Recently, Zhang Bo, an academician of the Chinese Academy of Sciences and honorary dean of the Institute of Artificial Intelligence at Tsinghua University, stated during his speech at the ISC.AI 2024 12th Internet Security Conference that current artificial intelligence lacks a theoretical foundation and only has developed models and algorithms targeting specific fields. Both software and hardware are specialized, leading to a small market, which is why a large-scale artificial intelligence industry has not yet emerged. This is the core issue.
Zhang Bo, Academician of the Chinese Academy of Sciences and Honorary Dean of the Tsinghua University Institute of Artificial Intelligence
In this speech, Academician Zhang not only pointed out the flaws and issues in current artificial intelligence technology but also provided directions for future improvements.
When considering foundational models,
three major capabilities and one major flaw must be considered.
In Academician Zhang’s view, due to theoretical limitations, the previous stage of the artificial intelligence industry must develop in conjunction with specific application fields. Therefore, the artificial intelligence developed in this stage is considered specialized artificial intelligence, or “weak” artificial intelligence. However, he also pointed out that the current foundational models have achieved generality in language issues, stating, “When we consider foundational models, it is crucial to consider three major capabilities and one major flaw; this is very important and serves as the starting point for our future industrial development.”
He explained that the strength of large language models lies in their powerful language generation capability, strong human-machine natural interaction ability, and strong ability to reason from one example to another. “The language generation of large language models belongs to an open domain and can generate diverse results, all of which can be understood by humans. Even if it is ‘nonsense’, we can still understand what it is nonsensically saying, which is very important. Human-machine natural language dialogue in an open domain was previously thought to require generations of effort to achieve, but unexpectedly, this goal was reached in 2020.”
Academician Zhang stated that the flaw of large models is “hallucination”, “Because we require diverse outputs, errors will inevitably occur. This error is very different from the errors machines make, which are often controllable. This error is intrinsic and will happen, and it is uncontrollable, so this is also a consideration we need to keep in mind for future applications.”
Combining the three major capabilities and one major flaw, Academician Zhang summarized that the current suitable application scenarios for large models require a high tolerance for errors. He indicated that from an industrial perspective, the application of large models presents a “U” shape—requiring diversity in the planning and design phase, while the service and recommendation phases also require diversity and a high tolerance for errors, but the middle phase needs to consider usage based on the situation.
Despite these issues, Academician Zhang still stated that “the model must be used,” “because once there is a model foundation, the efficiency and quality of applications will certainly improve. In the past, application scenarios were developed on empty computers, which are equivalent to being illiterate, but now with large models, the platform is at least a high school graduate, so development efficiency will certainly improve; this will be the future direction.”
Academician Zhang focused on analyzing the fundamental reasons for the occurrence of hallucinations, believing that the fundamental limitation of models lies in the fact that all the work done by machines is externally driven, where humans teach them how to do it, rather than the machines acting autonomously. Meanwhile, the results they generate are significantly influenced by prompts, which is a clear distinction from humans completing tasks under internal intentions.
The four development directions for large models are:
Alignment, Multimodality, Intelligent Agents, Embodied Intelligence
Academician Zhang introduced that there are four development directions for large models in the future, which are very important for improving large models.
The first is alignment with humans, “Large models lack the ability to judge right from wrong and cannot self-update; they are updated under human guidance. Without breaking through this point, machines cannot self-evolve. Large models require external prompts, so correcting the errors of large models under human guidance is our first task.”
The second is multimodal generation, “Multimodal generation will be very important for industrial development in the future, as we see that large models primarily generate text, but using the same methods to generate images, sounds, videos, and code will yield results comparable to human levels. The reason we can generate images so well now is mainly due to linking images with text. Thus, the essential breakthrough lies in text processing.”
The third is the concept of AI Agents, “We need to integrate large models with surrounding virtual environments to allow the environment to prompt them about their errors because one can only know right from wrong after performing a task. Therefore, the concept of intelligent agents is very important, allowing the environment to prompt the agents and giving them the opportunity to reflect and correct errors.”
The fourth is embodied intelligence, “By incorporating robots, we can enable large models to operate in the physical world. In the future, how to develop general-purpose robots? I believe it should be ‘software general, hardware diversified’. While Musk promotes humanoid robots, I believe the future will go beyond just humanoid robots.”
In Academician Zhang’s view, to develop the third generation of artificial intelligence, we must first establish a theory. The existence of large models cannot be explained by any theory, which causes various confusions and misunderstandings. As the scale of machine development increases, the lack of theoretical explanation leads to panic, so achieving safe, controllable, trustworthy, reliable, and scalable artificial intelligence technology is crucial. Until this field is fully developed, artificial intelligence will always have safety issues.
Source: Xinjingbao Beike Finance
Disclaimer: The purpose of this public account (ID: cnzifi) in publishing content and images is to disseminate more information. Copyright belongs to the original author and is not for commercial use. If there is any infringement, please contact us.
The application for the 2024 Hong Kong Fintech Week exhibition is now open!
The 2024 Zhongguancun “Fan Taike” Fintech International Innovation Competition is now open for registration!
