In 1991, Professor Yu Ruizhao, then the head of the Academic Affairs Office at Zhejiang University, personally taught us a course titled “Artificial Intelligence and Expert Systems.” More than 20 years later, when artificial intelligence has once again become a hot topic in society, few people mention “expert systems.” In recent years, I have spent a considerable amount of time researching intelligent manufacturing in factories. My feeling at the time was: the so-called “intelligence” in intelligent manufacturing has little overlap with the currently popular artificial intelligence. Therefore, I have repeatedly emphasized that the “intelligence” in intelligent manufacturing is Smart rather than Intelligent.
At the same time, I also vaguely felt that intelligent manufacturing seems to have deep roots with expert systems. About 20 years ago, my company sent me to Beijing to learn a software called G2. Looking back now, this was a software platform for expert systems. If intelligent manufacturing and expert systems are related, we should explore why this direction did not continue, what difficulties or even traps were encountered.
Thus, I posed a question: “Why have expert systems declined?”
To my surprise, my mentor, Mr. Hu Shangxu, also joined this discussion. Mr. Hu believes that an expert system should be a specific software combination that implements artificial intelligence technology for a particular real-world object; in this sense, AlphaGo also incorporates expert system technology. Expert systems are targeted at specialized fields and specific problems, rather than being general-purpose. An expert system must be applied appropriately and solve the right problems to be effective.
Mr. Hu’s viewpoints prompted me to think.
Back then, a premise of the G2 software emphasized that professional technicians were not good at computer programming. Thus, programming was made to express like natural language, which, while easy to understand, was very cumbersome. However, for programmers, the troubles brought by this programming outweighed the benefits. Therefore, enterprises might think: Why spend a lot of money on this software when we don’t lack programmers? So, in this sense, G2’s market positioning was problematic. According to Mr. Hu’s statement, it seems that the usage scenario was incorrect.
Thinking from Mr. Hu’s perspective, enterprises have actually used expert systems. At Baosteel, we imported a software called “Continuous Casting Quality Anomaly Model” from Japan, which is essentially an expert system; the CAQC from the Austrian Steel Association and the recent US River Company have also taken similar paths. However, unlike the traditional view of “expert systems”: this knowledge hardly requires deep reasoning, and is almost entirely described using simple “production rules” (which is essentially IF THEN statements).
Intelligent manufacturing may indeed be along these lines. But how can simple knowledge exert such a significant effect?
When I was doing data analysis, I spent a lot of time only to obtain some “well-known” knowledge, which left me very confused. My mentor, Mr. Wang Hongshui, enlightened me: putting the knowledge from human brains into computers and letting computers execute it is itself a remarkable task. After more than ten years, my understanding of this statement has gradually deepened: if we digitize the knowledge in human brains and put it into computers, we can achieve knowledge refinement, improve real-time performance, promote sustainable improvement, reduce human errors, facilitate knowledge inheritance, and enable large-scale complex optimization, even promoting a reduction in workforce.
Therefore, although the knowledge remains the same, the value manifests differently depending on where it is applied. This might be what Mr. Hu means by using it in the right place?
Since this valuable work has progressed so slowly in the past, why is that?
I think the root cause may be the issue of value manifestation: that is, the input-output ratio of this idea may not have been suitable in the past, but in the process of promoting intelligent manufacturing, the cost-effectiveness of this work will continue to improve, leading to a transformation from quantitative change to qualitative change.
For example, the result of information integration is that on-site problems have become more complex, people have become overwhelmed, and the necessity of knowledge digitization has increased; the level of basic automation has improved, making knowledge application or intelligentization more convenient; the development of ICT technology has expanded optimization space and reduced the cost of technical implementation; labor costs have increased…….
Thus, although the logic remains the same, the economics have changed. OK?