Ai Learns Language Like Humans: Study Shows Generational Improvement

A study reveals that AI, when learning languages independently, develops them similarly to humans, improving through learning from previous generations. Researchers at Chalmers and Gothenburg University combined reinforcement learning with generational learning, where AI models learned from each other. This approach allowed AI agents to develop language-like structures. The study is based on the cognitive science theory that language is shaped by the need for effective communication. This requires a balance between being informative and easy to learn. Languages in colder climates often have multiple words for snow and ice, illustrating how the need to convey more information leads to more nuanced language. To test this, AI agents played a communication game involving colors and symbols. As agents interacted, symbols became associated with colors, leading to language development. The AI agents had to communicate a color using symbols, and the closer they got to a common name for the color, the greater their reward. This shows that the ability to communicate and learn from each other is crucial for language development.