AI Isn’t Smarter Than a Child—Yet

If you think and artificial intelligence model running on thousands of cutting-edge computers chips he is smart, let me introduce you to the 1 year old.

Well, so babies can’t write computer programs, solve advanced math problems, or argue philosophical ideas. But unlike today’s AI models, which use the knowledge of ocean education as well a lot of energy for a small countryBabies learn to perceive the world with extraordinary skill. They recognize new things after seeing them once or twice, and they learn through visual observation and physical interaction.

When it comes to controlling AI, babies – and their brain structure – can hold important information. Developing a baby-like AI model would make edge models cheaper and less energy-intensive, and would also be important if AI-powered robots could learn their environment naturally.

To explore this new frontier, researchers from Meta, the University of Stanford, the University of Tokyo, and France’s École Normale Supérieure. developed a new test that assesses the learning abilities of infants is pushing AI researchers to develop algorithms that match them.

The EgoBabyVLM Challenge they judge well how visual language models, or VLMs, who study both texts and images, can understand the world as a baby sees it. It takes an example to describe the world after ingesting about a thousand hours of video taken from cameras attached to the heads of infants and young children. (Yes, really.)

It is clear that advanced models fail miserably when presented with a vivid and distorted picture, suggesting that there may be something unique to the structure of a child’s brain that enables it to learn quickly from limited information.

Instead of having stored data, babies learn from the old perspective: parents talk about things that are no longer visible, show things with their eyes or hands, or talk about past or future events rather than what is happening at the time. Children learn not only from language but from many modern and modern things, says Michael Frank, an experienced scientist at Stanford University who works in language learning and was involved in the development of EgoBabyVLM.

Tests show that when it comes to AI, “it’s clear that there’s more (than language) that’s needed,” says Frank.

Learning a Language

EgoBabyVLM is the latest example of how scientists are using AI to explore human intelligence. A Challenge called BabyLMintroduced in 2023, it provided AI models for language learning using roughly the amount of data a 10-year-old takes in—tens of millions of words, compared to trillions for AI models. Interestingly, it seems that transformer-based AI models—which use language to manage the relationships between words in different sentences—can do this better, finding out what’s difficult. The views of Noam Chomsky about how syntax can be encoded in the human brain.

Ryan Cotterell, a linguist at ETH Zurich who pioneered BabyLM, says things are different when it comes to understanding the world. “There’s not going to be a lot of social interaction — there’s no social networking,” he says.

Joshua Tenenbaum, a cognitive scientist at the Massachusetts Institute of Technology, says BabyLM has shown that models do not have “intelligence” about the world, social situations, or emotional intelligence.

“Transformers are very good at finding patterns in data,” says Tenenbaum. “But it seems that effective learning strategies are not able to capture the type of data that an infant or child receives and learns from everything they do.”