On October 23, during the 2024 World Science and Technology Development Forum held in Beijing, the theme of “Innovative AI Governance to Build International Trust in Technology Governance” took center stage. At this event, academician Qiao Hong from the Chinese Academy of Sciences and Chairman of the World Robot Organization unveiled the “2024 Outlook on the Top Ten Frontiers of Artificial Intelligence Technologies.” This report includes four common AI technologies, three large-scale pre-training models, two embodied intelligence innovations, and one area of generative AI, aiming to spark both scholarly and public discussion on advancing AI development and its applications.
The report highlights key trends in AI technology such as: the rise of “small data and high-quality data,” “human-AI alignment: Building Trusted AI Systems,” “AI ‘Constitution’: Ensuring Compliance and Safety,” “Explainable Models: Making AI More Transparent and Trustworthy,” “Revolution of Pre-training Models under the Scale Law,” “Omni-modal Large Models: Breaking Data Barriers,” “A New Era of AI-Driven Scientific Research,” “Embodied Cerebellum Models: Providing Real-time Response Capabilities to Robots,” “Entity AI Systems: Empowering the Physical World,” and “World Simulators: Creating Infinite Possibilities in the Digital Realm.”
With regard to human-AI alignment, Qiao emphasized that simply relying on data and algorithms is not sufficient. It is vital to translate human values and ethics into reinforcement learning reward functions that guide models toward behaviors that align with human expectations.
As concerns regarding compliance, safety, and ethical issues associated with current AI systems grow increasingly prominent, establishing a supervisory model framework is essential. This framework aims to set clear standards and guidelines to ensure that all AI systems adhere to established principles during their development and use, thereby mitigating risks from the unregulated usage of AI technologies.
Qiao also discussed the significance of large-scale pre-training models, which, utilizing vast parameters and training data, enhance human-AI interaction and reasoning capabilities, ultimately increasing the variety and richness of tasks that can be accomplished.
In the realm of AI-driven scientific research, advanced models and generative technologies can significantly boost the efficiency of hypothesis generation, experimental design, and data analysis, leading to faster and more accurate research outcomes. This agile research methodology holds the potential to markedly increase the likelihood of discovering new scientific patterns, thereby accelerating the progress of scientific exploration.
The embodied cerebellum model, integral to robotic movement, focuses on addressing the integration of software algorithms with physical spaces and reconciling the trade-off between high-performance singular systems and versatile capabilities, enabling intelligent robotic systems to meet the intricate operational and real-time control demands of the real world.
Humanoid robots represent the pinnacle of entity AI systems, equipped with multimodal perception and understanding abilities that allow for natural interaction with humans. They can also make autonomous decisions and act independently in complex environments, promising wider application in challenging job scenarios in the future.
World simulators enhance model generalization capabilities by integrating data quality, diversity, training strategies, and regularization techniques, offering immersive and high-fidelity experiences through digital interactive engines. These simulators create richer and more diverse digital worlds, applicable in education, entertainment, and various other fields, while also enabling the creation of advanced digital environments.
According to Qiao Hong, AI’s applications span industries from smart manufacturing to smart cities, healthcare, and financial services. Its profound impact presents both challenges and opportunities. Therefore, it’s crucial to contemplate AI’s developmental trajectory, promote technological innovations and industry upgrades, and ensure the sustainable advancement of AI technologies.
In summary, AI’s evolution is accelerating technological progress and social transformation at an unprecedented pace. The fields of common AI technologies, large-scale pre-training models, embodied intelligence, and generative AI are teeming with possibilities and potential. The advancements in these technologies promise to offer more convenient and efficient lifestyles while driving innovation and growth across various sectors.
About the author