Researchers instilled curiosity in robots similar to that of toddlers, which enabled them to acquire language at twice the speed.

      Endow a robot with curiosity, and it begins to acquire language (and misbehave) in a manner akin to a toddler.

      Scientists have long sought to understand how children learn language so quickly, and a recent study from the Okinawa Institute of Science and Technology (OIST) may have uncovered part of the answer: curiosity.

      Researchers created a virtual robot equipped with a brain-inspired neural network and placed it in a simulated 3D environment filled with shapes, colors, and simple commands such as “push left magenta dumbbell.”

      Some robots were rewarded solely for correctly completing tasks, while others received extra rewards for displaying curiosity, effectively giving them a little internal boost whenever they encountered something that challenged their pre-existing understanding of the world.

      How does curiosity facilitate a robot's language learning?

      The robots exhibiting curiosity didn’t just surpass their indifferent peers; they outperformed them significantly. The study, published in Science Advances, revealed that curious robots achieved a true grasp of language in roughly half the time.

      Study author Theodore Tinker likened this experience to trying white chocolate for the first time, despite already enjoying dark chocolate. Taking that risk leads to a broader understanding of chocolate as a whole.

      Halfway through the training, things took an even more intriguing turn. The curious robots began to knock things over and experiment with unrequested actions, essentially playing. This behavior was not programmed; it emerged spontaneously.

      Do robots really make the same mistakes as toddlers?

      The robots also exhibited a familiar pattern seen in how children acquire language. Children often initially use certain verb forms correctly, then apply grammar rules too broadly, leading to mistakes in verbs they had previously used correctly, before eventually sorting out exceptions and self-correcting. The robots demonstrated the same U-shaped dip in performance.

      It’s also a stark contrast to how modern chatbots learn. Large language models like ChatGPT train on vast datasets and predict the statistically probable next word. The robot's brain operates more similarly to ours, prioritizing accuracy while attempting to maintain its beliefs, updating them only when confronted with surprising new information.

      This does not imply that robots comprehend language in the same way humans do. However, it suggests that curiosity combined with diverse experiences may significantly contribute to how toddlers decipher language with minimal information.