화학 분야의 인공 지능: 발전과 응용

최근 몇 년간 인공지능의 급속한 발전은 화학에 혁명을 일으키며 효율적인 분자 발견을 가능하게 하고 과학적 진보를 가속화하고 있습니다.

### 화학의 미래: 인공 지능과 분자 설계
* 강원대학교 이주영 교수는 인공지능과 화학의 통합에 대해 발표하며, **인공지능이 분자 설계에 혁신을 가져올 수 있는 잠재력**을 강조합니다.
* 🔬 1940년대의 초기 이론적 제안부터 **2012년 이후 딥러닝 알고리즘이 주도한 최근의 재부상**까지, 화학 분야 AI의 역사에 대해 논의합니다.
* 🎨 가상의 캐릭터에서 사실적인 사람의 얼굴을 합성한 예시를 통해 **AI의 생성 기능**을 선보이며, 새로운 데이터를 생성하는 **패턴 인식의 힘**을 보여 줍니다.
* 🌌 광활한 '화학 공간'을 탐색하여 기본 화학 법칙의 경계 내에 존재할 수 있는 **잠재적인 10^60개의 서로 다른 분자**를 추정합니다.
* 🤖 '완전 자동화된 분자 설계'라는 개념이 도입되어 사람의 개입 없이 **AI가 새로운 분자를 예측, 설계, 합성, 테스트**하는 미래를 상상합니다.
* 🧠 자동화를 통해 화학자들이 더 복잡한 문제와 창의적인 솔루션에 집중할 수 있게 될 것이라는 관점에서 **화학의 전통적인 역할을 대체하는 AI에 대한 우려**를 해결합니다.
* 🌐 화학에서 AI의 잠재적 응용 분야는 **분자 발견을 넘어 신약 개발, 재료 과학 및 환경 솔루션**에까지 영향을 미칩니다.
* 👩 강의는 **분자 구조, 성질, 다양성**을 연구하는 화학의 본질에 대해 논의하며 **물리와 생물학 사이의 독특한 위치**를 강조하는 것으로 마무리됩니다.

First, could you briefly explain why artificial intelligence is gaining attention in chemistry? Well, not only in chemistry but in other fields as well, a lot of data has accumulated over a long period of experimental research. Because we have gathered extensive data on what properties certain molecules possess, there is a growing recognition that we can make predictions that were previously impossible using theoretical methods by learning from this data. The relationship between artificial intelligence and chemistry seems to be a very interesting topic. Now, let's invite Professor Lee Joo-young to give a mini-lecture on science for about 8 to 10 minutes. The title of the lecture is Artificial Intelligence and Chemistry. Let's give a big round of applause.

I would like to ask for the professor's lecture. Hello, everyone. Today, I will be giving a lecture on the topic of artificial intelligence and chemistry. My name is Lee Joo-young from the Department of Chemistry at Kangwon National University. Recently, artificial intelligence has become a significant topic across various fields. Artificial intelligence is also referred to as machine learning, or in Korean, 기계학습. If we consider its dictionary definition, machine learning is a statistical method that enables computer systems to learn from data. However, while this explanation may seem abstract, the reality is that AI is rapidly integrating into our daily lives. Various applications of AI are now becoming part of our everyday experiences. However, there are still many challenges to address.

Not only that, but recently in the field of artificial intelligence, one of the very hot and highly discussed themes is generative models. Here, you can see eight portraits of individuals. In fact, these photos are a mix of real people and computer-generated images of fictional characters. Can you distinguish which photos are of real people and which are of fictional characters? It's not easy, right? So, I will reveal the answers at the end. Here, AI works by analyzing thousands, millions, or even tens of millions of photos of people. It recognizes patterns, understanding what a human face looks like, and then, based on those patterns, it can artificially create a new face of a fictional character.

Now, artificial intelligence has developed significantly in recent years. To briefly understand the recent boom in artificial intelligence, we should look back at its history. In fact, the concept of artificial intelligence was first proposed in the 1940s. However, at that time, computers were so slow and simple that they could not be used practically; it remained at the level of ideas. Then, in the late 1980s and early 1990s, as computers and algorithms advanced, the first boom in artificial intelligence occurred. Those in the audience who are probably over 30 may remember that in the early 1990s, discussions about artificial intelligence and fuzzy logic were prevalent in home appliances. This was when the first boom in artificial intelligence took place.

However, soon people realized that when they applied artificial intelligence to somewhat difficult problems, it did not work as well as expected. So, during that time, they thought it would be effective, but when they actually tried it, they found it to be quite useless. As a result, from the mid-1990s onwards, people lost interest in it. Then, starting in 2012, there was a sudden resurgence of interest in artificial intelligence. The reason for this was the emergence of a method known as deep learning, which has been widely discussed lately. In the field of computer science, there is a competition called ImageNet. In this competition, computer programs are given image files and tasked with identifying what is in the image, whether it is a dog or a frog, for example. In fact, until 2011, the competition...

The accuracy has now dropped to only 75%. In fact, this level of accuracy was already unsatisfactory. However, since the introduction of deep learning methods in 2012, the accuracy has rapidly improved to about 96% or more, almost reaching human-level performance. Now, there are three main reasons why artificial intelligence has developed so quickly in recent times. First, there has been significant advancement in deep learning algorithms. Second, various hardware technologies have progressed, leading to rapid improvements in hard disk storage devices. Nowadays, it has become possible to install about ten hard disks in a typical home PC. Lastly, graphics cards, which were originally used for graphic processing, have turned out to be...

People have come to realize that calculations used in AI can be applied very effectively, allowing for rapid processing of essential computations. As these three elements combine well, the accuracy of artificial intelligence has been improving rapidly in recent times. Consequently, chemists are now considering how to utilize this enhanced AI in their work. Recently, chemists have been trying to explore these possibilities, and one prominent approach is to use AI to discover new molecules. The concept of inverse design has been introduced in this context. Traditionally, there were two main methods for discovering new molecules or substances. The first method starts with existing known molecules.

Let's slightly modify the molecules so that they become the ones with the desired properties. In this diagram, the white part represents the area that possesses the properties we want. One way to achieve this is to mix various existing compounds randomly to create more molecules, and if we're lucky, we might end up with one or two that have the desired properties. These are the two main methods that have been widely used. However, recently, with the advancement of AI, we can teach chemistry to AI and have it predict the properties of molecules. By doing this, we can create molecules with the desired properties much more smartly and efficiently with the help of AI.

People have started to think about this, and so we now refer to it as inverse design. We usually call chemistry the science of diversity, and I would like to briefly explain how diverse the molecules in chemistry can be. In chemistry, you may have often heard about basic chemical laws in your classes. For example, carbon has four bonds, while oxygen has two. As long as we do not violate these basic chemical laws, we can freely imagine and design molecules. Chemists have speculated about the number of theoretically possible molecules that do not violate these basic chemical laws, estimating that there could be around 10 to the power of 60 different molecules that could exist.

하고 있습니다. 그런데 10의 60승이라는 숫자가 얼마나 큰 거냐 하면, 전 우주에 있는 모든 별의 개수를 현재 천문학자들이 약 10의 24승 개 정도 있다고 추측하고 있습니다. 그래서 이 말인즉슨, 우리 전 우주에 있는 별의 개수보다 사실은 가능한 존재, 가능한 분자의 개수가 훨씬 더 많다는 것입니다. 즉, 분자의 우주가 실제 우리가 살고 있는 우주보다 더 복잡하고 더 넓을 수 있다는 것을 의미합니다. 이렇게 넓은 분자의 우주를 탐색하기 위해서는 반드시 효율적인 방법이 필요합니다. 그래서 이런 효율적인 탐색을 도와줄 수 있는 중요한 툴로서 인공지능이 큰 역할을 할 수 있지 않을까라고 많은 화학자들이 믿고 있습니다. 그래서 이런 AI를 이용해서, AI가 미래의 화학을 어떻게 바꿀 것인지에 대해 최근 화학계에서 많은 사람들이 논의하고 있습니다.

One of the themes I am interested in is fully automated molecular design. This can be broadly categorized into three parts. First, just as we synthesized a person's face earlier, we teach AI chemistry so that it can create many plausible molecules that do not currently exist. Second, we will develop AI to predict what properties these created molecules will have. With this, we can select molecules that we believe will have the desired properties, and then, with the help of AI and robots, we can automate the synthesis and testing of these molecules. This is the dream we are pursuing. If these processes are completed, everything will be fully automated.

I expect that a new laboratory will emerge where we can continuously test new molecules without any human intervention. Currently, all these steps are performed manually, but if all these processes become automated, we could test new molecules at a speed that is dozens or even hundreds of times faster than we do now. This would allow us to discover new materials or potential new drug candidates that we are currently unaware of much more quickly. Many chemists are hopeful about this. I, too, have great expectations for how AI will transform chemistry in the future. Thank you for listening, and thank you, Professor Lee Joo-young, for the interesting lecture. Please give a big round of applause.

How did Professor Seokja view Professor Lee Joo-young's lecture? I found it very interesting. While listening, I thought that instead of me giving a one-hour lecture, it would have been better if I had done a three-minute science segment and Professor Lee Joo-hee had given the one-hour lecture. It seems like I'm overly praising him, but there is a reason for that. In fact, Professor Lee Joo-young is the first doctoral student of our Professor Seokja. You might not remember, but when we had our second lecture with Professor Lee Young-min from KAIST, he also invited his own student, Professor Park Jae-ho, as a panelist. Both of them are computational chemists, and it seems that scholars in computational fields have a special affection for their students. So, I can't help but ask how Professor Lee Jae-yong was as a student. Could you share a bit about that, Professor Seokja?

Can you tell me about it? It's a very difficult question, but it seems like it was just like now. I was a student who was just as fun, articulate, open to new things, and got along well with those around me, collaborating effectively in research. The next question might be a bit challenging: could you share what Professor Lee Joo-young was like as a professor? It's a very tough question, but the truth is that during research, especially in a degree program, there are times when it can be quite difficult. There are moments when research doesn't go well, and it can be mentally taxing. However, during those times, Professor Suk Jae-hak, as you may have felt during his lectures, is a very warm-hearted person.

It feels like she has always embraced me warmly, like an older sister or a neighborhood sister, which allowed me to complete my thesis well. So, I think she has always guided me warmly. It seems like I'm speaking very difficultly because I am in the same situation as I mentioned earlier, and since I was right next to them, I could see how those two interacted as a professor and a student at that time. I can confirm that what they discussed was not an exaggeration at all. However, while listening to Professor Lee Joo-young's lecture earlier, I started to feel a bit worried. The concern arises from the fact that whenever we talk about artificial intelligence, there is always the question of whether, if AI can truly do everything, there will be no jobs left for humans. So, now, regarding AI in chemistry...

If it becomes useful, won't that mean that chemists will have nothing to do? However, I think there will still be some work left for computational chemists like us. But if you look at the last slide, there were no organic chemists mentioned, so it seems like their roles might disappear. This raises concerns about whether all chemists will have their jobs taken over by AI. What do you think about this, Professor Lee Jeong-hyun? 

Yes, in fact, whenever this topic comes up, especially among experimentalists, there is a lot of worry about whether we will eventually have nothing to do. In my opinion, I am not 100% sure that AI will reach that level. Moreover, as I mentioned earlier, automated labs and automated experiments are currently at a stage where people are still somewhat...

We are now prioritizing the automation of simple repetitive tasks. These simple repetitive tasks can be entrusted to robots and AI, allowing people to focus on more complex problems and larger molecules. This means that we can explore more difficult issues or delve into the fundamental rules and principles of molecules that we previously could not tackle. Ultimately, I believe this will enable us to spend more time thinking about more creative problems. Our professor in Sokcho is highly capable and serves as the graduate committee chair in our department, overseeing the guidance of all graduate students. I would like to ask, regarding the future of chemistry, how do you see the application of artificial intelligence in the field of chemistry for the graduate students who will become chemists?

If you could give me advice on what preparations to make and which direction to take, I would first like to clarify that I am not in this position because of my own abilities; rather, I was appointed by Professor Shin Seong-min, the head of the department. Recently, we also told the new graduate students who will be joining us next semester that chemistry is not something you do alone. It is essential to communicate with those around you and learn new things together. Within the field of chemistry, there are various other stakeholders, and it is crucial to engage with people from different disciplines to foster collaboration. The field is changing rapidly, so if it seems like there is no answer, it is important to escape quickly. These various aspects cannot be handled alone; the era of studying and experimenting in isolation is over. I believe we need to work together, and that is what I would like to convey.

We talked about the use of artificial intelligence in chemistry, but in fact, the field of artificial intelligence has a very wide range of applications. If we think about it, although we didn't discuss it today, it seems that it could also be applied across the entire field of natural sciences and, broadly speaking, in all areas of science and technology. Professor Lee Joo-young, could you share your thoughts on how artificial intelligence might be utilized and what its prospects are in the broader context of natural sciences and science and technology? As I mentioned earlier, in chemistry, we are teaching artificial intelligence about chemistry so that it can help us discover molecules that could potentially become drugs, or molecules that, while they do not currently exist, possess very useful properties.

With the help of AI, if we can discover things much faster than we currently do, we may be able to assist in treating various intractable diseases that we cannot currently cure. Additionally, we could find new materials with functions that can help solve the various environmental issues we are facing today, such as components for batteries or solar cells. If we can identify such substances, I believe that in the future, this could serve as a great tool to significantly accelerate the advancement of chemistry. The topic of artificial intelligence will likely continue to be a hot topic, and I think we will keep discussing it. You will probably hear a lot about it in the Gauss lectures as well. However, one concern I have is that you all are now...

Today, as we listen to the lecture, we realize that this is precisely a lecture series on chemistry. However, it seems like we have discussed very different topics. So, I would like to hear from both of you, as computational chemists in all fields of chemistry, about your thoughts on chemistry itself. Professor Seokjeok, what do you think chemistry is? I always teach and believe that chemistry is the study of the structure and properties of molecules. In exploring these structures and properties, I think it can involve any mathematical theory, physical theory, or even artificial intelligence as a tool. Through these, we can create new substances with useful properties. Recently, you returned to the Korean chemistry community from your position at the University of Im Yongdae. What are your thoughts on chemistry?

What do you think? I have always thought of chemistry as being fundamentally two things. First, I believe that chemistry is the study of change. It explores how one substance transforms into another and how one state changes into a different state. Secondly, I see chemistry as a field that deals with diversity, examining how various atoms mix together to create interesting phenomena. In this sense, it is a study of atoms. Physics tends to delve into a single fundamental principle or the world of smaller particles, while biology addresses very large and complex problems. However, chemistry occupies a space in between.

Now, I think it's an interesting field of study that deals with various molecules, specifically three of them, which are very closely related to our daily lives.

[카오스 짧강] 인공지능과 화학_by이주용｜2018 가을 카오스 강연 '화학의 미스터리, CheMystery' 9강

KAOS는 ‘과학.지식.나눔’을 추구하는 공익재단입니다.

카오스 사이언스는 ‘기초과학’을 위한 채널입니다.

물리, 천문, 화학, 생명과학, 지구과학 등 자연과학과 수학 등에 관한 다양한 내용을 흥미롭게 소개하고, 
‘기초과학’과 ‘과학적 사고방식’의 중요성을 함께 실감하고자 합니다.

카오스 사이언스는 과학자들의 채널입니다.

실험실과 연구실의 과학자를 무대 위에 세워 그들이 어떤 연구를 하고 있는지 직접 들을 수 있습니다. 
또한 과학자 인터뷰, 실험실과 연구실 탐방 등을 통해 과학의 현장을 생생하게 전달합니다.

카오스 사이언스는 여러분의 채널입니다.

카오스의 핵심 가치는 ‘질문과 토론’입니다. 온라인과 오프라인의 다양한 강연과 과학 영상을 통해 여러분과 소통하고, 
다양한 질문에 대한 과학자들의 답변을 직접 들을 수 있습니다. 앞으로도 여러분의 다양한 목소리에 열심히 귀 기울이겠습니다.

Knowledge Awakening On Stage = ‘무대 위에서 깨어난 지식’

Since 2012 : 2012 제1회 카오스 과학콘서트 ‘유리알유희’


카오스 사이언스

인공 지능이 화학에 미치는 영향

인공 지능은 분자 특성에 대한 방대한 양의 축적된 데이터를 통해 이전에는 불가능했던 예측을 가능하게 함으로써 화학 분야에서 주목받고 있습니다.

AI를 화학에 접목하면 다양한 응용 분야와 과제를 해결할 수 있으며, 현재는 생성 모델에 중점을 두고 있습니다.

1940년대로 거슬러 올라가는 역사를 바탕으로 최근 몇 년간 AI의 발전은 화학 분야에 큰 영향을 미쳤습니다.

인공 지능의 부활과 화학에 미치는 영향

인공 지능은 1990년대 중반부터 관심이 감소했지만 딥러닝 방법과 하드웨어 기술의 급속한 발전, 효과적인 계산 응용 프로그램의 실현으로 인해 2012년에 다시 부활했습니다.

이러한 부활로 인해 인공지능의 정확도가 급격히 향상되면서 화학자들은 새로운 분자를 발견하기 위해 향상된 인공지능의 사용을 모색하고 있으며, 전통적인 방법에서 벗어나 인공지능에게 분자 특성을 예측하도록 가르치고 있습니다.

화학에는 관측 가능한 우주의 별 수를 능가하는 10^60개의 분자가 존재할 수 있는 것으로 추정되는 광활한 분자의 세계가 있습니다.

화학자들은 이 광활한 분자 공간을 효율적으로 탐색하는 데, 특히 완전 자동화된 분자 설계에서 AI가 중요한 역할을 할 것이라고 믿습니다.

이 자동화된 프로세스는 새로운 분자의 합성과 테스트에 혁명을 일으켜 잠재적으로 새로운 물질과 신약 후보 물질의 발견을 가속화할 수 있습니다.

이주영 교수의 강연에 대한 석자 교수의 견해

석자 교수는 이주영 교수의 강의에 감탄하며, 자신은 과학 부분을 짧게 하고 이주영 교수의 강의는 길게 했으면 더 좋았을 것 같다고 소감을 밝혔습니다.

박사 과정 중에 이주영 교수님이 든든한 멘토가 되어주셨다며 칭찬을 아끼지 않았습니다.

인공지능이 화학자의 역할에 미칠 영향에 대해 일자리 대체에 대한 우려도 있었지만, 계산 화학자의 지속적인 필요성에 대한 낙관론도 있었습니다.

화학과 자연 과학에서 인공지능의 역할

인공지능은 화학 분야에서 단순 반복적인 작업을 점점 더 많이 처리하여 더 복잡한 문제와 새로운 분자의 탐구에 집중할 수 있게 해줍니다.

화학 분야에서 AI를 적용하면 환경 문제를 해결하기 위한 신소재뿐만 아니라 신약 개발도 가속화될 것으로 기대됩니다.

또한 AI는 자연과학과 기술 전반에 걸쳐 광범위한 전망을 가지고 있어 다양한 분야의 발전을 가속화할 수 있는 잠재력을 가지고 있습니다.

화학은 새로운 물질을 만들기 위해 수학적 이론과 인공지능을 활용하여 분자의 구조, 성질, 변형을 연구하는 학문입니다.

또한 원자의 변화와 다양성을 연구하고 원자들이 어떻게 섞여 다양한 현상을 만들어내는지 탐구하며 물리와 생물학 사이의 독특한 공간을 차지합니다.

화학은 우리 일상과 밀접한 관련이 있는 분자를 이해하는 데 중요한 역할을 합니다.