# When Brain-inspired AI Meets AGI

**Authors:** Lin Zhao, Lu Zhang, Zihao Wu, Yuzhong Chen, Haixing Dai, Xiaowei Yu, Zhengliang Liu, Tuo Zhang, Xintao Hu, Xi Jiang, Xiang Li, Dajiang Zhu, Dinggang Shen, Tianming Liu  
**arXiv:** [2303.15935](https://arxiv.org/abs/2303.15935)  
**Date:** 2023-03-28  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.15935v1/

## Key points

- Artificial General Intelligence (AGI) aims to create machines capable of performing any intellectual task that humans can do
- AGI researchers draw inspiration from the human brain and seek to replicate its principles in intelligent machines
- Brain-inspired artificial intelligence is a field that combines insights from neuroscience, psychology, and computer science to develop more efficient and powerful AI systems
- Incorporating training signals from multiple modalities into Language Models (LLMs) is a promising approach to building multimodal AI systems
- Cross-modal alignment is essential for various tasks such as text-to-image and image-to-text generation, visual question answering, and video-language modeling
- Most existing works in the field of large language models (LLM) reasoning adopt prompt-based methods which provide technical methods for LLMs to better demonstrate their ability during problem-solving
- The evolution of AGI systems involves both algorithmic and infrastructural perspectives
- Challenges include scaling up current hardware infrastructure and developing new algorithms that can handle multimodality, reasoning, and generalization
- AGI has the potential to revolutionize various fields such as healthcare, education, and transportation but ethical implications must be considered.

## Summary

Artificial General Intelligence (AGI) has been a long-standing goal of humanity, with the aim of creating machines capable of performing any intellectual task that humans can do. To achieve this, AGI researchers draw inspiration from the human brain and seek to replicate its principles in intelligent machines. Brain-inspired artificial intelligence is a field that has emerged from this endeavor, combining insights from neuroscience, psychology, and computer science to develop more efficient and powerful AI systems. One promising approach to building multimodal AI systems is incorporating training signals from multiple modalities into Language Models (LLMs). This requires aligning internal representations across different modalities to enable the AI system to integrate knowledge seamlessly. Multimodal AI systems have been experimenting with aligning text/NLP and vision into an embedding space to facilitate multimodal decision-making. Cross-modal alignment is essential for various tasks such as text-to-image and image-to-text generation, visual question answering, and video-language modeling. In recent years, most existing works in the field of large language models (LLM) reasoning adopt prompt-based methods which can be divided into three technical routes: zero-shot Chain of Thought (CoT), Few-Shot CoT, and Least-to-most prompting. These methods provide technical methods for LLMs to better demonstrate their ability during problem-solving. The evolution of AGI systems involves both algorithmic and infrastructural perspectives. In terms of algorithms, there are two main approaches: using code or using prompts to enhance LLM reasoning abilities. The former represents a strategy of directly enhancing LLM reasoning abilities by increasing the diversity of training data while the latter provides a technical method for LLMs to better demonstrate their ability during problem solving. From an infrastructural perspective, there are several challenges that need addressing before achieving AGI. One challenge is scaling up current hardware infrastructure as modern deep learning models require significant computational resources. Another challenge is developing new algorithms that can handle multimodality, reasoning, and generalization. Despite these challenges, AGI has the potential to revolutionize various fields such as healthcare, education, and transportation. However, it is important to consider the ethical implications of creating machines that can perform any intellectual task that humans can do. As such, researchers must prioritize developing AGI systems with safety and ethical considerations in mind.