[The World on the Page] For ChatGPT Too, 'Imitation Is the Mother of Creation'

100 Million Users in Just Two Months
Mechanical Patchwork Answers: Fact-Checking Remains a Human Task
Actively Utilizing AI for Idea Organization, Learning Assistance, and Creative Activities

The interactive AI ChatGPT craze is no ordinary phenomenon. Developed by the nonprofit AI research institute OpenAI and released on November 30, 2022, this chatbot surpassed 100 million users in less than two months, becoming the fastest-growing app in the world. The previous record was held by TikTok, which took nine months.

According to Park Sang-gil's "AI Knowledge Understandable Even to Non-Majors," ChatGPT is an AI utilizing the massive language model GPT-3. ChatGPT's training data amounts to a staggering 45TB, with 570GB of refined data alone. It essentially memorizes all the sentences accumulated by humanity. The model is so large that it costs over 1.2 billion KRW just for one round of AI training.

Thanks to this, ChatGPT answers questions as if a human wrote the text. It provides quite accurate answers in sentence form to various questions. Explaining principles, it can code; given a topic, it can create PowerPoint materials. When the improved GPT-4 is released, answers will be much more accurate, and conversations will feel even more realistic. There is concern that students might submit assignments written by ChatGPT in schools.

First, there is a point to be cautious about. All answers ChatGPT produces are patchworks. It responds by mechanically stringing together the most plausible sentences from the learned data. If AlphaGo predicts the next move in Go and selects the best one, ChatGPT predicts the next sentence and mechanically outputs the most plausible sentence. Since there is no original creation, plagiarism is unavoidable. ChatGPT can write news articles, exam answers, and scientific papers, but it is not creative at all. In narrow topics, experts can easily find which texts were plagiarized.

There are predictions that ChatGPT will replace search engines like Google, but this seems premature. ChatGPT does not judge factual accuracy or consider semantic structures; it simply provides answers that sound plausible. Therefore, only by great luck can one encounter a good answer. When answers get longer, the content becomes jumbled, and sentences often repeat. In short, the satisfaction level with search results is not high yet.

When asked about the value or prospects of something, it becomes even more chaotic. No matter how advanced the AI is, it does not know what is meaningful. Without understanding meaning, it cannot know value or make predictions. Using ChatGPT often leads to the conclusion that meaning or prediction is not the AI's job. It either cannot answer important questions or merely rehashes obvious answers.

According to philosopher Kim Jae-in, ChatGPT has a fundamental limitation as long as it relies on language. Language does not convey only facts. It can present nonexistent things like dragons, unicorns, or round squares as if they exist. In short, language contains fiction, errors, and falsehoods, and sometimes conveys truth through falsehood. Artistic language is the pinnacle of this. ChatGPT is powerless against this power of language. It cannot judge facts or understand what truth is in context.

That ChatGPT has low understanding of plagiarism does not mean it is insignificant. In fact, humans are mostly not creative either and do not always engage in creative work. They also do not always think about meaning or make high-level value judgments. If used well, ChatGPT can greatly help lead a better life.

Above all, ChatGPT is a powerful brainstorming machine. When appropriate questions are posed on a specific topic to elicit decent answers, ChatGPT organizes existing ideas or materials well and produces documents. Most are ideas we are already familiar with, but there are also quite a few previously unknown ideas. This helps us make better decisions and stimulates and triggers our creativity.

ChatGPT is an excellent teacher for writing (coding, music, art, etc.). It is very good as an auxiliary creative tool. It generates dialogues suitable for given characters in specific situations and constructs story flows plausibly. It corrects spelling in documents we write, proofreads inaccurate sentences, filters rough expressions, and improves the quality of writing. It writes lyrics matching melodies, draws pictures fitting themes and styles, and codes according to conditions. The emerging search engine You.com operates similarly in a conversational manner but provides source documents as footnotes to avoid plagiarism.

This will change the way we work by entrusting ChatGPT with organizing existing ideas or materials, allowing humans to focus on creative thinking. Writing or learning will evolve into a process where humans add new expressions or ideas to the results of conversations with ChatGPT. Accordingly, the role of schools will also change to fit a hyperconnected society where constant dialogue with AI occurs. Rather than worrying about plagiarism and emphasizing simple transmission and reproduction of knowledge, students will be encouraged to develop critical literacy to doubt and verify given materials themselves and foster creativity.

For political parties or companies, using ChatGPT allows fast and accurate processing of large amounts of data such as voter or consumer information, and appropriate analysis to generate content with plausible response strategies. Customer services like public services or call centers are effectively automated.

In fact, writing news reports or summarizing public opinion on specific issues also becomes easier. This enhances responsiveness to specific policies or social events and reduces the likelihood of proposing wrong measures that go against public opinion. Moreover, by broadly reviewing alternatives submitted worldwide for given problems, it is possible to prepare measures and gain the power to lead the future with policies or strategies containing creative ideas.

Of course, the answers ChatGPT proposes are never creative. ChatGPT cannot think beyond mechanically outputting learned sentences and has no power to come up with new ideas. Fact-checking and human value judgment of content generated by ChatGPT are always essential. "Plausibility" and "correctness" are different. Furthermore, creativity arises only when well-organized ideas are taken a step further in reasoning. No matter how advanced AI is, it is not creative at all. Only humans can make AI creative.

Jang Eun-su, Publishing Culture Critic

This content was produced with the assistance of AI translation services.