物理人工智能是未来最大的机遇 - 2026-08-27

With the advent of superintelligent agents, there is a consensus that true intelligence goes beyond understanding static texts and requires grasping dynamic physical causality. This has given rise to physical artificial intelligence. It excels in multimodal environmental perception — seeing, listening, touching — and real-time interaction, bridging the gap between the digital and physical worlds.

Currently, AI development is converging toward physical AI, exploring it through multiple technological pathways.

By integrating LLMs with multimodal capabilities, physical AI provides intelligent brains for embodiment, bringing digital agents into the real world and heralding a new era of the intelligent economy.

Digital LLMs and intelligent agents are rapidly advancing toward self-iteration. The United States' models such as GPT-5.6, Claude Fable 5 and Grok 4.6, alongside Chinese open-source models from Zhipu AI, Moonshot AI and MiniMax, rank among the world's top performers. China's multimodal large models have closed the gap with the US, and in some areas even surpassed them. Seedance 2.0, for instance, has raised concerns in Hollywood.

From Manus to Openclaw, Harness and Hermes, disembodied agents are gaining autonomy and interconnectivity, emblematic of agentic AI.

In physical AI, advancements in vision-language-action alignment, world models and spatial intelligence are enhancing the sophistication of AI "brains", while lightweight visual and radar-based sensor navigation are boosting precise physical perception.

Humanoid robots, autonomous vehicles, and components such as dexterous hands, motors and advanced skins are rapidly enhancing agility. Currently, humanoid robots are the primary research focus, while autonomous vehicles are the most mature, with significant commercial potential.

The "jagged intelligence" in physical AI, highlighted by Moravec's paradox, means AI can surpass doctoral students in logic, math, and coding but still struggles with the perception, action, and emotional interaction skills of a toddler.

While LLMs create limited new jobs and rapidly displace a wide range of jobs, causing anxiety among coders and clerks, the "jagged intelligence" of physical AI will slow its economic integration and job creation.

The 2026 World Artificial Intelligence Conference signaled a trend toward physical AI. China's high-performance, low-cost open-source models, especially multimodal large models, are bolstering this trend, enhancing the capabilities of physical AI and embodied agents.

This advancement is expected to significantly boost the efficiency and safety of China's manufacturing, services, and mining sectors while creating new demand and jobs.

The rise of physical AI is an opportunity to shift the economy from K-shaped divergence to K-shaped integration. By coordinating industries — from new growth drivers to traditional drivers — within the "scissor gap" of K-shaped divergence, this trend can drive sustainable economic growth and facilitate effective industrial upgrading.

One imperative measure is to launch a "physical AI+" initiative in the scissor gap to boost emerging industries and upgrade traditional sectors to prevent their collapse. By integrating traditional, emerging, and future industries and phasing out outdated capacities, such an initiative can give momentum to sustainable growth.

Enterprises, especially small and medium-sized ones, along with financial institutions, should embrace this trend. By adopting a long-term perspective, they can establish themselves within the embodied intelligence industry chain, similar to how today's "new trio" — electric vehicles, lithium batteries and solar products — were developed. This will prepare them for a pivotal moment in the AI industry.

The embodied intelligence industry, well-aligned with China's manufacturing strengths, offers a strategic opportunity for industrial upgrading.

Models are advancing, enhancing both the cognitive and behavioral layers of embodied intelligence. Rapid developments in body components include the shift from hydraulic to electric actuation, offering lightweight, fast-response and precise control.

Progress also spans the perceptual layer with sensors, radar, vision and tactile sensing, as well as components such as dexterous hand joints, batteries, motors, harmonic drives, servo systems, lead screws, materials, and thermal management systems.

"Physical AI+" has the potential to transform industries fundamentally by reshaping equipment, spatial layouts, and human-machine interactions, among others.

This transformation can lead to logical restructuring and systemic enhancement, creating significant opportunities. Importantly, not every sector needs the most advanced AI.

Often, integrating stronger AI technologies with weaker ones results in higher efficiency and greater benefits.

Also, the governance of physical AI can generate new demand and employment opportunities. Its ethical, safety, and social governance is complex and involves multiple disciplines and stakeholders.

This can lead to the creation of various sectors and roles, such as model safety auditing, data privacy review, data cleansing, system security assessment, ethical evaluation, and social impact assessment.

AI is designed to benefit humanity. Embracing embodiment is crucial to enhancing industries, expanding demand, and boosting employment.

Source: China Daily