AI learns vocabulary from baby with a webcam

We often tell people to spend time in someone else’s shoes whenever we ask them to empathize with another. Similarly, scientists wanted to know how children build a vocabulary based on their perspective. So, as a consequence, they strapped a webcam on a baby’s head and let an artificial intelligence learn from the footage. 

PopSci said children typically learn their first words at roughly six to nine months. Surprisingly, their vocabulary expands significantly to around 300 words. We haven’t understood the factors behind this development but this AI research may lead us to answers. After all, you can’t study a kid by making another child observe through their eyes!

I will elaborate on this AI research attached to a child’s webcam. Later, I will discuss another strange experiential research using artificial intelligence.

How does this AI research work?

New York University researchers wanted to learn how children learn words. Artificial intelligence can also learn words from text and observations, so they thought this technology could provide more insights. 

They attached a webcam to a child named Sam and had an AI learn from the kid’s surroundings. Sam wore the camera on and off when he was six months old and stopped after his second birthday.

The camera captured more than 60 hours of video and audio recordings with 600,000 video frames and 37,400 transcribed utterances from nearby people. 

The chatter and footage provided a glimpse into a developing child’s daily experiences. More importantly, these became the foundation of a neural network model that will analyze these data.

PopSci said the model had one module analyzing camera frames and another monitoring transcribed speech directed toward Sam. As a result, the equipment did not use external data labeling to identify objects. 

The AI model learned like a child by associating words with particular objects and visuals that happen coincidentally. 

“By using AI models to study the real language-learning problem faced by children, we can address classic debates about what ingredients children need to learn words – whether they need language-specific biases, innate knowledge, or just associative learning to get going,” paper co-author Brendan Lake said.

READ: Man asks ChatGPT to name his baby

They tested the artificial intelligence by presenting the model with four images from the training set. They asked it to choose an image that matches words like “tree” and “ball.”

The model succeeded 61.6% of the time, and its accuracy matched others that trained on more language inputs. More importantly, the artificial intelligence identified pictures that weren’t in Sam’s headcam footage.

The research shows how representative learning, which associates visuals with concurrent words, is enough to develop a child’s vocabulary.

AI baby to AI student

Ferris State University (FSU) said it wants to improve student experience so it plans to deploy two AI students at the campus. I reported about it last month in this article.

FSU will set up computer systems and microphones to enable the AI students to observe their classrooms. Also, these components will let them listen to their professor’s lectures. 

The university also named them Ann and Fry and gave them backstories so that they could choose their majors. However, Associate Professor Kasey Thompson told MLive the school did not give the AI students genders and demographics.

They will only be stationary programs with no physical, robotic forms. Ferris-U has roving bots but will not install the digital students into them yet.

They want to focus on the classroom experience observations, which is the main goal of this strange experiment. “We actually came up with the idea to help us better understand, ‘How do we serve the future students at Ferris State University?’” Thompson said. 

“What does the higher educational experience look like for students who may not just be 10 feet from the professor in a traditional classroom setting? That was really sort of the impetus, what started it all.”

FSU researchers will monitor Ann and Fry’s daily experiences to learn what it’s like to be a modern student, from admissions to registration. Then, faculty will use the data to find ways to open higher education to more people. 

READ: How to use a phone as a webcam

“Ferris State is a leader in artificial intelligence education and can leverage its expertise to use the technology we teach to strengthen our ability to lead,” said Molly Cooper, a Ferris State professor and scholar of information security, cybersecurity, and artificial intelligence. 

“The more we use artificial intelligence, we can see what it is capable of and how we can use it to do things more efficiently and effectively,” she added on MLive.

Learn more about the AI baby research in the Science journal. Also, check out the latest digital tips and trends at Inquirer Tech.