(Image Source: Unsplash)

TMTPOST -- Embodied intelligence, world models, synthetic data, and other revolutionary advancements in artificial intelligence (AI) are among the major artificial intelligence trends in 2025, according to the Top Ten AI Technologies and Application Trends released by Beijing Academy of Artificial Intelligence (BAAI) on Thursday.

Wang Zhongyuan, the director of BAAI, said that AI is now at an inflection  point. The accelerating advancementof large language models are heralding  the arrival of the era of Artificial General Intelligence (AGI).

The integration of unified multimodal systems, embodied intelligence, and AI for science will strengthen AI’s ability to perceive, understand, and reason about the world, he noted.

These technologies will connect the digital and physical worlds, enabling breakthroughs in scientific research. Wang pointed out that BAAI, a new research institution focused on AGI, intends to ride on these top ten trends as a starting point to guide the development of AI technology, fostering collaboration across the industry.

Lin Yonghua, the vice president and chief engineer of BAAI, noted that there is shared anticipation of AI surpassing human intelligence and transitioning from the digital to the physical world. “We may take different paths to achieve AGI. The challenge lies in determining which approach will ultimately lead to AGI and how far we are from realizing this goal,” Lin said.

 The top ten AI technologies and application trends released by BAAI are as follows:

AI for Science: Revolutionizing Research Paradigms

The first trend identified by BAAI is AI for Science (AI4S), which is set to drive a paradigm shift in scientific research. According to recent data, nearly half of researchers worldwide predict that AI will positively impact their fields. This is in stark contrast to only 28% of U.S. researchers and 41% of Indian researchers who feel the same. These figures indicate a rapidly growing awareness of AI’s positive transformative effects on research methods and processes.

AI has already started reshaping scientific research, evidenced by the granting of the Nobel Prize in Physics and Chemistry to AI-related scientists. In 2025, multimodal large models are expected to further integrate into scientific research, enabling the analysis of complex multidimensional data. These developments will point to new directions in biomedical, meteorological, materials discovery, life simulation, and energy research.

Embodied Intelligence: The Dawn of a New Era

In 2025, embodied intelligence will extend beyond the mechanical subject to more advanced embodied cognitive systems. In terms of industry dynamics, nearly 100  humanoid robot startups in China may go through an industry consolidation, with fewer players staying in the market. Technologically, end-to-end models will continue to evolve, with breakthroughs in the development of smaller brain large models. On the commercial front, there will be a surge in embodied intelligence applications across industrial settings, with humanoid robots entering mass production. 

Multimodal Large Models: The Next Frontier

The third trend is the unification of multimodal large models, which allows for more efficient AI systems. While current language and multimodal models can simulate human cognitive processes, they still face limitations. New technologies that bridge multimodal data—integrating visual, audio, and 3D data from the start—are paving the way for the next generation of multimodal AI models. This shift promises to align various modes of data during training, creating more cohesive and powerful multimodal systems.

Scaling Laws and Reinforcement Learning: Advancing Model Generalization 

The fourth trend is about the application of Scaling Laws in the context of reinforcement learning (RL) and large language models (LLMs). As foundational models continue to scale, the "cost-effectiveness" of traditional pre-training approaches diminishes. Instead, after-training processes and scenario-specific scaling are being actively explored. RL will play a key role in discovering these new pathways, further enhancing the performance of AI models. 

World Models: Enabling Advanced Causal Reasoning

The fifth trend revolves around the accelerated deployment of world models. These models emphasize causal reasoning, allowing AI to develop more sophisticated cognitive abilities and logical decision-making capabilities. These advancements will drive AI applications in fields such as autonomous driving, robotics, and intelligent manufacturing, while also expanding the boundaries of traditional tasks and exploring new possibilities in human-machine interaction.

Synthetic Data: A Catalyst for AI Model Iteration

The sixth trend is that synthetic data will play a crucial role in AI model iteration and real-world application. High-quality data is increasingly seen as a bottleneck to scaling large models. Synthetic data can reduce the cost of manual curation and labeling while addressing privacy concerns. It will also mitigate issues related to data monopolization and high acquisition costs, enabling the broader application of large models.

Optimization for AI Native Applications 

As large models expand to cutting-edge devices like smartphones and PCs, this trend underscores the need for optimization to ensure practical deployment. AI models will face challenges in terms of inference costs and hardware limitations on these devices. Continuous iteration in algorithm acceleration and hardware optimization will be essential for the successful application of AI across diverse platforms.

Agentic AI: A New Product Paradigm

The eighth trend is the rise of agentic AI, which will reshape product applications. More autonomous and general-purpose AI agents will penetrate both work and life scenarios, becoming integral to everyday business operations. In 2025, we expect to see more intelligent systems capable of understanding complex business processes and providing customized solutions.

AI Super Apps: Who Will Dominate the Market?

The ninth trend is  the emergence of AI-driven super apps, though competition for dominance in this market remains fierce. AI capabilities in image and video processing are rapidly improving, alongside cost reductions in inference optimization. This convergence of advancements is laying the foundation for the emergence of AI super apps, though it remains unclear who will ultimately succeed in capturing the largest share of this market. 

AI Security and Governance: Ensuring Safe Development

The tenth trend is related to the ongoing efforts to improve AI security and governance. The improved autonomy of AI models raises concerns about potential risks and the unpredictability of complex systems. Experts stress the need for enhanced security measures to safeguard against potential failures while continuing to foster the safe application of AI technologies.

As AI continues to evolve, industry leaders are optimistic about its potential to drive economic and social change. With AI applications expanding across sectors such as finance, healthcare, and retail, the global AI market is expected to reach $227 billion by 2025, contributing nearly $19.9 trillion to global GDP by 2030.