AI as the "Trump Card" for Technology Succession in Manufacturing: Its Challenges and Solutions | AI Annotation Blog

Table of Contents

1. Introduction
2. The Essence of Skilled Techniques: The Hard-to-Verbalize "Sense of Judgment"
3. Identification AI Supporting the "Inheritance of Judgment Skills"
4. Bottlenecks of Technical Inheritance AI: High-Difficulty Annotations
5. The Role of Specialized Services Supporting High-Difficulty Annotations
6. Summary
7. Features of Human Science Annotation Services

1. Introduction

In Japan, a manufacturing powerhouse, the aging of skilled technicians and the shortage of human resources in the manufacturing industry have been long-standing issues. Efforts to pass on skills using AI are not just a current trend but have been an ongoing area of exploration.

There have been many attempts to replicate the discerning ability of skilled technicians using AI technologies such as image recognition and anomaly detection, but widespread adoption at the field level is still far from sufficient, and challenges remain.

The skills, know-how, and judgment criteria possessed by skilled technicians are often personal and held in an ambiguous form without being verbalized. This is said to be a root cause of many cases where AI development stops at the PoC stage, and it is also a major factor hindering the development and introduction of AI that inherits skilled techniques.

Recently, in the context of technical succession, solving challenges through the utilization of generative AI is often discussed as a trend within manuals and knowledge management. However, this article separates generative AI and focuses specifically on the transmission of skilled techniques through discriminative AI from a more on-site perspective, explaining the current situation, challenges, and key points for effectively leveraging AI in technical succession.

2. The Essence of Skilled Techniques – The Difficult-to-Verbalize "Sense of Skilled Technicians" –

The skills of skilled technicians often include a large amount of "tacit knowledge" that is difficult to express in numbers, manuals, or other texts. In situations requiring fine judgments, such as visual inspections and finishing processes, there are "intuition" and "tricks" that instantly discern slight differences in color or shape, which ultimately have a significant impact on quality. As mentioned earlier, such skilled techniques tend to be difficult to verbalize and easily become personalized and tacit knowledge. Even if attempts are made to formalize them into manuals or the like, it is challenging to fully reproduce judgments based on senses and experience, making it not easy to pass them on to the next generation.

Identification AI is expected to be an effective means to pass on the “unseen judgments of skilled technicians” to the next generation and to utilize them widely. Development and implementation are currently progressing; however, in such AI development and deployment, the success greatly depends on how well the “sensibilities of skilled technicians” can be incorporated and prepared in the training data that determines AI accuracy.

3. The Inheritance of "Skilled Technicians' Discerning Ability" Supported by Identification AI

By the way, identification AI refers to AI technology specialized in "identification," using object recognition, region detection, and pattern recognition techniques for tasks such as image recognition and anomaly detection. In manufacturing sites, it is utilized for "discrimination," "classification," and "prediction" by analyzing cameras, sensors, and various data.

● Reproducing the Expert’s “Eye” with Image Recognition AI (Automation of Visual Inspection Using Image Recognition AI)
Image recognition AI, a representative example of identification AI, is a technology that learns the OK/NG judgments of skilled inspectors and automates the reproduction of those standards. In companies that have introduced this technology, by repeatedly adjusting the AI’s judgment results while comparing them with the decisions of skilled inspectors, the AI analyzes camera footage and still images to identify surface scratches, paint unevenness, and other defects, enabling the AI to approximate the “perspective of the expert.”
Through these efforts, a system has been established to digitally inherit the “expert’s eye,” preventing the reliance on individual judgment, and there are increasing cases where AI reproduces the judgment criteria of skilled inspectors in visual inspections and similar tasks.

Significantly improving inspection accuracy while reducing manpower for inspections that require expertise and are time-consuming (Toyota Motor Corporation)

● Reproducing the "Sense of Discomfort" of Skilled Technicians with Anomaly Detection AI
Skilled technicians detect abnormalities by consciously or unconsciously using their five senses to notice subtle differences in sounds or vibrations. AI learns the normal state from sensor data and detects deviations, quantifying the "sense of discomfort" to enable understanding, thereby contributing to early anomaly detection and quality stabilization.

These cases indicate one direction for the future manufacturing sites in terms of "passing down skilled techniques," but it is rare that AI can achieve the nearly 100% accuracy demanded by the manufacturing industry, and completely replacing skilled technicians remains difficult. It is important that AI implementation functions as a means to "complement and reproduce the judgments of skilled workers." Also, in the example of image recognition AI mentioned above, if image training data that takes into account site-specific factors such as light reflection, shooting conditions, and individual differences is not prepared and trained, false detections or over-detections will occur as a result. In other words, the design and preparation of training data—specifically, "what kind of data and at what granularity it is taught"—greatly influence the accuracy of AI.

4. Bottlenecks in AI Development for Skill Transfer – Highly Challenging Annotation

As mentioned earlier, for AI to accurately learn skilled techniques, high-quality and relatively large amounts of training data that embody the skilled techniques and experience are indispensable. However, in reality, converting the judgments of skilled technicians into data is not easy. Expertise is required from the AI design stage, such as deciding which parts of the images to focus on for learning and at what level to assign labels. Additionally, since skilled techniques are often not verbalized and remain ambiguous, they cannot always be fully captured in the annotation work specifications for creating training data, which can result in AI failing to reproduce the "sensibility of skilled technicians."

The expertise of skilled technicians is also required during the annotation work phase. However, having skilled technicians themselves perform a large volume of annotation work in-house is quite difficult in terms of both time availability and cost, and there is a risk of development delays. (To avoid this, companies often have no choice but to assign the work to available personnel within the company, and cases where the “sense of skilled technicians” could not be captured in the data as a result are frequently cited as reasons for outsourcing to annotation vendors.)

On the other hand, when outsourcing to annotation vendors, it is possible to expect securing a large amount of training data; however, due to a lack of knowledge and experience regarding the relevant technology, ensuring the accuracy of training data that guarantees skilled techniques becomes a challenge. Therefore, whether internally or externally, balancing the two elements of "quality" and "quantity" of training data is a major bottleneck in AI development aimed at the transmission of skilled techniques.

5. The Role of Specialized Outsourcing Services Supporting High-Difficulty Annotation

In AI development and implementation like this, whether conducting annotation by securing personnel in-house or outsourcing/delegating to a vendor, it is necessary and important to carry out the work after accurately passing on the skilled technicians' expertise and judgment criteria to the workers.

Each company will decide whether to perform the work in-house or outsource it based on their policies and priorities, but when outsourcing annotation and inheriting the know-how and techniques such as the judgment criteria of skilled technicians, it is not sufficient to simply carry out the work faithfully according to the verbalized annotation work specifications.

As mentioned at the beginning, it is difficult to verbalize the "intuition" and "tips" of skilled technicians, and it is neither feasible nor realistic to fully incorporate them into the specifications. Therefore, it is necessary to repeatedly perform small-lot annotations, have the customer's skilled technicians provide feedback each time, convey the subtle nuances of judgment criteria verbally in meetings or similar settings, and have the vendor pass these on to the workers using the same methods.

In other words, it can be said that it is difficult to achieve this unless the annotation vendor closely communicates and collaborates with the client, empathizing with them and flexibly resolving and responding to issues and requests.

Annotation vendors rarely possess specialized technical expertise in the relevant domain. However, annotation work in any field or subject inherently involves ambiguous and subtle nuances that cannot be fully verbalized. For this reason, human involvement in the work is unavoidable. Many annotation vendors, by necessity, have extensive experience dealing with such nuances, and are often well-versed in the methods of work and management needed to ensure and stabilize accuracy, as well as the key points to focus on. Therefore, even if they do not possess the specialized skills themselves, annotation vendors like those described above can be considered an effective option for outsourcing or delegated services.

6. Summary

As mentioned earlier, while identification-type AI is continuously being developed for the transmission of skilled techniques, it is also true that significant challenges still exist that hinder its introduction and success. Although the media tends to emphasize only the positive aspects, the development and implementation of AI require highly accurate and large volumes of data, and AI that cannot guarantee 100% accuracy is not a replacement for humans but rather a support tool. Therefore, it is necessary to enable collaboration between AI and humans. By confronting these challenges and realities and overcoming them step by step, the introduction and utilization of AI that broadly inherits the "intuition" and "know-how" of skilled technicians to the next generation and within the company will steadily evolve.

7. Human Science Teacher Data Creation, LLM RAG Data Structuring Outsourcing Service

Over 48 million pieces of training data created

At Human Science, we are involved in AI model development projects across various industries, starting with natural language processing and extending to medical support, automotive, IT, manufacturing, and construction, just to name a few. Through direct business with many companies, including GAFAM, we have provided over 48 million pieces of high-quality training data. No matter the industry, our team of 150 annotators is prepared to accommodate various types of annotation, data labeling, and data structuring, from small-scale projects to big long-term projects.

Resource management without crowdsourcing

At Human Science, we do not use crowdsourcing. Instead, projects are handled by personnel who are contracted with us directly. Based on a solid understanding of each member's practical experience and their evaluations from previous projects, we form teams that can deliver maximum performance.

Generative AI LLM Dataset Creation and Structuring, Also Supporting "Manual Creation and Maintenance Optimized for AI"

In addition to labeling for data organization and creating training data for identification-based AI, we also support structuring document data for generative AI and LLM RAG construction. Since our founding, manual production has been our main business and service, and we now also assist with organizing business knowledge and manual creation aimed at future generative AI and RAG implementation and utilization. We provide optimal solutions leveraging our unique expertise in the structure of various documents.

Secure room available on-site

Within our Shinjuku office at Human Science, we have secure rooms that meet ISMS standards. Therefore, we can guarantee security, even for projects that include highly confidential data. We consider the preservation of confidentiality to be extremely important for all projects. When working remotely as well, our information security management system has received high praise from clients, because not only do we implement hardware measures, we continuously provide security training to our personnel.

In-house Support

We also provide personnel dispatch services for annotation-experienced staff and project managers who match our clients' tasks and situations. It is also possible to organize teams stationed at the client's site. Additionally, we support the training of your workers and project managers, selection of tools tailored to your situation, automation, work methods, and the construction of optimal processes to improve quality and productivity. We assist with any issues related to annotation and data labeling that our clients may face.