Huawei Researchers Propose 'Embodied AI' for Enhanced Real World Interaction and Learning
Summary:
Huawei's Noah’s Ark Lab's researchers have proposed a framework for "embodied artificial intelligence" (E-AI) as a crucial advancement towards artificial general intelligence (AGI). They argue that current large language models like OpenAI’s ChatGPT and Google’s Gemini cannot fully understand the real world due to a lack of direct interaction. They suggest creating AI agents that can perceive, act, learn, and remember within a physical embodiment to gain better comprehension and function like a general intelligence. Though the team outlined a theoretical structure for creating embodied AI, they also highlighted current technological challenges, including the fact that most potent AI models exist on vast cloud networks.
The Noah’s Ark Laboratory of Huawei in Paris recently introduced a groundbreaking framework for "embodied artificial intelligence" (E-AI), which they consider to be a critical development in the quest for artificial general intelligence (AGI). AGI, often referred to as "human-level AI" or "strong AI," is an AI system that has the capacity to execute any task given adequate resources. While there's no universal agreement on what makes an AI system a general intelligence, several companies, like OpenAI, focus exclusively on this technology.
Towards the end of the 2010s, the emerging generative pre-trained transformer (GPT) technology led many AGI specialists to adopt the notion that "scale is all you need," implying that they foresaw that transformers, scaled beyond current capabilities, could eventually pave the way to an AGI model. The Huawei researchers, however, contradict this belief in their paper by arguing that large language models, like OpenAI’s ChatGPT and Google’s Gemini, fail to comprehend the real world due to their lack of engagement with it. As per the publication:
"Many believe that merely expanding these models in terms of data and computational power can lead to AGI. We disagree with this view. We suggest that genuine understanding is only possible through E-AI agents who interact and learn from the world they inhabit."
According to the scholars, for AI agents to truly engage with the real world, they must be housed in some sort of body that enables perception, action, memory, and learning. Perception, in this sense, is the capability of an AI system to acquire raw data from the real world in real-time, process it, and encode it into a latent learning environment. Essentially, for an AI to understand the real world to the extent of a general intelligence, it needs to have the ability to focus on it via its own sensory systems.
In addition to perception, agents should be capable of taking actions and evaluating their results. Current AI models are mostly "pre-trained," akin to a student given a test with the answers beforehand. By enabling AI to act autonomously and perceive the consequences of its actions as new memories, the team proposes that agents could learn about the world through experience and mistakes, similar to living organisms.
The researchers propose a theoretical scheme by which a large language model or fundamental AI model could be embodied to achieve these goals in the future. However, they also underscore the numerous obstacles that lay ahead. One significant challenge is that the most potent current large language models operate on expansive cloud networks, making embodiment a complex task with present technology. Related: A breakthrough in nuclear fusion could transform artificial intelligence.
Published At
2/7/2024 11:23:13 PM
Disclaimer: Algoine does not endorse any content or product on this page. Readers should conduct their own research before taking any actions related to the asset, company, or any information in this article and assume full responsibility for their decisions. This article should not be considered as investment advice. Our news is prepared with AI support.
Do you suspect this content may be misleading, incomplete, or inappropriate in any way, requiring modification or removal?
We appreciate your report.