Overview

Historical Context and Importance of AGI

Building artificial general intelligence (AGI) has been a defining challenge in the realm of AI research. Since the dawn of the computational era, the aspiration to achieve AGI has profoundly shaped the vision and mission of AI researchers. Broadly defined, AGI refers to machines that can perform any intellectual task that a human being can do. Over the decades, we have witnessed substantial progress in narrow AI, where systems excel in specific tasks. Still, the quintessential goal of creating an AGI, a system with broad and flexible cognitive abilities, remains elusive.

Recent Advances in AI: Towards AGI?

The recent advances in large language models (LLMs) like GPT-4 and LLama-2, present an intriguing twist in this narrative. These models have displayed remarkable capabilities, sometimes on par or even surpassing human abilities in specific domains, hinting at a form of the Turing test being passed. Furthermore, the scope of these models is no longer limited to just natural language processing (NLP). We’re seeing promising signs of these models branching out, such as AI agents utilizing tools, retrieval from external database, displaying reasoning abilities, competing complex tasks via writing code, multimodal learning with text and images, etc.

The Road Ahead: Challenges and Limitations

However, as promising as these advances are, the gap between current LLMs and true AGI is still significant. Notable limitations include diminishing returns and potential constraints to the scaling law, lacking robust reasoning capabilities, hallucination and factual inaccuracies, lack of commonsense reasoning capabilities, and many others. Furthermore, as we move closer to AGI, it becomes paramount to address the critical concerns of safety, ethics, and regulatory implications, for example, aligning AGI’s values with humanity’s diverse set of beliefs, navigating the moral dilemmas and ensuring AGI can make ethical decisions, and addressing security risks of generative AI.

Workshop Scope and Topics

This workshop aims to become a melting pot for ideas, discussions, and debates regarding our proximity to AGI. We invite submissions on a range of topics including, but not limited to:

Frontiers of AGI research: examples include AI agents, embodied AI, retrieval-based and tool- augmented LLMs, knowledge-enhanced AI, and multi-agent AI.
Classic AGI Attempts as Inspiration: Delving into historical methods such as expert systems, symbolic AI, Type I and Type II reasoning for insights that can guide LLM research further.
Interdisciplinary Insights for AGI: Drawing parallels from fields like psychology, sociology, and neuroscience to inspire and inform the development of LLMs towards AGI.
Fundamental Limitations of LLMs: Analyzing the intrinsic capabilities or lack thereof in LLMs that might impede their progression to AGI. This includes discussions on reasoning, planning, and more.
Practical Limitations of LLMs and Foundation models: Addressing external challenges like system constraints, computational costs, data acquisition barriers, and privacy concerns.
Safety, Ethics, and Regulation in AGI Development: Exploring the complexity of moral, safety, and regulatory concerns that will shape AGI’s evolution.
AGI’s Economic and Societal Impacts: Probing the potential changes AGI might initiate into our societies, economies, and daily lives.

Should you have any questions, please reach out to us via email:
agiworkshop@googlegroups.com