TMTPOST -- There are three key challenges confronting the AI technology and applications, which are engineering feasibility, profitability and digital divide, Liu Xiangming, Co-founder and Co-CEO of TMTPost noted on Friday at the panel "Gaining insights into a future where humans coexist with AI" during the 2024 World Internet Conference in Wuzhen, Zhejiang province.

As a major driver of the ongoing technological revolution, AI has reshaped production and lifestyles at the expense of potentially significant risks. The panel highlighted the global imperative to ensure “AI for good” and foster the responsible development of artificial intelligence.  

Liu pointed out that while large AI models show promising capabilities, deploying them in real-world business scenarios often proves difficult. For example, the media industry demands about one out of ten thousand error rates, far exceeding the 99% accuracy typically achieved by large models. Bridging this gap involves immense engineering effort, underscoring that the challenge is as much about implementation as technology itself.  

The industry is exploring how AI can generate sustainable profits, with two primary paths emerging: To C (Consumer) market leverages scalability to rapidly achieve unicorn status. However, the challenge lies in finding common ground among diverse user applications; To B (Business) market targets specific scenarios to deliver solutions. Identifying precise use cases can contribute to faster profitability.

Traditional enterprises may have the necessary scenarios and data but often lack the required talent and resources. The leap to AI is significantly more challenging than the earlier shift to digitalization, as AI requires more specialized expertise. Addressing how to help these companies bridge the gap remains a critical issue in the future.

Liu also noted that while the “killer app” for the internet era was search engines and for the mobile internet era was short videos, AI has yet to produce a definitive breakthrough application. In the age of AI, the emergence of numerous applications is essential. Only when a sufficient number of applications are successfully implemented, will a true “killer app” for AI arise.

Other speakers echoed Liu’s concerns while exploring additional dimensions of AI development. Li Anmin, the deputy director of the China Telecom Research Institute, highlighted the importance of multi-modal AI, cross-disciplinary tools, and security advancements. He stressed that the integration of spatial-temporal models and tools for high-quality data processing would be critical for AI’s development at the next stage.  

Fang Xingdong, Executive Vice Dean of the School of Media and International Culture at Zhejiang University, emphasized the need for global cooperation in AI governance. He pointed to the United Nations' Global Digital Compact as a landmark effort to align international AI strategies with human values. According to Fang, AI governance should prioritize human-centric approaches over purely technological fixes.

Han Meng, Director of the Intelligent Fusion Research Center at Zhejiang University, discussed the emergence of embodied intelligence. He explained that combining pre-trained models with control systems could bridge the gap between the digital and physical worlds, enabling AI to drive real-world applications. However, he warned that the integration of AI with physical systems comes with new risks, necessitating enhanced oversight and safety mechanisms.

As AI continues to reshape industries and societies, participants at the forum agreed on the need for multi-stakeholder collaboration and proactive governance to harness its benefits while mitigating risks.