From April 10 to 11, the 2025 China Mobile Cloud and Intelligent Computing Conference was held in Suzhou, Jiangsu. During the conference, Academician Tan Tieniu, Secretary of the Party Committee of Nanjing University, delivered a speech. He said, "The development of general artificial intelligence still has a long way to go, and the brute force AI development model that relies on piling up computing power and data is unsustainable."
Tan Tieniu Nanjing University Official Website
Elon Musk predicted in 2024 that general artificial intelligence would be achieved within two years; Sam Altman, CEO of OpenAI, believes that general artificial intelligence will be achieved within over 1000 days, or a few years; Demis Hassabis, last year's Nobel Chemistry Prize winner, also predicted that general artificial intelligence is about ten years away from us.
However, in Tan Tieniu's view, general AI that fully rivals human intelligence remains out of reach.
"They are overly optimistic," Tan Tieniu believes that artificial intelligence has just begun, and there is still a long way to go. "Even the most advanced AI systems have a significant gap compared to human intelligence, and there is still much that AI cannot do."
At the beginning of April, Sam Altman, CEO of OpenAI, publicly stated on social media that the release of GPT-5 was being delayed and might be released in a few months.
Tan Tieniu analyzed that the brute force AI development model that relies on piling up computing power and data is not sustainable.
Firstly, its performance improvement is not sustainable, meaning that with the same amount of data and computing power, the performance improvement is not as significant as before, so performance improvement is not sustainable; secondly, energy consumption is not sustainable, as much "intelligence" requires much "energy," which clearly contradicts the goals of energy conservation, emissions reduction, and sustainable development; finally, data support is not sustainable, as generative data has illusions and incorrect information, bringing noise rather than additional knowledge.
"Therefore, I believe that the development path should shift from 'brute force' artificial intelligence to 'clever' artificial intelligence. It is crucial to develop efficient and lightweight artificial intelligence," Tan Tieniu analyzed. "This can be achieved through algorithm innovation, theoretical innovation, engineering innovation, etc."
Meanwhile, Tan Tieniu believes that the development of artificial intelligence cannot be blinded by a single leaf. Generative artificial intelligence based on basic large models is not the entirety of AI, and AI empowerment is not limited to large model empowerment. He believes that artificial intelligence research should return to its roots.
In conclusion, Tan Tieniu mentioned that artificial intelligence is not omnipotent and has many limitations, but in today's world where artificial intelligence is widely permeating and the era of intelligence is approaching, it is absolutely unacceptable not to learn, understand, embrace, and apply artificial intelligence.
The full text of the speech is as follows:
You know, the concept of artificial intelligence was proposed in 1956. In nearly 70 years since then, artificial intelligence has roughly gone through three rises and two falls in three stages.
When artificial intelligence started, it proved a large number of mathematical theorems through logical reasoning, which was very exciting and made many people full of anticipation and optimism for the future development of humanity. As a result, these two predictions (see the following picture) made by the two regarding the development of artificial intelligence now seem unrealistic. These two predictions have not been fully realized to this day.
Because of such excessively high expectations and the huge gap between actual capabilities of artificial intelligence, it led to a low point in the development of artificial intelligence. So after sober reflection, people felt that they should strengthen the connection between artificial intelligence and applications.
In the second stage, breakthroughs were again made in applications, which once again filled some people with anticipation and excessive optimism. As a result, Japan invested national resources in developing what was called the fifth-generation computer, ultimately failing in the early 1990s, leading to the second winter in artificial intelligence research. Now we have entered a new phase of data-driven development. Similarly, everyone is full of anticipation and optimism.
I think the nearly 70-year history of artificial intelligence development gives us many insights. The most important thing I want to share with you is the last point: we must adhere to rationality and pragmatism.
Artificial intelligence has indeed made breakthrough progress in theory, technology, and application, and it is profoundly influencing the course of human civilization and the global landscape. I try to summarize the basic development trend of artificial intelligence with nine sentences.
First, specialized artificial intelligence is becoming increasingly mature. Second, significant breakthroughs have been made in large model technology. Third, generative talent is revitalized. Fourth, embodied intelligence and humanoid robots are receiving much attention. Fifth, scientific research driven by artificial intelligence is developing rapidly. Sixth, artificial intelligence is accelerating the empowerment of thousands of industries. Seventh, international competition in the field of artificial intelligence is becoming increasingly intense. Eighth, the social impact of artificial intelligence is becoming more prominent.
The ninth sentence may differ from the views of many colleagues; perhaps I am too pessimistic. In my opinion, general artificial intelligence still has a long way to go. There is no consensus on the current state of general artificial intelligence development. I quote the views of three senior figures.
The first is Elon Musk: he said that general artificial intelligence will be achieved within two years, as he mentioned last year.
Although there is no consensus on the definition of general artificial intelligence, it is generally considered that artificial intelligence that can rival or even surpass human intelligence is called general artificial intelligence. According to this definition, he believes it can be achieved within two years.
Altman seems quite optimistic, saying that general artificial intelligence will be achieved within over 1000 days, or a few years. The third person, Hassabis, who won the Nobel Prize last year, said that we are about ten years away from general artificial intelligence. These are their three viewpoints.
These predictions reminded me of the predictions from the 1960s and 1970s that I mentioned earlier. In my view, they are overly optimistic. General artificial intelligence has just begun, and there is still a long way to go. Because even the most advanced artificial intelligence systems have a significant gap compared to human intelligence, and there is still much that artificial intelligence cannot do. I can give many such examples.
General artificial intelligence that fully rivals human intelligence is still far off. Perhaps I am overly pessimistic, but this is truly how I feel.
Should we develop general, specialized, or multi-purpose artificial intelligence? In my opinion, practical applications require more multi-purpose artificial intelligence. Because every profession has its expertise, but this does not mean we don't need large models at all.
I believe that building on the refinement of specialized artificial intelligence, promoting research in multi-purpose artificial intelligence oriented towards practical applications is a feasible approach. Additionally, the brute force AI development model that piles up computing power and data is unsustainable. The reason is simple: performance improvement is not sustainable. The release of GPT-5 has been delayed, meaning that with the same amount of data and computing power, the performance improvement is not as significant as before, so performance improvement is not sustainable.
Moreover, energy consumption is not sustainable. "Intelligence" makes me feel that the more "intelligence" you want, the more "energy" you need to consume. This clearly contradicts the goals of energy conservation, emissions reduction, and sustainable development.
Thirdly, data support is not sustainable. What happens when the data runs out? We use generative data. You know, generative data has illusions and incorrect information, bringing noise rather than additional knowledge. Think about what kind of results training large models with such data would yield.
Therefore, I believe that the development path should shift from "brute force" artificial intelligence to "clever" artificial intelligence. Developing efficient and lightweight artificial intelligence is crucial. How to achieve this? By bio-inspired approaches, through algorithm innovation, theoretical innovation, engineering innovation, etc. DeepSeek is a great example.
Of course, promoting the development of artificial intelligence cannot blind us to the big picture. Generative artificial intelligence based on basic large models is not the entirety of AI, and AI empowerment is not limited to large model empowerment. Therefore, I believe that artificial intelligence research should return to its roots.
So for the next steps in development, I have a few personal opinions for your criticism.
First, there is no doubt that artificial intelligence is burgeoning. We must seize this opportunity firmly. First, we should firmly establish a rational and pragmatic development philosophy. We must not be overly optimistic. When there is a craze, we must calm down and see how it should be developed.
Second, we still need to build an ecosystem.
Third, we need to deepen the integration with industrial applications. Humanoid robots are currently very popular. In my view, if humanoid robots are only used for dancing, waving, or flipping, without finding an application entry point, I believe it is not sustainable.
Fourth, we must vigorously promote the empowerment of artificial intelligence in scientific research. Fifth, we need to focus on the cultivation of basic research talents. Sixth, we must unwaveringly promote open cooperation.
Finally, I want to say that artificial intelligence is not omnipotent and has many limitations. Of course, in today's world where artificial intelligence is widely permeating and the era of intelligence is approaching, not learning, understanding, embracing, and applying artificial intelligence is absolutely unacceptable.
This article is an exclusive article by Observer Network and cannot be reprinted without permission.
Original source: https://www.toutiao.com/article/7491988687404810806/
Disclaimer: The views expressed in this article are solely those of the author. Welcome to express your attitude in the button below to show your support or opposition.