Finding optimal color preference — Through the Machine Learning Technology
Master of Design Thesis progress report
Overview
In the era of the 4th industrial revolution, the importance of big data and Artificial Intelligence technology come to the fore. These technologies make us convenient in many aspects, however, the voice of concern is also not small. Especially in terms of automation, the experts are expecting in the future world, AI will take a lot of jobs. According to the World Economic Forum’s “The Future of Jobs Report2020”, AI is expected to replace 85 million jobs worldwide by 2025. Though that sounds scary the report goes on to say that it will also create 97million new jobs in the same timeframe [1]. Also, In an essay posted on Medium, AI guru Kai-Fu Lee — CEO of Sinovation Ventures and author of the 2018 book “AI Superpowers: China, Silicon Valley, and the New World Order” — posits that 50% of all jobs will be automated by AI inside of 15 years. Accountants, factory workers, truckers, paralegals, and radiologists — just to name a few — will be confronted by a disruption akin to that faced by farmers during the Industrial Revolution [2]. The job designer is one of the considerations. As many people know, some designer works were already replaced by AI. For example, there are many websites that generate logo design, many webpage building platforms offer proper design templates and photoshop can automatically be aware of the boundary of backgrounds and remove them.
In my opinion, in the design field, working with AI technology is an inevitable flow to current and future designers. The designers should understand AI and use it as a convenient tool, not being overwhelmed. However, the designers are not familiar with AI and some of them have a vague fear about this. Participants almost all stated that they “know very little” about how ML works. They characterized their ML literacy as “understanding at a very high level… [at the level of] knowing what a classifier is and what a label is.” Only one participant (P6) had taken any ML course; an online course taken well after graduation (Yang et al. 2018).
My initial idea of the paper was started from here.
I moved to Seattle this September. Since I took my entire first grad school year online in Korea, where has a completely different time zone, I was so happy to move and excited to meet my cohorts in person. Apart from that, the moving was tough and big stuff to me. I brought 6 baggage from my home country, so my apartment was full of my stuff. I wanted to buy something for organizing my things, so I turned on the amazon online app and searched the living boxes. At that point, what I realized was the recommendation system was not that helpful to my situation. Since I brought some stuff (such as mattress and knockdown drawer) from my home, I had a kind of theme color in furniture– gray. However, what the amazon recommendation system showed me was random shapes and colors of living boxes.
It seems the system was only considered the category, not which design the user wants. This problem made me more distracted and spend more time choosing the design I wanted. As a result, I had to finish shopping without finding the design I wanted. Based on this experience I became to think that how can we get reinforced recommendation results which considered people’s actual visual preferences — especially focused on the color and color combination — so that it can provide more satisfactory results.
Findings
Paradox of choice
We live a plenty life than ever before. With money, we can easily get the stuff that we need anytime, anywhere. There are many kinds of items on display in many stores and we can choose what we want from them. We are surrounded by plenty of choices.
Nonetheless, though modern Americans have more choice than any group of people ever has before, and thus, presumably, more freedom and autonomy, we don’t seem to be benefiting from it psychologically (Schwartz, n.d.).
This phenomenon is called the paradox of choices. The paradox of choice stipulates that while we might believe that being presented with multiple options makes it easier to choose one that we are happy with, and thus increases consumer satisfaction, having an abundance of options requires more effort to decide and can leave us feeling unsatisfied with our choice. [3] According to Schwartz in his book “The paradox of Choice”, this is induced by people’s regret about the opportunity cost they couldn’t experience. To prevent this phenomenon and induce consumers’ quick decision-making, many companies introduced recommendation algorithms to provide personalized recommended items. For example, Netflix and Spotify recommend new content based on the user’s play history, and Amazon also provides recommended items results through a similar process. Nowadays, recommendation systems have become the most core technology to almost all platforms.
Recommendation system operation
There are two main filtering methods for this recommendation system. One is content-based filtering and the other one is collaborative filtering.
First, Collaborative filtering is a popular recommendation algorithm that bases its prediction and recommendations on the rating or behavior of users in the system. This operates based on the user’s taste or liking information from their online history and predicts their future choice.
Second, As the name suggests, content-based filtering operates based on the similarity of content while collaborative filtering is based on the similarity of opinions on documents read in the past. It is recommending items in the same categories that users previously purchased or liked.
I also interviewed an AI expert, who is a former professor in computer engineering and currently teaching at Samsung right now. We discussed how the recommendation systems in current days work and is there anything we can reinforce. What I learned from this interview is current machine learning system is almost based on the visual part of the object. For example, there is a machine learning technology named Clustering. Clustering is the process of dividing given objects into several individual clusters (subgroup). In this process, there are no criteria to classify so the systems must figure out the similarities based on the object value themselves.
So, the system has to figure out the similarity based on the object’s visual values. This is why the AI confuses chicken and poodle, chihuahua and muffins.
This is the same for the recommendation system. The system mainly operates based on the categorized data. So when the people searched for a certain item, the algorithm considers items within the category as a similar item and provides them as a recommended for you result.
MBTI test and Color Image Scale
MBTI test — Myers-Briggs Type Indicator is a personality type test, which classifies people’s personality into 16 types and defines them as one of them through tests.
The purpose of the Myers-Briggs Type Indicator® (MBTI®) personality inventory is to make the theory of psychological types described by C. G. Jung understandable and useful in people’s lives. The essence of the theory is that much seemingly random variation in the behavior is actually quite orderly and consistent, being due to basic differences in the ways individuals prefer to use their perception and judgment. [4]
MBTI is popular because it relates to Type and once you know your Type (nature) you will be benefited throughout life. For example, people can be aware of their personality and tendency, and they recognize what they have to be careful related to the cons of personality, and how to improve their pros.
A Color Image Scale was devised by the use of an original color-projection technique, analysis of variance, cluster analysis, factor analysis, and the semantic differential method.
On this scale, every color has three attributes: warm or cool, soft or hard, and clear or grayish, which correlate with the notation hue, value, and chroma. The Color Image Scale is useful for describing similar and contrasting images of colors. The scale also allows the classification and correlation of various objects (shapes, patterns, clothing, foods, etc.) and the study of personal preferences in these and other areas (Kobayashi, 2009).
Both notions are meaningful because they materialized vague notions (characteristic and color) by classifying and categorizing them. People are usually insensitive to things that they can’t recognize. When the concept becomes tangible, people can easily understand and access the concept. Through my thesis project, I want to build some method that can provide people’s color preferences in a tangible way like the MBTI test. By justifying their color preference, they can bring awareness about their potential color preference to their daily life and apply them. Also, the recommendation system can be reinforced by adding certain color preference items in its system. It can provide more satisfactory recommended results based on this.
Poster Show Insight
On December 10th, MDes students had the poster show that can exchange feedback about the thesis progress.
I received a lot of feedback from various guests and the main opinion was, I had to narrow down more the thesis topic. It seems I need to pick the specific boundary (ID, IxD, VCD) and more drill down to the certain item. In addition, the necessity of one detailed scenario was pointed out. For example, in the painting the wall situation which color do I have to pick and that color-matched to furniture? Or I and my partner have different color preferences for home decoration. in this situation, through the AI recommendation system, that problem can be negotiated. Can I get proper color combination stuff?
I spent this quarter exploring the current AI technology and trying to narrow it down to my thesis topic. Since the field of AI has not been familiar to me, I rambled a lot. At the same time, although it is a piece of the iceberg, I am confident that I have been considerable progress in understanding some parts of AI technology.
References
Schwartz, Barry. n.d. “The Paradox of Choice,” 282.
Yang, Qian, Alex Scuito, John Zimmerman, Jodi Forlizzi, and Aaron Steinfeld. 2018. “Investigating How Experienced UX Designers Effectively Work with Machine Learning.” In Proceedings of the 2018 Designing Interactive Systems Conference, 585–96. Hong Kong China: ACM. https://doi.org/10.1145/3196709.3196730.
Kobayashi, Shigenobu. (2009). The aim and method of the Color Image Scale. Color Research & Application. 6. 93–107. 10.1002/col.5080060210.
[1] https://builtin.com/artificial-intelligence/ai-replacing-jobs-creating-jobs
[2] https://kaifulee.medium.com/10-jobs-that-are-safe-in-an-ai-world-ec4c45523f4f
[3] https://thedecisionlab.com/reference-guide/economics/the-paradox-of-choice/
[4] https://www.myersbriggs.org/my-mbti-personality-type/understanding-mbti-type-dynamics/
Kobayashi, Shigenobu. (2009). The aim and method of the Color Image Scale. Color Research & Application. 6. 93–107. 10.1002/col.5080060210.
