NC

I oversee planning, development, and operation of the research on language AI. The field of research of our office is comprised of natural language understanding and creation, conversational technology, machine translation, search technology, etc., and we develop services applying the aforementioned technologies. In 2018, we launched a webpage on baseball-related information (PAIGE: Click the link to connect) and have been providing machine learning based AI weather news articles (link) as a result of joint research for AI media with the national news agency, Yonhap News Agency, since 2020. In addition, we research and service opinion mining, automatic hashtag creation, and spam filtering technology for operation of games.

The concept of our NLP (Natural Language Processing) technology started from a sentence, “How was yesterday?” For instance, let’s assume that there are two NC Dinos fans that used to talk much about baseball. In case a friend asks, “How was yesterday?”, the other would explain how the game went. I thought it would be nice if we could have the same fun that we had when talking to another fan when talking to AI.

As you can see, the AI that we dream of is the one that knows all about me, including my situation and interest, and explains to me through conversation whenever it is necessary. We dream about an AI that could react properly when I get annoyed and talk to myself and provide immediate answers to my questions.

Those that study language AI should have a special mindset. Developing a language AI is like traveling through an endless ocean. You will deal with extremely diverse data and understand that language is highly productive.

There is no one in the world that talks in exactly the same way as I do. Researchers should divide the countless language data into proper categories and develop framework for analysis in order to enable ‘natural conversation’, which is a destination that is yet to be defined. If you are doubtful that human emotion could be classified and think it is to consider human beings as machine, you will get into trouble, even from the beginning.

We are taking one step at a time in the right direction even though it is not easy. In the beginning, there was not much data for Korean NLP. We started by accumulating data in Korean. We still make multi-faceted efforts to get high-quality data through collaborating with academia and agreeing on an MOU with Yonhap News. Also, our office is a team of professionals from various areas, including computer science, AI, linguistics, and psychology, studying each other’s field of expertise and creating synergies.

It is a difficult journey. However, we have aimed to enable pleasant conversations between human beings and AI from the beginning, because conversations are the most certain means of communication, through which we could understand each other. We dream of a future where we could have pleasant conversations with AI, exchanging knowledge, understanding each other, and discovering fun.

I feel the most rewarding, when users say they really like using our service. I felt really proud when I saw a user’s comment that wondered if a review prepared by AI was ‘really written by a machine or not.'

However, you get to experience a series of troubles until you could commercialize a service targeting the public based on research results, because a large number of exceptions occur, when the service is to be applied to reality. Commercialization means to be able to respond to the aforementioned exceptions that rarely occur. For instance, let’s talk about the weather forecast published through Yonhap News. There is not much data for a machine to learn in case a forecast is to be made about a once-in-a-decade heavy rain fall or heat wave. On PAIGE, a human being may talk about something that the machine has never heard of. However, even in this case, AI should not stay silent. We need to decide an appropriate answer to respond to such a case.

However, these problems are difficult to be discovered in case if we stop our work after publishing a paper or filing a patent application. The difficulties that we experience while working with the data in reality may suggest a new research subject or a direction for further research. If you talk to the people that actually use the service, you get to understand where AI is the most needed. Unless it is commercialized, a technology will forever stay inside the lab.

When we first started a research project with Yonhap News, many asked if their jobs would be gone or not. There are many people that are worried if they would lose their jobs because of the advent of AI technology. However, I believe that ‘AI technology is not to replace human beings, but to help them’. For instance, if a doctor had to analyze and diagnose a disease after viewing 100 X-ray copies, AI helps reducing the number to 50, and by doing so, we could reduce the amount of required labor by half. We may also improve the level of quality to near 100% through receiving assistance from a machine, after a person delivers 90% of the quality. Human beings could focus on what only they could do.

In addition, there are a countless number of environments and jobs that damage human dignity and infringe upon their basic rights, which could be done by AI.

The PAIGE app combines all the technologies that we are studying. There are more than 30 deep learning engines applied to conversational technology. However, we continue to develop the technology, since the current version does not represent an end-goal in conversation with AI.

The initial version did not make any sense; it was a rudimentary system developed by engineering majors. There was no concept of “a good conversation with AI”, and there were many trials and errors where we tried many things from different angles, since there was no one with relevant experiences. The system went through many changes, as we watched people utilizing it after launch. In the beginning, people could only ask questions and get answers, and not many of them utilized the conversation function. The user base expanded, as professional game designers applied contents to conversations and responded to users’ needs.

When you try to apply NLP technology to a new service, the first things you need to check are the research results and data that have been accumulated. Researchers may make two mistakes; thinking that ‘what I have been doing is the best’ and setting unrealistic targets. In order not to make the mistakes, you need to understand the current progress in the research of the technology and the limitation in its potential for improvement.

Despite the busy schedule, the staff in our office host seminars and study together to keep up with the latest trend in research. By doing so, we could understand the research methodology and results of many researchers and always picture the starting point in our minds. We need to think once again when we believe that the technological development is complete; we need to think what we should do to overcome technological limitations while analyzing errors and how we could develop a differentiated technology that no one could overtake. One may believe that this is the true beginning.

I worked as a developer for four years, after graduating from college as a computer science major and joining an SI company. At the time, I used to work the entire week, both weekdays and weekends, I took a half-day off on Sundays. I thought I was working so much that I could kill myself. Of course, now I would have a different perspective on such a situation, but it was how I felt at the time. Therefore, I went back to school. When I joined the lab, neither AI nor NLP was a promising area. I was not being ambitious. Rather, I wanted to try something that was fun and where I could take the lead.

NLP technology is not a complete technology. Rather, it is the one that we should keep researching, and it is still on the path to grow. I believed that in order for us to further develop the technology and make it successful, the company should be able to understand its nature.

Games keep evolving. You may create new things whenever you want, and there is no limitation in your imaginations. Also, many people have conversations inside the games. Games provided the right environment for AI application, since all of these are maintained as data. Also, not only the CEO, but also my fellow researchers had an in-depth understanding of NLP and were serious about the technology. NC was an interesting place to be even from the beginning.

I initially believed that I should be able to make judgments based on science in order to develop differentiated technologies and that I should be never be emotional in order to do so. I thought it was natural to get objective assessments, as it was the case back in graduate school. I believed that in case there was someone that falls behind, I should try to improve them no matter what.

However, it was not the answer. Once a judgmental atmosphere is created, nobody can speak up. I had to encourage them, introduce best cases, and find the areas where they could fulfill their potential. Since then, the staff have suggested good ideas without me having to push them. Business was not enabled by talent of individual staff. I belatedly came to understand that good results were produced based on collective intelligence.

Now our office is most excellent in collaborating with each other. Considering the nature of NLP, technologies should be connected with each other in a horizontal manner to complete development of a single service. Related staff are gathered, have meetings, and create results without me having to make any orders. That is because we talk to each other a lot and have an atmosphere where we are willing to listen to each other no matter what the story is about.

In many occasions, we develop technologies and are not able to use them. Sometimes, our staff get exhausted while going through such processes. Whenever it happens, I want to tell them enjoy the process where we would gather around, talk together, dream, and make things together. Of course, it is not easy. I could be the one that stop it from being realized. However, I believe that it is never enough to say that I am proud of the staff that are doing great.

If you ask me the reason I work, the answer would be simple. I want to do something that would benefit the world. Whenever I imagine how to develop this technology and where to utilize it, the things that I do get so interesting, and they energize me.

Of course, there are difficult situations. No matter how serious they are, they seem complex once I am in them. Then, I talk to myself: “Again, this is a difficult situation. But, have I ever been in a situation where I could be carefree?” I also think that I “would find the answer, eventually”. There may be a situation in the future where I should leave the Company out of a sudden. Or, there may be situation where I would completely be burnt out and be unable to work. Well, I am optimistic, but I will move on as I always have been, telling myself that “I would be able to find something fun, again.”

Yet, there are not many people complimenting the quality of conversation with AI. It will not be easy to satisfy people, because they have seen high-quality conversations with AI in movies. However, we will continue our work, because it is an end-goal for NLP. We will continue to walk our journey, aiming to provide the conversational experience that is better than what we provide now to users.