This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Read our privacy policy

The Role of Computing and AI in Building the Metaverse

With connectivity, computing, and AI at the core of its digital infrastructure, the metaverse is the Internet's next evolutionary step towards hyper-real immersive experiences. In part two of this series, we look at how computing and AI will evolve to underpin the metaverse.

By Zhao Shefeng, Strategic Marketing Expert, Carrier Marketing Dept
Oct 2022

In the first article, we explored the six defining features of the communications networks that will power the metaverse:

  • Cubic broadband network
  • Deterministic experience
  • AI-native
  • Harmonized communication and sensing (HCS)
  • Security and trustworthiness
  • Green and low-carbon

In this article, we look at the evolution of the computing and AI capabilities that will work with and alongside connectivity to underpin the metaverse. After all, it is the powerful combination of networks, computing, and AI that will facilitate the ubiquitous sensing, limitless exploration, pervasive AI, and huge innovation potential that will characterize not just the metaverse, but also the future intelligent world.

Computing and the Metaverse

Today, the computing power of a smartphone far exceeds that of the supercomputer used for the Apollo program – humanity's first moon landing in 1969. The seamless integration of the digital and physical worlds, sensing and emotional interactions between humans and machines, and pervasive AI are all helping us achieve the capacity to think and act beyond what would otherwise be capable of. These capabilities will also help numerous industries make the jump from digitalization to intelligence.

Huawei's Intelligent World 2030 report estimates that, by 2030, general computing power will reach 3.3 ZFLOPS and AI computing power will reach 105 ZFLOPS, a 500-fold increase over 2020, generating more than 1 yottabyte of data worldwide annually. Computing is approaching its physical limits, so innovations in software, architecture, and systems are crucial. More importantly, the industry must jointly explore a new computing foundation; overcome the physical limitations of semiconductors; and make computing greener, more secure, and more intelligent.

Innovations in basic software and hardware are necessary to build a foundation that can support the sustained development of the metaverse, thereby enabling upper-layer application ecosystems to flourish. This will drive the high-quality development of the digital economy.

In terms of general computing and AI computing, Huawei focuses on innovation and research in five foundational technologies: general computing, AI computing, basic software, terminal operating systems, and AI development framework.

(1) General computing: building full-stack IT infrastructure and industry applications and services based on Kunpeng series’ processors.

(2) AI computing: building full-stack, AI computing infrastructure and applications based on Ascend series’ processors.

(3) Basic software: developing basic software by making the openEuler OS and openGauss database fully open source, following years of large-scale commercial use in the telecom sector.

(4) Terminal operating systems: developing the OpenHarmony OS for smart devices that shares kernel technology with openEuler. openEuler-powered equipment can automatically identify and connect to OpenHarmony devices, ensuring digital applications are better served in all scenarios. OpenEuler and OpenHarmony can collaborate closely for industrial applications, with OpenEuler applicable to highly reliable and deterministic industrial equipment and OpenHarmony applicable to highly interactive industrial terminals.

(5) AI development framework: providing development tools and platforms for AI developers at different levels through Huawei's layered and decoupled AI development framework. MindSpore, an open-source AI framework, natively supports foundation models to help developers quickly develop and train their own models. The Ascend application enablement software, MindX, helps developers quickly develop AI applications, facilitating simplified AI development. Operator developers can use the heterogeneous computing architecture Compute Architecture for Neural Networks (CANN) to revolutionize performance.

AI and the Metaverse

AI and AI-based models are essential to the metaverse. AI-based computing centers provide high computing power and big data to enable foundation models and foster new applications. Foundation models are platforms by nature, meaning developers can quickly develop industry-specific AI models that offer high accuracy simply by customizing a pre-trained foundation model according to specific scenarios. Foundation models are key to promoting the AI technology ecosystem. Huawei's new industrial AI development model is already boosting the development of the AI industry.

  • Peng Cheng Laboratory released PCL-L, the world's first 200 billion-parameter NLP model, as well as PCL-G, a foundation model for bioinformatics research.
  • Wuhan University released Wuhan.Luojia, an AI framework for the intelligent interpretation of remote sensing images, based on the Wuhan AI Computing Center.
  • The Institute of Automation of the Chinese Academy of Sciences released Zidong.Taichu, the world's first three-modal foundation model (image, text, and speech). In the future, more foundation models will be incubated to address local needs.

There have been successful cases of cross-modal retrieval and the generation of multi-modal foundation models. The model that handles image search by text offers high precision, with high relevance between the input text and semantics of the returned image. When we input something like "small blue bowl on a table containing fruit and nuts", the cross-modal retrieval system will return a picture that fits this description. With the addition of the speech modal, a three-modal, pre-trained model will realize the unified representation and mutual generation of the three modals, enabling image-based speech generation and speech-based image generation. With image-based speech generation, the system automatically generates speech that describes a video or image input. With speech-based image generation, the system automatically generates an image that reflects a speech input, for example, "a person flying a kite on a beach".

With progress in content generation, display, and interaction technologies, an increasing amount of digital-native content will become available, and the experiences of the digital and physical worlds are converging. MetaStudio, Huawei's cloud-based digital content production line, provides cloud-based media infrastructure for different industries. The MetaStudio system consists of three layers: core engine, platform, and pipeline. The core engine layer provides Huawei's proprietary cloud rendering engine and space engine.

The platform layer provides cloud conferencing, cloud desktop, digital asset management, and media AI capabilities. More specifically, it provides the industry's first 4K HD cloud desktop to support 60 fps and is capable of replacing the workstations typically used in the design industry. The pipeline layer provides capabilities for generating cloud applications such as modeling, rendering, knowledge comprehension, exhibition, and live streaming.

These core platforms and engines enable the efficient generation and operation of digital content in different industry-specific scenarios. MetaStudio has already been applied in many scenarios, including digital host, digital home decoration, cloud-based movie production, and virtual exhibitions.

With computing power becoming more abundant and models increasing in maturity, AI is being adopted in the core production systems of numerous industries to create real value. However, the adoption of AI in industries faces four major challenges: scarce AI computing power, shortage of AI talent, challenges in AI development, and difficulties in industry AI application.

ModelArts, Huawei's AI development platform, has continuously improved its foundation model training and inference optimization capabilities. It offers full support for MLOps and provides capabilities for continuous and collaborative iteration in AI development. Huawei is also developing an open AI ecosystem for ModelArts, bringing in development tools from the ecosystem, and streamlining the supply, demand, and learning of AI applications. ModelArts provides end-to-end AI model development capabilities, including data processing, algorithm development, model training, model management, and model deployment. These capabilities address the needs of computing resource scheduling, AI service orchestration, AI asset management, and AI application commissioning and deployment management.

Based on the MLOps concept, ModelArts's Workflow tool provides functions such as operation logging, monitoring, and continuous operation. The Workflow tool can be used to manage the entire lifecycle of AI development, operation, and post-operation monitoring, accelerating iteration from AI development to application and continuously optimizing results and experience. In addition, developers can use HUAWEI CLOUD’s AI training community, AI Gallery, to quickly build AI assets like Workflow and share related capabilities.

In the past, it took an average of half a year for AI application developers to adapt to the chips, systems, and AI frameworks of different cloud-edge devices, which was time-consuming and labor-intensive. ModelArts shields the differences between underlying software and hardware, enabling one-time development and full-scenario deployment of AI applications. In an oil and gas field inspection scenario, ModelArts deploys AI applications on AR glasses, mobile phones, and equipment room servers with one click, enabling high-performance AI applications such as 3D reconstruction and location recognition to run smoothly with AR glasses. This solution helps customers adapt to different chips, systems, and inference frameworks during application development, shortening the development period of AI applications, such as pipeline inspection and device fault identification. ModelArts shortens the cross-platform development adaptation period by 80% and improves inference performance by 2 to 10 times.

Looking toward the future intelligent world, key computing technologies will be upgraded and evolve in six key directions: cognitive intelligence, intrinsic security, green and integrated computing, diversified computing, multi-dimensional collaboration, and physical-layer breakthroughs.

ICT innovation is upgrading from 1.0 to 2.0

ICT innovation is upgrading from 1.0 to 2.0. Huawei has made many engineering and technological innovations in wireless, optical networks, and smartphones, which have generated a huge amount of business and social value. Innovation 2.0 is about making breakthroughs in fundamental theories and new inventions based on basic technologies.

Looking to the future, Huawei will continue focusing on making breakthroughs in basic science and pushing the frontiers of technology. We will take a vision- and assumption-driven approach as we work to identify new industrial requirements for ICT and tackle some of the biggest challenges facing the industry. Throughout this process, we will continue to advocate for broader collaboration between the industry, academia, and research institutes to help light up the future through innovation.

We are now well on the way to the metaverse. Over the next decade, the ICT industry will see hundreds of billions of connections, 10 Gbit/s broadband speed per person, a 100-fold increase in computing and storage capacities, and a 50%-plus increase in the use of renewable energy. The technologies that power the generation, transmission, processing, and the use of information and energy must constantly evolve. Based on these predictions and assumptions, Huawei has released nine technological challenges and research directions for the next decade:

1. Defining 5.5G to support hundreds of billions of diverse connections

2. Nanoscale optics for an exponential increase in fiber capacity

3. Optimizing network protocols to connect all things

4. Advanced computing power strong enough for the intelligent world

5. Extracting knowledge from massive data for breakthroughs in industrial AI

6. Going beyond von Neumann architecture for 100x denser storage systems

7. Combining computing and sensing for a hyper-reality, multi-modal experience

8. Enabling continuous self-monitoring for more proactive health management

9. An intelligent Internet of Energy for the generation, storage, and consumption of greener electricity

Through innovations and breakthroughs in key ICT technologies, we will realize stronger connections, faster computing, and greener energy, leading us towards the intelligent metaverse.

References:

Huawei Intelligent World 2030 report, September 2021

https://www.huawei.com/en/giv