The online ”Korean Grammar Checker” is a program bookmarked by people, no matter which field they work for, from those who write for a living to those who do not. Latecomers such as Naver and Daum portal sites also provide grammar-checking functions. Nevertheless, they have not been able to replace the Korean grammar checker because it accurately and meticulously detects errors based on the Standard Korean Language Dictionary. This checker was developed by Kwon Hyuk-Chul (Prof. Dept. of Computer Science & Engineering, PNU) and is also called the “Pusan National University Grammar Checker.” 

Any other grammar checker has virtually been no match for the Korean grammar checker so far. However, one year before Prof. Kwon's retirement, arose a new rival; ChatGPT. Although its performance is still lower than the Korean grammar checker, it sometimes can recognize and correct sentences that the checker could not. Prof. Kwon is currently grappling with ChatGPT to make up for the weaknesses of the Korean grammar checker. "My expectations for ChatGPT have become too high in my mind, yet, in my heart, I hope that ChatGPT will not be able to modify sentences," he said. On February 20th, “Channel PNU” visited Prof. Kwon, who said he had met "the second challenge and crisis in his life."

Q. Does ChatGPT correct as well as the Korean grammar checker?

A. After doing some research on Chat GPT for a few days, it was pretty good. Thanks to the deep learning approach: the way is different from us, it can now correct sentences we cannot. As it would cost us an enormous amount of money to implement such a system, we are finding some ways to surmount this difficulty. "What on earth is my life for?" said a professor who had devoted his entire life to studying film after experiencing his effort of research gone to waste by the deep learning of AI 10 years ago. What people like me feel about it is the concern that everything we have built up could collapse at once.

Q. How is it possible for Chat GPT to correct spelling mistakes that the spell checker could not?

A. The operating method is slightly different. The spell checker generates rules like "This can be corrected like this," and then it starts to correct incorrect cases in accordance with the rules. Take "결제" (payment) and "결재" (approval) as an example. You have to put all the words related to documents in "결제" (payment), and all words related to money in "결제" (payment). On the other hand, deep learning of Chat GPT surprisingly learns the rules by itself when an example is shown. Whether “paying” by card or “paying” funds exemplifies. As such, if you provide Chat GPT five examples, it understands the context and resolves even unfamiliar issues such as slush funds and commission payment, not to mention fund payment. There appeared to have been more than a year of Chat GPT putting in modifying spelling mistakes.

Q. Is the overall principle of Chat GPT basically like this, or is this only applicable to grammar mistakes?

A. Yes. Deep learning is getting massive amounts of data and statistically generating the most common answer. AI (Artificial Intelligence) is not built out of anything but from the data, humans put in. Regardless of right or wrong, AI brings out the one with the most data. Does it mean that it has no shortcomings? No. Seen the other way around, in practice, statistically infrequent sentences are not subject to being corrected. An American amateur recently won against the most outstanding Go AI in the world, far more proficient than AlphaGo, with 14 wins out of 15 games. The secret lies in targeting weaknesses. It is like, once your opponent makes an impulse movement with an example that has not been seen statistically, you are not supposed to anticipate it. 

Q. It is said that there will be an "AI-native" generation that has been using AI since birth. How will ChatGPT and AI change the future? 

A. Looking at the handwriting of a young politician who served as the representative of a political party a while ago, it did not even reach the level of a second grader in elementary school. Bad handwriting was a big issue back then, but now no one judges a person negatively for it. This is owing to the fact that people could write using computers. Now, writing will face changes in the same way. With the aid of ChatGPT, people may make a draft and refine it to create their own writing. When that happens, when people are given a piece of paper and asked to write something from scratch, they may not be able to write at all. Though their writing may be great as long as they heeded the advice of GPT and revised it. 

I am giving a lecture on Chat GPT this Friday (February 24th). When I asked Chat GPT, "How should I explain you to teachers?" I was told to approach differently according to different perspectives in a way that explains “to children” and “to experts” respectively. Wouldn't the result be much better if I were to modify my speech based on this? In addition to this, machine translation might make a contribution to multiple achievements. If I were to write a book, it would translate it into every language in the world simultaneously and publish it. There will be no need for people to do calculus in engineering colleges. It is likely that the trend of art will also go through some changes since machines are already capable of drawing pictures and composing music. However, I think that once technology advances, humans will also change in different ways to adapt to it. No one can predict exactly how the future will change.

Q. There may be many people who feel uncomfortable about this. 

A. It was reported that a condolence message for a shooting incident at the University of Michigan in the United States turned out to be written by ChatGPT. I personally thought this was inappropriate, but my wife disagreed. She exclaimed, "When you write the officiating speech, you write it by taking advantage of templates, and the college president's congratulatory speech is also written by someone else instead." (The professor and his wife are both professors at PNU.) It is a matter of social consensus. Regardless of whether it is desirable or not, wouldn't 95% of people follow my wife's idea in the future?

What concerns me is that diversity in society could vanish. Individuals create diverse styles, expressions, and terminology in their way when writing. But now, this is not the case anymore. Noam Chomsky once indicated that writing from now on is nothing but plagiarism. 

Q. I heard Chat GPT was asked to do an assignment and helped to get A+. What should we do with that?

A. No need to be afraid. We ought to take advantage of it. Turn the problem upside down. We could ask students to analyze and criticize the result found by Chat GPT to explain the pros and cons. Statistically, Chat GPT offers the most common data. There is also a saying that it is tailored to white men in their mid-30s to mid-40s, which means it limits the terminology. Criticizing and adding your own thoughts might be even better. You can still obtain accurate information from Chat GPT.

Q. What do you think about AI ethics?

A. It is surprising that ethics is not something that can be divided as clearly as 0 and 1. Deep learning systems like ChatGPT tend to avoid answering if the ethics level is too high. On the other hand, if ethics are set too low, ChatGPT lets out a wide range of responses. It is humans who set the limits and teach AI. I remember a system named Tay created by Microsoft, which learns on its own and talks. However, when a group of neo-Nazis taught it, it once said, "Hitler was great." It is a dilemma.

Q. Do you think AI can outperform humans?

A. Not at all! I don't know if I am seen as someone who merely praises AI. I have always believed that AI cannot outperform humans. Some people say that Chat GPT seems to have a “soul”, but it is a simple technology that only requires basic calculus skills at the second grade of high school students. It just looks like there is something in it due to its vast amount of data. Even though it managed to bring about quantitative changes, there is no element that raises fundamental qualitative change. I am sure some people insist that quantitative changes lead to qualitative changes (when computer performance improves, self-consciousness, self-awareness, and soul follow), but I do not. Under circumstances where AI endlessly develops, there will be countless ways that humans will find and surmount it at the end of the day. 

Q. You seem to have a strong faith in humans. What can the qualitative changes be?

A. I do not know because I do not know the essence of humans. But, this is surely not something of  statistical kind. Being reckless in competition is not a calculating thing and average results. Take examples like the ego of people who can lose but try and who climb Everest Mountain at the cost of their lives. Indeed, it is still possible for AI to imitate even that. On the other hand, don’t humans have to have something special such as a soul, that can be apart from science?

Reporter Jun Hyung-Seo

Translated by Yoon Seo-Jin