AI Isn’t Smarter Than a Baby—Yet

AI isn't smarter than a baby yet.

Okay, so babies might not be able to write code

or solve advanced math problems,

but they learn to make sense of their world

with amazing efficiency,

identifying new objects

after seeing them just a handful of times

through observation as well as physical interaction.

But why are we making this comparison?

Well, it's because, when it comes to advances in AI,

the architecture of babies' brains might hold

crucial insights.

In fact, a more baby-like version of AI could make

frontier models cheaper and less energy-intensive.

It could also help AI-powered robots learn

to make sense of their environments in a more natural way.

Researchers at Meta and elsewhere developed a new test

that highlights the learning skills of babies

and pushes AI researchers

to design algorithms to match them.

The EgoBabyVLM challenge judges

how well vision language models, or VLMs,

which learn from both text and imagery,

make sense of the world as a baby sees it.

This requires a model to describe the world

after ingesting about 1,000 hours of video

collected from cameras strapped to the heads

of infants and toddlers.

Yes, really.

It turns out the cutting-edge models fail miserably

when fed this footage,

which suggests there may be something different

about the design of a baby's brain

that enables it to learn so rapidly

and from so little information.

Instead of curated data sets,

babies learn from a kaleidoscopic view of things,

parents talking about objects that are no longer visible,

or who are indicating things using a gaze or gesture,

or discussing events from the past or in the future

rather than whatever's happening right then.

A cognitive scientist at Stanford

who is involved in developing the test told me that,

when it comes to AI,

it's clear there's more than just language that's needed.

For more like this, subscribe to my newsletter,

AI Lab at WIRED.com/newsletters.