Bread That "tastes Like Love" Created By AI X Craftsmanship

Human-Computer Interaction 25/09/2024

3 main points
✔️ As a new approach to food development using AI, emotion vectors are extracted from conversation and lyrics data from romance TV shows, and "love bread" is developed based on the similarity between emotions and ingredients.
✔️ Tasting evaluation results showed a certain degree of similarity between AI-derived tastes and human sensibilities, with a particularly high accuracy rate of 65.6% for the "jealousy" flavor. Participants suggested the possibility of proposing new experiential value, not only in terms of taste but also concept.
✔️ It is expected that the findings of this study can be used to develop comprehensive product development, including appearance and package design, by combining more diverse AI technologies.

Food Development through Co-creation with AI: bread with a "taste of love"
written by Takuya Sera, Izumi Kuwata, Yuki Taya, Noritaka Shimura, Yosuke Motohashi
(Submitted on 19 Apr 2024)
Comments: Accepted to GenAICHI: CHI 2024 Workshop on Generative AI and HCI
Subjects: Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)

code：

The images used in this article are from the paper, the introductory slides, or were created based on them.

Introduction

The feeling of love can be sweet and sour, bitter and sometimes complex.What if we could express this feeling of love through taste? NEC and the long-established bakery Kimuraya Sohonten have collaborated on a project to realize such an unusual idea using cutting-edge AI technology.

From the conversation data of romantic reality TV shows and lyrics of songs about love, the AI extracts emotion vectors representing various love scenes such as "first meeting," "date," "jealousy," "lost love," and "both love". Meanwhile, another AI analyzed the emotion vectors represented by the names of 183 kinds of fruits and sweets from lyrics data of about 35,000 songs.

The AI recommended ingredients with high similarity in these two emotional vectors to the bakers, who then devised the taste and appearance based on these recommendations, and developed five types of "love bread.Do the flavors derived by the AI really express the human emotion of love? We take a look behind the scenes at this unique approach that uses AI to explore the relationship between emotions and taste.

Related Research

In the food industry, it has become increasingly important in recent years to develop products that appeal not only to taste and nutritional value, but also to consumers' experiences and emotions.For example, products made in collaboration with popular movies and TV programs have contributed to increased sales by building attachment to the work into the taste experience.Traditionally, however, the development of new products has relied heavily on the intuition and experience of craftspeople, limiting the ability to create new experiential value.

IBM's "Chef Watson" has used AI to analyze thousands of combinations of more than 50,000 ingredients to develop original granola products tailored to user preferences. This facilitates the exploration of new tastes, and the preference data is utilized in new product development.Also in Japan, research has been conducted on a unique food development method that uses AI to translate data from newspaper articles into tastes that reflect the atmosphere of the times.

Proposed Method

In this study, we developed an AI-based "love bread" through the following three processes.

Conversation Analysis

AI analyzed 15 hours of conversation data from the cast of a romantic reality show and classified them into five scenes: "first meeting," "date," "jealousy," "heartbreak," and "both love". Thirty-two different emotion tags were assigned to the conversational sentences of each scene, and a 32-dimensional emotion vector was calculated for each scene.

Lyrics Analysis

From a database of approximately 1 million song lyrics, we extract approximately 35,000 songs that contain 183 different food ingredients, such as fruits and sweets. As in Process 1, emotion tags were assigned to each lyric, and a 32-dimensional emotion vector was calculated for each ingredient.

Food Recommendations

We calculated the Cosine similarity between the emotion vectors of the love scenes and the emotion vectors of the ingredients. The top 50 ingredients that were close to the emotional distribution of each scene were recommended to the baker as suitable ingredients for expressing romantic feelings.The baker selected the most compatible combination of the 50 ingredients recommended by the AI and developed five types of "love bread. For example, "First Encounter" uses cotton candy for the dough, apple for the marble, and blue crunch for the topping.

In this research, AI was also used to generate product descriptions based on each scene and ingredient, using NEC cotomi, a large-scale language model developed by NEC, to automatically generate descriptions and manually make minor modifications to create text for use on packaging and special websites.

As described above, by making combined use of various AI technologies such as natural language processing and machine learning, we were able to develop a novel product that expresses romantic feelings through taste.

Experiment

In this study, the developed "love bread" was evaluated in two ways: a tasting evaluation by 31 general participants and an interview with the developer.In the tasting evaluation, participants tasted five kinds of bread and were asked to guess which bread represented which love scene.

The results are shown in the confusion matrix in Table 1, with an overall correct response rate of 43.8%. This is much higher than the expected value of 20% for a random selection, indicating that there is a certain degree of similarity between the AI-derived tastes and human sensibilities.In particular, the "jealousy" flavor had a high accuracy rate of 65.6%, suggesting that AI may be able to accurately express specific emotions through specific tastes.On the other hand, the taste for "mutual love" had a low hit rate of 26.7%, suggesting that it could easily be confused with the taste for "date. It seems that it is difficult in some aspects to distinguish and express similar emotions through taste.

Participants commented, "I think couples and families could buy a variety of flavors and enjoy sharing them," and "The bread itself was very tasty. I thought it would be interesting to develop the same taste and name into other foods and drinks." These comments suggested the possibility of proposing new experiential value to consumers, not only in terms of taste but also in terms of concept.

In an interview with the developer, we learned that he came up with the flavor and hue combinations based on the list of ingredients recommended by the AI.In particular, for the "jealousy" flavor, they struggled to adjust the color to express the emotion.They also indicated that working with AI stimulated their creativity, saying, "By utilizing AI, we were able to challenge ingredients and combinations we had never tried before, and discover new possibilities in baking.

These evaluation results indicate that the AI-based food product development method proposed in this research has the potential to contribute to the creation of new taste experiences that appeal to people's senses.Furthermore, by combining various AI technologies, such as image generation AI, in addition to taste and ingredient recommendations, it is expected to be deployed for comprehensive product development, including appearance and package design.In addition, it will be possible to develop products that better meet consumer needs through two-way communication, for example, after a craftsman creates a prototype, he can input feedback to the AI and ask it to suggest further optimized ingredients and designs.

Conclusion

Inthis study, we proposed a new approach to food development using AI and confirmed its effectiveness,showing that the food ingredients proposed by AI have high similarity to human sensibilities and can contribute to the creation of new taste experiences sought by consumers.

In the future, the use of AI technology not only for taste, but also for more diverse aspects such as appearance and package design, is expected to further expand the possibilities for food product development.Inaddition, two-way communication between artisans and AI will make it possible to develop products that better meet consumer needs, andcollaboration between AI and humans will play an important role in accelerating innovation in the food industry and providing new value to consumers.