Abstract:Social learning plays an important role in the development of human intelligence. As children, we imitate our parents' speech patterns until we are able to produce sounds; we learn from them praising us and scolding us; and as adults, we learn by working with others. In this work, we survey the degree to which this paradigm -- social learning -- has been mirrored in machine learning. In particular, since learning socially requires interacting with others, we are interested in how embodied agents can and have utilised these techniques. This is especially in light of the degree to which recent advances in natural language processing (NLP) enable us to perform new forms of social learning. We look at how behavioural cloning and next-token prediction mirror human imitation, how learning from human feedback mirrors human education, and how we can go further to enable fully communicative agents that learn from each other. We find that while individual social learning techniques have been used successfully, there has been little unifying work showing how to bring them together into socially embodied agents.

What problem does this paper attempt to address?

The paper aims to explore the application of social learning in machine learning, especially how to better achieve this goal after advancements in Natural Language Processing (NLP). Specifically, the paper focuses on the following points: 1. **The Importance of Social Learning**: Humans learn by observing and imitating others, and this social learning is crucial for intellectual development. For example, children learn to speak by mimicking their parents' language patterns. 2. **Social Learning Techniques in Machine Learning**: The paper investigates which current machine learning techniques can simulate the human social learning process. For instance, Behavioral Cloning and Next-Token Prediction are similar to human imitation learning; learning from human feedback is akin to the human educational process; and studying how to construct fully communicative agents that can learn from each other. 3. **Classification of Social Learning**: According to Tomasello's theory, social learning can be divided into Imitative Learning, Instructive Learning, and Collaborative Learning. The paper is organized according to this classification. 4. **Development of Large Language Models (LLMs)**: The development of LLMs provides new opportunities for achieving social learning. Early experiments show that agents built entirely on LLMs can exhibit significant zero-shot performance on physical tasks, and these agents have a natural ability to utilize human demonstration data and cultural knowledge. 5. **Application of Social Learning in AI**: The paper discusses in detail the application of Imitative Learning, Instructive Learning, and Collaborative Learning in AI. Particularly, imitation techniques are more mature in training generative models, so the paper focuses more on instructive and collaborative forms of social learning, which are more directly driven by the emergence of LLMs. 6. **Future Research Directions**: By carefully analyzing the gap between human social learning paradigms and the current techniques used to train agents, the paper hopes to inspire more research to fully leverage the advantages of social and cultural learning. In summary, this paper aims to provide guidance for the future development of advanced agents that can learn through interaction with other agents by systematically reviewing social learning-related technologies and their applications in AI.

Social Learning through Interactions with Other Agents: A Survey

Learning through Dialogue Interactions by Asking Questions

Dialogue Learning with Human-in-the-Loop.

Teaching Machines to Converse

Uman-in-thel oop

Socialized Learning: Making Each Other Better Through Multi-Agent Collaboration

Why Robots Should Be Social: Enhancing Machine Learning through Social Human-Robot Interaction

SocialAI: Benchmarking Socio-Cognitive Abilities in Deep Reinforcement Learning Agents

Towards Social AI: A Survey on Understanding Social Interactions

Exploring Collaboration Mechanisms for LLM Agents: A Social Psychology View

Listen, Interact and Talk: Learning to Speak via Interaction

The SocialAI school: a framework leveraging developmental psychology toward artificial socio-cultural agents

Intelligent Social Learning

SocialAI 0.1: Towards a Benchmark to Stimulate Research on Socio-Cognitive Abilities in Deep Reinforcement Learning Agents

Conversational AI for multi-agent communication in Natural Language

SOTOPIA-$π$: Interactive Learning of Socially Intelligent Language Agents

Learning from humans to build social cognition among robots

The SocialAI School: Insights from Developmental Psychology Towards Artificial Socio-Cultural Agents

A framework for the emergence and analysis of language in social learning agents

Exploring Social Learning in Collaborative Augmented Reality With Pedagogical Agents as Learning Companions

Learn2smile: Learning non-verbal interaction through observation