How does AI work in the realm of love and marriage?
*Note: These are Jeffrey Ding’s informal and unofficial translations -- all credit for the original goes to the authors and the original text linked below. Jeff is a Rhodes Scholar at Oxford, PhD candidate in International Relations, Researcher at GovAI/Future of Humanity Institute, and Research Fellow at the Center for Security and Emerging Technology. These are informal translations and all credit for the original work goes to the authors. Others are welcome to share excerpts from these translations as long as my original translation is cited. Commenters should be aware that the Google Doc is also publicly shareable by link. These translations are part of the ChinAI newsletter - weekly-updated library of translations from Chinese thinkers on AI-related issues: https://chinai.substack.com/
Source: Tencent Research Institute
Author ：老木 (Old Wood) -- author on light topics for Tencent Research Institute
Date: July 25, 2019
Original Japanese version of the marriage questionnaire and report: 日本内阁府.令和元年版少子化社会对策白皮书[OL]. https://www8.cao.go.jp/shoushi/shoushika/whitepaper/measures/w-2019/r01pdfgaiyoh/pdf/01gaiyoh.pdf
As we all know, the aging of Japanese society is very serious. For many in the younger generations, don’t even talk about giving birth, there’s not even an impetus for dating and marriage. This situation has caused distress in Japan.
Starting in December 2018, the Japanese government conducted a questionnaire surveying 3,980 unmarried men and women aged 20-40 who have a desire to get married. The Japanese media released the results of the survey on Japanese marriage preferences on January 8 this year. . According to the survey, only 27% of people think that "marriage is a natural matter of course,"[a] 68% of the respondents “are okay with not getting married", and 60% believe "even if they are married, they are okay with not wanting children.” The age segment of 30-39 years old had the highest proportion of those who can accept never being married, reaching 88%[b][c]. Reaching 70%, the ratio of those who believe it’s okay to not get married set a record high since the survey was first launched 25 years ago.
Directed at the people who answered the most questions, when asked if they had adopted any specific measures to find a match, 61.4% said respond that “they had not especially taken any measure.” In particular, 72% of men in their 20s answered “no specific measures.” According to Japanese media, it can be seen that young males now have a more negative/passive mindset toward finding a match.
And in mentioning the reasons why people are not planning to fall in love and get married, this report shows that 46.8% of the respondents replied, “have not met a suitable match,” 26.6% answered “not enough money to get married”, and 24% answered “cannot interact well with the opposite sex.” Thus, the report concludes that a major reason why 20-40 year-old men and women are not getting married is "they have not met a suitable match."
Based on the results of this survey, the Japanese government published the White Paper on “Societal Countermeasures for the Declining Birthrate” on the 18th[d]. The white paper points out that many people have the intention to marry but do not take particular measures (to find a match). The government needs to support these people, such as by creating a variety of activities in and out of the workplace for young people[e][f], and increasing their chances of finding the right person.
More importantly, the white paper also introduced a novel initiative in Ehime Prefecture, Japan: using AI to help young people match up with a suitable partner in love and romance.
How AI can address issues of love and marriage is not a completely new concept. One episode called “Hang the DJ,” of the fourth season of Black Mirror,” describes a sci-fi scene in an AI dating app: the computer, through understanding the user's personality characteristics in detail, constructs AIs that possess users’ personalities, and then lets the two users' AIs experience countless lives in the virtual world in order to observe the degree of match between the two.
In reality, AI naturally will not be that "science-fiction-y" in addressing issues of love and marriage. However, in recent years, there has been the emergence of many dating apps that do matching based on big data including one’s personality and appearance, but most of them stop in the direction of dating. Japan, on the other hand, hopes that AI can help users find a partner for the purpose of marriage and a longer lasting relationship.
*Summary of next two paragraphs -- modern dating scene has changed as path methods of introductions via friends no longer work. The piece goes on to describe how this specific trend manifests in Japan due to three main reasons:
First, the total amount (of friends) is small. Usually, our circle of friends is much smaller than our circle of people we do activities with. Those (possible matches) recommended by friends are usually their friends who have no partners but there are not many of these people. Thus the pool of (possible matches) that one can screen is very small. In today’s Japan, everyone’s life is relatively comfortable and content, so it is easy to hold on to the attitude of “I’d rather go without a ‘match’ than put up with something shoddy,” and keep looking (for a better match). You look and look and then it just passes. By contrast, through big data matching, you can basically test out the conditions of all possible matches in a specified area and time period. The amount of (possible matches) will be much larger.
Second, the time costs. Usually, if a friend recommends someone to you, to preserve face it’s necessary to go and meet them. If work is relatively busy, you have to dash to the other side of the city to see someone who there’s a high probability that you will not see again in this life. For many people this consumes a lot of energy. If you schedule three dates a week, then there may be a shadow hanging over the matter of meeting again afterwards. Especially for the otaku [g]boys and girls who find it very difficult to meet a match, a date could push them into an abyss of fear. For online dating, in comparison, the two sides meet each other in person only when their level of mutual approval is relatively high and there are no time concerns.
Third, (preserving) face. If a possible match is introduced by a friend, then there are quite a lot of things to worry about before you meet them. For example, after the date, if you don’t feel good, is that a refutation to your friend’s face? If you feel that there is a 10% chance that you could date, and the two sides try it but the end result is a breakup, will it be impossible to even be friends with your friend anymore?
As mentioned above, the model of "friend recommendation" has gradually become ineffective in love and marriage for people’s modern lifestyle[h][i]. It is no wonder that people have lamented that "being single for a time is cool, always being single is always cool."
In fact, this issue is also the biggest pain point of a group of Japanese white papers. As a result, Japanese officials began to explore the possibilities and practical implementations of how AI addresses issues of love and marriage. Ehime Prefecture, which was mentioned in Japan’s White Paper on “Societal Countermeasures for the Declining Birthrate,” is a relatively successful case of putting (these implementations of AI + dating) into practice.
Ehime Prefecture started a “joint date” (collective blind date) for single young men and women in 2008 -- organized by volunteers -- but due to the above factors, the initial results were not satisfactory. Starting in 2011, in addition to the collective blind date activity, AI was introduced to screen and match people to each other, and the success rates of the blind dates increased by 16%.
Product thinking tells us that every complex problem can be solved by breaking it down into several small problems. Here, we use the way of thinking of "discover problems - break down into small problems - solve these small problems" to see how we can solve the problem of love and marriage.
A pair of possible future lovers who don’t know each other must take a fancy to each other while browsing limited blind date information. This so-called "fancying the other person" usually needs to fulfill the three main conditions of "love at first sight," social status[j][k][l], and personality matching.
In traditional dating organizations and websites, these three conditions face several segmented problems:
1. Although dating and marriage websites have photos, the total number of users is also very impressive, where does a new user even start to look?
2. Social status such as occupation, salary, and education level can be directly filled in, but people are often reluctant to expose too much information about themselves to strangers.
3. This is not to mention someone’s personality, under the traditional model, “personality” is more like metaphysics, as there are no standards that can be visualized.
The way AI can address these three problems is through the three steps of quantification, feature extraction, and matching.
Take the visual of “love at first sight” as an example -- love at first sight seems to be a rather complicated issue as users often can't express what they like before they meet the right person.
But when broken down into characteristics, all the human appearance is is just a nose, a mouth, two eyes and two ears. Everyone's preferences in terms of appearance are ever-changing but they can’t be separated from changes in the data of each facial feature.
Although people can't accurately describe their exact figure for each element, we can “save the country through indirect means” through inputting elements of a “dream lover.”
The AI can first use the information the user inputs to see if they have a favorite celebrity or anime [m]character, and then retrieve whether other users’ photos in the database correspond to the key characteristics of the celebrity or anime character, thereby filtering the database to see if there is a suitable "dream girl/dream guy.”[n]
In addition, the "successful cases" of love and marriage can also be a training set for AI matching: we can enter the wedding photos of successful married couples[o] in the district office or city hall (civil affairs bureau) into the photo library (training set) for the AI to learn from. If party A starts the query, we first find a person similar to A in the photo library, then start to look up the photo of the other party B, and then match the person who approximates B in our registered database.
If we still can't find a suitable match to recommend at this point, what can be done? Don’t panic -- there is a saying that "people always like those who look like themselves." Those who are like ourselves is not only referencing similarity in appearances with ourselves but also personality and interests. So here we are going to consider how to solve the problem of personality matching: in terms of personality, we can still use the above feature point extraction method to deal with it.
Social status can be matched according to user's lifecycle or physical footprint -- e.g. consumption levels, activity areas, credit card bills, etc., and this information can stealthily emerge from the process of recommending two users to each other, which avoids the direct disclosure of salary, job position, and other information to strangers.
Of course, personality certainly becomes even more important when it comes to living together. In the past, the person who introduced both parties to each other, judged the level of personality matching according to their own subjective judgment. As with "love at first sight", the quantification, feature extraction and matching of personality are actions that cannot be done in the manual era. This is also the reason why many people encounter tragedy in love.
In this area, you can start with some behavior data of users surfing the Internet - what type of emoticons do they like to use when chatting, what kind of games users like to play, what kind of websites they like to browse, what kind of videos they watch, what kind of music they listen to or what movies they like, etc. From this, the reverse-engineered training set from the successful marriage cases can also contribute to the accuracy of this match.
Ok, at this point, imagine that we are one of the people in this matching system, then through the algorithm, we can solve the problem of traversing the most difficult phase -- going from getting to know someone to getting along with someone: there is a potential match who you connect with in both personality and eyes-affinity[p] — oh, they are not far from me, we can go out and meet — ahh, I also like to eat the dishes you ordered — right, right, I am also following this drama show recently — you actually play this game too, which region are you (in the game), let’s team up together (in the game). Afterwards, the two sides meet each other’s parents, and because they are well-matched in social and economic status[q], the parents of both sides nodded and agreed right away, and they are happily married.[r]