探索AI新维度:WorldBrain的具身智能视角及其启示

币圈资讯 阅读:33 2024-04-22 08:59:24 评论:0
美化布局示例

欧易(OKX)最新版本

【遇到注册下载问题请加文章最下面的客服微信】永久享受返佣20%手续费!

APP下载   全球官网 大陆官网

币安(Binance)最新版本

币安交易所app【遇到注册下载问题请加文章最下面的客服微信】永久享受返佣20%手续费!

APP下载   官网地址

火币HTX最新版本

火币老牌交易所【遇到注册下载问题请加文章最下面的客服微信】永久享受返佣20%手续费!

APP下载   官网地址

AI的世界正在以惊人的速度跃进,创造着一个全新的智能时代。

2024春节人工智能话题疯狂发酵。美国OpenAI公司正式发布其首个文本-视频生成模型Sora,能够根据文字指令创造出既逼真又充满想象力的场景,微软凭借OpenAI翻身市值超过3.1万亿美元,超越苹果成为全球市值最高公司。

随后,美国科技巨头谷歌(Google)宣布推出全球性能最强大、轻量级的开源模型系列Gemma,性能远超主流开源模型Llama 和Mistral,直接登顶Hugging Face开源大模型排行榜。

在人工智能BCI方面,马斯克宣称首个植入Neuralink大脑芯片的人类已经完全康复,患者仅用意念就可以控制鼠标,开创的脑机接口新时代。

英伟达发布了2023年四季度财务报告显示,英伟达去年四季度营收221亿美元,超各项数据均显著超出市场预期。技术创新催化下,全球人工智能概念股近期大涨,多股获机构扎堆关注。高盛预测2025年全球人工智能投资增长至1584亿美元,这只是冰山一角,根据最新的人工智能统计数据,截至 2023 年,全球人工智能市场规模估计为 4541.2 亿美元。按照每年 19% 的增长率,到 2032 年人工智能市场价值可能超过2.5 万亿美元。

我们可以确定,这个社会最终是要走向人工智能,那么人工智能下一步将会走向哪里?

人工智能和加速计算正在共同改变技术行业。下一波人工智能浪潮将是一种被称为具身AI的新型人工智能,即能够理解、推理并与物理世界互动的智能系统。英伟达创始人CEO黄仁勋指出了方向。

在探索智能与现实的交融上,WorldBrain坚持长期研究,建立技术储备,并取得突破新发展。WorldBrain是目前唯一一个将具身智能理念融入Crypto的卓越创新项目。通过模仿人脑建模的大型智能神经网络系统,WorldBrain将智能引入到具体的物体和机器中,赋予它们感知、理解和互动的能力。它的独特之处在于将智能从虚拟世界引向现实,探索智能与现实的交融之道。

具身智能的由来

具身智能的思想萌芽于人工智能诞生之初。

具身智能(Embodied Intelligence)这一概念,最早由人工智能之父图灵(Alan Mathison Turing)于1948年提出,embodied即人体的,intelligence即智能,结合在一起即具身的、离不开人体的智能。同时,图灵还提出了另一个与其相对的概念Disembodied Intelligence,即体外智能、离开人体的智能。 

1950年,图灵在其开创新论文《Computing Machinery and Intelligence》中,提出人工智能可能的两条发展路径:一条路是聚焦抽象计算(比如下棋)所需的智能,另一条路则是为机器配备最好的传感器,赋予它们与人类沟通的能力,并像婴儿那样学习的能力。这两条道路便逐渐演变成了非具身智能和具身智能。

什么是具身智能?

具身智能(Embodied Artificial Intelligence,Embodied Intelligence,Embodied AI,简写:EAI),又称“具身AI"“具身人工智能”,指一种基于物理身体进行感知和行动的智能系统,其通过智能体(可以是生物或机械)与环境的交互获取信息、理解问题、做出决策并实现行动,从而产生智能行为和适应性。

具身智能可以简要理解为通过创建软硬件结合的智能体,强调有物理身体的智能体通过与物理环境进行交互而获得智能的人工智能研究范式。我们拆分来看:一是“具身”,二是“智能”。具身的基本含义是指认知对身体的依赖性,即身体对于认知具有影响。认知不能脱离身体单独存在。此外,“具身”相对的概念是“离身”(Disembodiment),指的是认知与身体解耦。人工智能学科领域对智能的理解多偏人的抽象、逻辑能力及适应性能力。

atUSo5kNKdfZy3Bg1zgojwpbStAsxX3Hh0uaKiER.png

探索AI新维度 WorldBrain的具身智能新视角

具身智能是相对于传统的抽象化的人工智能,是一种能够感知并交互自身环境并具有自我感知能力的智能形态。具身智能可以以多种形式存在,如机器人、自动驾驶等。它们不仅能理解环境中的信息,还能以更高层次从复杂的环境中获取并处理信息。

WorldBrain是基于WorldModel世界模型创造出一个前所未有的分布式智能神经网络系统,通过模拟人脑的运作机制和思考过程,以更好地理解人工智能和认知科学之间的联系。它是未来实现“强人工智能”的重要组成部分,通过采用复杂的计算模型,包括神经元网络的建模、信息处理、学习和决策制定等,有助于深入研究人类认知的基本原理,为AI系统提供更有效的灵感。

WorldBrain为人类带来了更加智能化、高效的信息处理方式,为未来的数字社会构建起一个全新的智能生态系统。

首先,具身智能可以理解和响应自己的环境。这是通过先进的感知技术,比如视觉传感器、语音识别系统或者触觉传感器,使得具身智能能够获取环境中的信息。WorldBrain借鉴大脑地图和参考系的机制,通过感知输入(例如视觉和位置传感器数据通过使用人脑中的地图和参考系作为学习世界模型的基础,获得更准确的空间感知和环境理解能力。这将使得 WorldBrain 能够更好地与物理世界进行交互,并实现更高级的认知和智能功能。

其次,具身智能能通过机器学习自我改善并学习新的技能。这通过大数据和深度学习技术,使得具身智能不仅仅是一个静态的系统,而是一个不断改善和学习的动态系统。WorldBrain 可以以无监督的方式快速训练,以学习环境的压缩空间和时间表示。通过使用从世界模型中提取的特征作为代理的输入,可以训练一个非常紧凑和简单的策略来解决所需的任务。

最后,具身智能能够和用户进行自然交互,理解并处理复杂的任务。人机交互上升到新的高度。不同于当前火热的大语言模型多模态多为文字、图片、音频、视频等之间的交互,具身智能重要表现是多模态+跨模态感知交互,尤其是视觉、语言和肢体行为的跨模态感知;这不仅需要人工智能,还需要人机交互、自然语言处理等多种技术。WorldBrain 可以通过感知和行为模拟来模仿人类的认知能力和行为表现,通过运动和身体表征模拟来模仿人类的运动能力 和身体表达,使得 WorldBrain 能够像人类一样进行复杂的操作和交互现。

具身智能是产生超级人工智能的一条可能路径。

具身智能是人工智能的新前沿,是向感知、理解、学习和适应等更高级别智能进化的关键一步。

《智源人工智能前沿报告》(2021-2022年度)指出:未来三年,基于虚拟世界、实时时空环境训练的具身模型会取得较大的发展,如自动驾驶、机器人、游戏中数字人等······未来五到十年,超大规模预训练模型(信息模型)和具身模型将会结合,成为‘数字超人’,在知识能力以及跟环境的互动程度上,将比以往的人类都要强,具身模型和机器人也将结合,在物理世界出现能力比人类还要强的无人系统,即‘具身超人’。乐观估计,在未来三十年,数字超人和具身超人可能会结合,最终诞生超级人工智能。

Sora一镜到底及多镜头画面功能强大,逼真且充满想象力。然而Sora不懂玻璃破碎和液体溢出的顺序,不明白蚂蚁有6条腿的常识,也并不能推理时间和因果关系。妄图用拟合的方法来认识世界,纯拟合路线成不了AGI,AGI最终需要自监督地去触摸和学习世界。

人工智能的终极梦想是实现超级人工智能。

具身智能是产生超级人工智能的一条可能路径。具身智能是能够提升当前的“弱人工智能”认知能力的重要方式。人工智能可以通过与环境交互的渠道,从真实的物理或虚拟的数字空间中学习和进步。

WorldBrain将继续沿革图灵所开拓的具身智能(Embodied Intelligence)的道路前进,追求在人工智能技术与现实世界之间构建更加紧密、更加自然的联系。尽管目前还处在初级阶段,但其打破人工智能与现实世界之间的障碍探索已经展现出无限的潜力。


The world is leaping forward at an amazing speed to create a brand-new intelligent era. The topic of artificial intelligence in the Spring Festival is crazy fermentation. American companies officially released their first text and video generation model, which can create realistic and imaginative scenes according to text instructions. Microsoft surpassed Apple to become the world's highest market value company by turning over the market value of more than one trillion dollars. Subsequently, American technology giant Google announced the launch of the world's most powerful and lightweight open source model series, which far exceeded the mainstream open source model and directly. Top of the list of open source big models In artificial intelligence, Musk claims that the first human implanted with brain chips has fully recovered, and the patient can control the new era of brain-computer interface initiated by the mouse only with his mind. NVIDIA released the financial report for the fourth quarter of last year, showing that NVIDIA's revenue in the fourth quarter of last year exceeded the market expectation. Under the catalysis of technological innovation, global artificial intelligence concept stocks have recently soared, and many companies have gathered together to pay attention to Goldman Sachs' prediction that global artificial intelligence investment will grow to 100 million in 2008. This is just the tip of the iceberg. According to the latest statistics of artificial intelligence, the global artificial intelligence market is estimated to be billion dollars by the end of 2008. According to the annual growth rate, the value of the artificial intelligence market may exceed one trillion dollars by 2008. We can be sure that this society will eventually move towards artificial intelligence, so where will artificial intelligence go next? Artificial intelligence and accelerated computing are jointly changing the technology industry. The next wave of artificial intelligence will be a new type of artificial intelligence called embodied, which can be understood. Huang Renxun, the founder of NVIDIA, an intelligent system for reasoning and interacting with the physical world, pointed out the direction. Insisting on long-term research, establishing technical reserves and making breakthroughs in new development is the only outstanding innovation project that integrates the concept of embodied intelligence. It introduces intelligence into concrete objects and machines by imitating the large-scale intelligent neural network system of human brain modeling, giving them the ability to perceive, understand and interact. Its uniqueness lies in leading intelligence from virtual world to reality. The idea of embodied intelligence sprouted at the beginning of the birth of artificial intelligence. The concept of embodied intelligence was first put forward by Turing, the father of artificial intelligence, in 2000, that is, the combination of the intelligence of the human body, that is, the intelligence of the human body can not be separated from the intelligence of the human body. At the same time, Turing also put forward another opposite concept, that is, the intelligence in vitro leaves the human body. In his pioneering paper, Turing put forward two possible development paths of artificial intelligence: focusing on abstract computing. For example, the other way of intelligence needed for playing chess is to equip machines with the best sensors and give them the ability to communicate with human beings and learn like babies. These two roads have gradually evolved into non-embodied intelligence and embodied intelligence. What is embodied intelligence? It is also called embodied artificial intelligence. It refers to an intelligent system that senses and acts based on the physical body. It can be biological or mechanical interaction with the environment to obtain information, understand problems, make decisions and realize actions. Moving to produce intelligent behavior and adaptive embodied intelligence can be briefly understood as an artificial intelligence research paradigm that emphasizes that agents with physical bodies gain intelligence by interacting with the physical environment. We split it into two parts: one is embodied, and the other is intelligent embodied. The basic meaning refers to the dependence of cognition on the body, that is, the body has influence on cognition, and cognition cannot exist alone without the body. In addition, the concept of embodied relative refers to the decoupling of cognition and body. The understanding of intelligence in the field of industrial intelligence is biased towards people's abstract logical ability and adaptability. A new perspective of embodied intelligence is explored. Compared with traditional abstract artificial intelligence, embodied intelligence is an intelligent form that can perceive and interact with its own environment and has self-awareness ability. Embodied intelligence can exist in many forms, such as robot autopilot. They can not only understand the information in the environment, but also obtain and process the information from the complex environment at a higher level. It is based on the world model. Type I creates an unprecedented distributed intelligent neural network system to better understand the relationship between artificial intelligence and cognitive science by simulating the operating mechanism and thinking process of the human brain. It is an important part of realizing strong artificial intelligence in the future, and it is helpful to deeply study the basic principles of human cognition by adopting complex computing models, including neural network modeling, information processing, learning and decision-making, which provides more effective inspiration for the system and brings more intelligent and efficient letters to mankind. The way of information processing builds a brand-new intelligent ecosystem for the future digital society. First, embodied intelligence can understand and respond to its own environment. This is through advanced perceptual technology such as visual sensor, voice recognition system or tactile sensor, which enables embodied intelligence to obtain information in the environment and learn from the mechanism of brain map and reference system, through perceptual input such as visual and position sensor data, and through using the map and reference system in the human brain as the basis for learning the world model. Accurate spatial perception and environmental understanding, which will make it possible to better interact with the physical world and realize higher cognitive and intelligent functions. Secondly, embodied intelligence can improve itself and learn new skills through machine learning. Through big data and deep learning technology, embodied intelligence is not only a static system, but a dynamic system of continuous improvement and learning, which can be quickly trained in an unsupervised way to learn from the compressed space and time of the environment. The features extracted from the model can be used as the input of the agent to train a very compact and simple strategy to solve the required tasks. Finally, the embodied intelligence can naturally interact with the user to understand and handle complex tasks. The human-computer interaction has risen to a new height, which is different from the current hot big language model. The important performance of the embodied intelligence is multi-modal cross-modal perception interaction, especially the cross-modal perception of visual language and limb behavior, which not only requires Artificial intelligence also needs human-computer interaction, natural language processing and other technologies. It can imitate human cognitive ability and behavior performance through perception and behavior simulation, and imitate human athletic ability and body expression through movement and body representation simulation, so that it can perform complex operations and interactions like human beings. Realistic intelligence is a possible path to produce super artificial intelligence. Realistic intelligence is a new frontier of artificial intelligence and a key step in the evolution of higher-level intelligence such as perception, understanding, learning and adaptation. The frontier report of Zhiyuan Artificial Intelligence points out that the phantom will be trained based on the real-time space-time environment of virtual world in the next three years. 比特币今日价格行情网_okx交易所app_永续合约_比特币怎么买卖交易_虚拟币交易所平台

文字格式和图片示例

注册有任何问题请添加 微信:MVIP619 拉你进入群

弹窗与图片大小一致 文章转载注明 网址:https://netpsp.com/?id=60049

美化布局示例

欧易(OKX)最新版本

【遇到注册下载问题请加文章最下面的客服微信】永久享受返佣20%手续费!

APP下载   全球官网 大陆官网

币安(Binance)最新版本

币安交易所app【遇到注册下载问题请加文章最下面的客服微信】永久享受返佣20%手续费!

APP下载   官网地址

火币HTX最新版本

火币老牌交易所【遇到注册下载问题请加文章最下面的客服微信】永久享受返佣20%手续费!

APP下载   官网地址
可以去百度分享获取分享代码输入这里。
声明

1.本站遵循行业规范,任何转载的稿件都会明确标注作者和来源;2.本站的原创文章,请转载时务必注明文章作者和来源,不尊重原创的行为我们将追究责任;3.作者投稿可能会经我们编辑修改或补充。

发表评论
平台列表
美化布局示例

欧易(OKX)

  全球官网 大陆官网

币安(Binance)

  官网

火币(HTX)

  官网

Gate.io

  官网

Bitget

  官网

deepcoin

  官网
关注我们

若遇到问题,加微信客服---清歌

搜索
排行榜
扫一扫,加我为微信好友加我为微信好友