Brain Inspired AI

Brain Alignment

The discussion on brain-based alignment in AI explores how neural networks can mimic human learning processes to ensure proper behavior. Lin Zhao highlights that current AI models, like OpenAI's, use reinforcement learning from human feedback to align their behavior, similar to how humans teach children 1. Despite these advancements, Lin notes that neural networks still differ significantly from the human brain in complexity and data requirements. He explains, "For the neural network we need to use a lot of data to train this neural network... but for the human beings it's different" 2.

   

AI Customization

Customizing AI behavior presents exciting possibilities for future AGI systems. Lin Zhao suggests that as AI evolves, we might not need traditional alignment methods, as we could directly customize neural network behavior 1. Lu Zhang emphasizes the importance of multimodality, integrating text, voice, and video to enhance AI's understanding of the world 3. Lin envisions a future where AI systems seamlessly convert sensory inputs into semantic understanding, stating, "We just need a convention from the image of the voice to the semantics" 3.