Dentsu Inc. and ISID Launch "Project Takumi Tech" to Preserve Artisans' Skills Using AI
The text of the Dentsu Inc. news release distributed on May 29 is as follows.
May 29, 2019
Jointly Develops Image Analysis AI "TUNA SCOPE™" with Sojitz to Assess Tuna Quality
Image Analysis AI "TUNA SCOPE™" Developed
Dentsu Inc. (Headquarters: Minato-ku, Tokyo; President: Toshihiro Yamamoto) and Dentsu International Information Services, Inc. (Headquarters: Minato-ku, Tokyo; President: Ryoichi Nawa; hereinafter ISID) have launched "Project Takumi Tech" through their Open Innovation Lab. Against the backdrop of societal changes such as a declining birthrate and aging population, this initiative aims to preserve the skills of master craftsmen across various industries facing succession challenges by leveraging technologies including artificial intelligence (AI).
As part of this initiative, in collaboration with Sojitz Corporation (Headquarters: Chiyoda-ku, Tokyo; President and CEO: Masayoshi Fujimoto), the group developed "TUNA SCOPE™," a system where AI judges the quality of natural tuna based on images of its tail cross-section. This system was developed to preserve the skill of tuna appraisal, which faces a severe shortage of successors, and a proof-of-concept experiment was conducted in March this year.
The Dentsu Group will continue efforts to further improve the accuracy and practical application of "TUNA SCOPE™," including the ongoing collection of training data for the learning model and optimization of the analysis algorithm. Furthermore, by applying the acquired know-how to the preservation of "expert judgment" through AI in other industrial fields, we will contribute to solving societal and corporate challenges.
[Background and Overview of the Initiative]
Across Japan's traditional industries, the skills cultivated by artisans over centuries represent invaluable knowledge resources, the accumulation of humanity's experiential wisdom. Often described as "artisan intuition," this know-how constitutes tacit knowledge that cannot be systematized or verbalized. With the aging of the artisans who hold this knowledge, its preservation is increasingly threatened.
Project Takumi Tech was launched to enable AI to independently interpret artisans' tacit knowledge through the collection of vast amounts of training data and the application of image analysis technology utilizing deep learning, aiming to pass on these valuable skills to future generations.
This time, focusing on the expertise of tuna wholesalers—a profession requiring a decade to master—whose aging workforce poses a critical challenge in the fisheries industry, we developed the AI system "TUNA SCOPE™." This system inherits vast data derived from the artisanal skill of "tail-cut inspection" (※1), which involves visually assessing tuna quality by examining the tail cross-section, through machine learning.
[Details of the Proof-of-Concept Experiment]
This proof-of-concept experiment was conducted in two phases: ① Applying the beta version of "TUNA SCOPE™" to Marumi Foods Co., Ltd.'s (※2) tail-cutting inspection operations to verify its accuracy, and ② Verifying market viability by branding tuna rated as the highest quality by the system as "AI Tuna" (trademark pending).
① Development and Application of the "TUNA SCOPE™" Beta Version
A system was built to link cross-sectional photographs of tuna tails with the results of 4-5 tiered quality assessments by craftsmen, thereby acquiring tail-cutting inspection data for image analysis. Furthermore, based on the collected data, we performed tuning and selected deep learning algorithms to develop the beta version of "TUNA SCOPE™," implemented as a smartphone application. Test operations of this system during inspection tasks at Marumi Foods' Yaizu Factory resulted in successful tuna quality assessments with an 85% agreement rate compared to human inspectors.
② Sales and Marketability Verification of "AI Tuna" Rated Highest Quality by AI
Tuna rated highest by "TUNA SCOPE™" was branded "AI Tuna," with a dedicated product logo developed. It was served for five days at "Sanchoku Gourmet Kaiten Sushi Kantaro Tokyo," selling approximately 1,000 plates. Survey results showed high satisfaction with "AI Tuna" from about 89% of customers who ordered it.
<Implementation Overview>
Date & Time: Wednesday, March 27, 2019 – Sunday, March 31, 20 19, 11:00 AM – 11:00 PM
Location: Sanchoku Gourmet Kaiten Sushi Kantaro Tokyo
Furthermore, Marumi Foods Co., Ltd., which handles tuna processing and sales, is cooperating in this proof-of-concept experiment by applying the system to inspection operations, while Koozyt, Inc. (※3) is providing image analysis technology.
Furthermore, this project is being carried out by Dentsu Inc.'s creative team as part of the activities of "AI MIRAI," a company-wide, AI-specialized project team that brings together cutting-edge AI expertise and internal/external networks.
*1 Tail-cutting inspection: The process where craftsmen assess the quality of frozen tuna by cutting off the tail and examining the cross-section.
※2 Marumi Foods Co., Ltd.:https://www.yskf.jp/company/marumi-foods.html
※3 Koozyt, Inc.:https://www.koozyt.com/
"TUNA SCOPE" Website ( https://tuna-scope.com/ )
＜About Sojitz Corporation＞
Company Name: Sojitz Corporation
Representative: Masayoshi Fujimoto (President and Representative Director)
Headquarters: 2-1-1 Uchisaiwaicho, Chiyoda-ku, Tokyo
Established: April 1, 2003
Capital: ¥160,339 million
Business Activities: As a general trading company, Sojitz conducts diverse global operations across various sectors including automobiles, plants, aviation, medical infrastructure, energy, metal resources, chemicals, food, agricultural and forestry resources, consumer goods, and industrial parks. These activities encompass the sale and trade of goods, the manufacturing and sale of various products and provision of services both domestically and internationally, the planning and coordination of various projects, investment, and financial activities.
URL:https://www.sojitz.com/jp/
End
Dentsu Inc. News Release
http://www.dentsu.co.jp/news/release/2019/0529-009827.html
Was this article helpful?
Share this article
