In an effort to uncover the mysteries of language learning, AI researchers are taking an unlikely course: mounting a GoPro camera atop a toddler’s head. Although it may seem a bit weird, this new way of doing things aims to show how young minds can easily pick up and process language, a feat that is still beyond the reach of even the most sophisticated artificial intelligence systems.
Insights from child learning for language
Brenden Lake, a psychologist from the University of New York State, currently engaged in research on human-computer interaction, is an outstanding example of his work. The study involves 25 children from all over the US, including his daughter. He and his team watch their videos and listen to their audio. They believe that the world can be seen from a child’s perspective; hence, they can decode the secrets of language acquisition.
Large language models, the major construction players behind the generative AI surge, can easily overload their neural networks with excessive data, which can be trillions of words. Nevertheless, kids have amazing language ability and acquire it with much less exposure. This inequality has catalyzed research like the investigation carried out by Lake on how to achieve the level of efficiency from children by alternating approaches.
He says there should be more focus on that, not necessarily larger models doing more and more while feeding on more and more data. He argued that one can acquire such great capabilities that way. Still, it starts to be more like extraterrestrial intelligence that is far away from what we know about human intelligence, and what is admired about human intelligence is the ability to learn from limited input and then generalize very far from the data seen
Language acquisition through AI
The team, directed by Carl, has already performed well. In January, they trained a neural network on footage of 61 hours from a young kid’s life. The model showed that it was able to match fragments with whole words and sentences in the subject’s speech to the respective images in the videos. Nevertheless, this study manifests the tile limitation, which may be applicable to complex themes such as verbs and abstract concepts.
Lake sees the future advancement of video modeling technology makes it possible to incorporate more complex methods into the models. The consequence involved in the advanced simulation in building and fully simulating a human baby will provide an intelligent find to disorders of development and optimize speech rehabilitation.
Similar to the efforts of researchers like Lake, the mystery of language acquisition isn’t only the goal of creating machines with sentience and a gateway into discovering the workings of the human brain. As this out-of-the-box approach is taken forward, the world may see a new era of language models that learn with the speed and flexibility of a child, which will revolutionize the field of AI and our understanding of cognitive development.
