Chinese expert: The three biggest challenges for AI development

Artificial intelligence (AI) continues to rapidly transform a wide range of industries and areas of life, but it also faces a number of challenges. The renowned Chinese academic and professor at Peking University, Mei Hong, recently shared his in-depth assessment of the current state of AI development in China at a specialist conference. He identified three key problems and outlined possible solutions. His perspectives provide valuable food for thought for the industry and the future direction of AI technology.

Three central problems of AI development

Excessive expectations and exaggerated hype Mei Hong emphasized that AI is currently still at the peak of the technological hype curve. Exaggerated expectations lead to investments and resources being bundled in an unsustainable manner. He argued that the industry needs a “cooling off period” in order to be able to act more realistically and sustainably.
Generalization of success cases According to Mei Hong, the focus on a few success stories often leads to exaggerated generalizations and unrealistic promises. These exaggerated claims could disappoint both users and investors in the long term and harm the progress of the technology.
Excessive user expectations Many users see AI technologies as a kind of “miracle cure” that can solve all problems. These unrealistic expectations put development teams under immense pressure and make it more difficult to present practical, feasible solutions.

Lack of diversity in technological approaches Mei Hong criticized the current tendency of the industry to focus on a few dominant technologies such as large language models (LLMs). This one-sided focus could inhibit innovation in other potentially promising areas. However, diversity in technological approaches is crucial to ensure stable and sustainable progress in AI development in the long term.

The importance of data accumulation One key to the success of AI technologies lies in the availability and quality of data. Mei Hong emphasized that companies should focus on comprehensive data collection when developing AI. Even if there are currently no clear use cases, a strategy of “collect what you can, store what you can” could be of great benefit later on. One example of the challenges in this area is the so-called “Information cocoon” problem. Platforms that work with AI algorithms tend to suggest content that only corresponds to users’ existing interests. This can lead to people remaining in a bubble and not gaining access to more diverse information. This is where better data strategies and smarter algorithms are needed.

The limits of large language models and their future potential Mei Hong pointed out that current large language models are heavily based on statistical probabilities and often do not go beyond mere pattern recognition. This limits their ability to generate truly creative or context-specific content. For the future, he suggested that, like the Internet, these models should be a Open source development should go through. A globally shared, collaboratively maintained model could accelerate innovation and at the same time lower the barriers to entry for smaller companies and research institutions.

The reality behind the challenges

The problem of the information cocoon Mei Hong warned of the danger that users could remain trapped in a “bubble” due to AI-driven recommendations. Such algorithms hinder access to broader, potentially relevant content and pose a challenge for platform operators.
Success of content AI such as “text-to-video” Technologies such as “text-to-video”, which are based on large language models and extensive data sets, have made impressive progress. However, in areas without a sufficient database, such as in specialized industries, comparable progress has yet to be made.
Long-term data collection Mei Hong emphasized that collecting and storing large amounts of data over long periods of time is crucial for the success of AI applications. Companies should show patience, as the implementation of AI technologies often requires a phase of intensive adaptation and accumulation.

A realistic look at the future of AI

Despite its impressive progress, AI faces fundamental challenges that cannot be ignored. Mei Hong’s analyses show that the development of AI technologies requires a balanced combination of innovation, data strategy and realistic expectations.

He urges caution when it comes to exaggerated expectations and emphasizes diversity in technological approaches. At the same time, he calls for a focus on long-term data strategies and open collaboration. Only in this way can AI develop its full social and economic impact and change the world in the long term.

Sources: