then encode it into a digital representation that the algorithm can access later. For the system to be effective, it must learn in a self-supervised way — that is, directly from unlabeled data such as images or sounds, rather than manually labeled data sets.

Meta is one of the most popular publishers of open-source AI research through its own internal research lab, according to Reuters, and its new model could be a game-changer in an industry that continues to advance.

The American company stated that this computer vision technology “learns (…) by creating an internal model of the external world by comparing abstract displays of images (instead of comparing the elements of the images themselves). The technology focuses on facilitating learning that reflects human observation capabilities.

Based on Lucon’s vision, the Image Joint Impeding Predictive Architecture allows a machine to think abstractly like a human being. It will be available to any researcher who wishes to test it.

I explained to the “Reddit” website how the new technology functions by using the example of a picture of a dog where some parts are obscured.

According to the American site, completing the hidden parts of the image of the dog will depend on guessing that there is a leg or a head, for example; Because that makes sense given the rest of the same image can be seen.

The i-JEPA technology is designed to learn how to predict the contents of the hidden parts of an image in a way that closely resembles how humans would expect it.

Meta is looking forward to developing the technology in the future to enable the user to predict the places and times of long-term future events in a video clip from a short context, and to adapt these predictions at the level of voices or written texts, according to the company itself.

The company stated that this is a crucial measure to apply and expand self-supervision techniques for developing a comprehensive understanding of the world through learning.

What are the differences from previous technologies?

Google has created AI programs like “ChatGBT” and “Bard” that use large databases to predict the next words in a sentence and generate various types of texts (such as theses or poems).

Programs including “D-A” and “Medjourney” work on the same principle to produce images.

Lucon said that a revolution has already been achieved. He mentioned that a model trained on billions of symbols may seem to understand, but in reality, it can still make silly or logical mistakes.

Meta stated that generative methods tend to prioritize detailed aspects instead of encompassing predictable concepts and, as a result, encounter difficulties in generating precise human imagery.

Mark Zuckerberg, the CEO of Meta, announced on his Facebook page that this technology aims to develop an AI system that “mirrors human understanding of the world” more accurately.

Lucon commented that generative models are outdated and will be replaced by predictive techniques, similar to the model introduced by the American group.