과대광고가 정당한가? 왜 모든 시선이 AI에 집중되는가

AI 분야 노벨상

The list of 2024 Nobel Prize winners has brought a lot of attention to AI. First, it was with the Nobel Prize in Physics, which rewarded the fundamental physics-inspired theory and early prototype for the building of neural networks, the basis of most of today’s AI technology. We discussed in detail how these early neural networks worked in “노벨상 성과에 투자하기—인공 신경망, AI의 기반”.

이번에는 화학 분야 노벨상이 계산 중심 연구에 수여되었습니다. 구체적으로는 David Baker에게 “계산 단백질 설계” 공로로 절반이 주어졌고, 나머지 절반은 Google의 AlphaFold 핵심 연구원인 Demis Hassabis와 John M. Jumper에게 “단백질 구조 예측” 공로로 공동 수여되었습니다.

따라서 2024년이 AI 노벨상 연도였다고 과장하는 것이 아니라, “자연과학”(물리학, 화학, 의학/생물학) 분야의 세 개 노벨상 중 두 개가 AI 관련 프로젝트에 수여되었습니다.

This was obviously praised by researchers in the field and people enthusiastic about the future of AI-driven research. But it also 컴퓨팅을and AI를 “진정한 물리학”이 아니라고 보는.

노벨상이 무엇을 보상해야 할까?

This is part of a larger and older debate of what constitutes “pure” science worthy of a Nobel Prize. Some will say that only theoretical advancement and true intellectual breakthroughs are worthy of the most prestigious international prize in science.

I’m speechless. I like ML and ANN as much as the next person, but hard to see that this is a Physics discovery. Guess the Nobel got hit by AI hype.

Jonathan Pritchard – 런던 임페리얼 칼리지 천체물리학자

Others will say that sciences should also be rewarded for their impact on the real world, especially when it comes to the Nobel Prize, whose founder, Alfred Nobel made to give a prize “전년도에 인류에게 가장 큰 혜택을 제공한 사람들에게” in physics, chemistry, physiology or medicine, literature, and peace.

In that context, it might be argued that neural networks have indeed contributed greatly to the benefit of humankind, will likely continue to do so even more in the future, and are therefore worthy of this year’s Nobel Prize.

컴퓨팅이 물리학인가?

Regarding the 2024 Nobel Prize In Physics specifically, the method developed to create neural networks was deeply rooted in physics. More specifically, it drew heavily from statistical physics, a field describing things with many elements like gases or liquids.

It also took inspiration from the observation that collective properties in many physical systems are robust to changes in model details. Notably magnetic materials derive their special characteristics thanks to their atomic spin – a property that makes each atom a tiny magnet.

Overall, the models rewarded by the Nobel Prize, The Boltzmann machine invented by Hinton and the Hopfield network, “are both energy-based models” that do not differ much from the mathematical description used to understand the physics of real-life material.

Still, some “real” physicists felt some dismay at the idea that computing is taking the spotlight from physics, maybe speaking more about how little society gives respect and attention to this field than about this year’s Nobel Prize in itself.

“It falls into the field of computer science. The annual Nobel prize is a rare opportunity for physics — and physicists with it — to step into the spotlight.

It’s the day when friends and family remember they know a physicist and maybe go and ask him or her what this recent Nobel is all about. But not this year.”

Sabine Hossenfelder- Physicist at the Munich Center for Mathematical Philosophy in Germany

AI로 돌아오는 물리학

The Nobel Prize reward also takes into account a growing development in computing to go back to physics-based methods to improve machine learning, neural networks, and AI.

“We need the way of thinking we have in physics to study machine learning.”

Lenka Zdeborová – Statistical physicist at the Swiss Federal Institute of Technology in Lausanne.

The early work on neural networks was truly interdisciplinary, taking also inspiration from the latest advancement in our understanding of how biological neurons work, and bringing together math, physics, computer sciences, and neurobiology.

“I think that the Nobel Prize in Physics should continue to spread into more regions of physics knowledge. Physics is becoming wider and wider, and it contains many areas of knowledge that did not exist in the past, or were not part of physics.”

Giorgio Parisi- Physicist at the Sapienza University of Rome, who 2021년 물리학 노벨을 공동 수상.

So while maybe a little surprising, and to the dismay of the most purist type of physicists, the attribution of the Nobel Prize to John Hopfield and Geoffrey Hinton is not as far from physics and “hard sciences” as it might look at at first glance.

상처에 독을 더하다?

Maybe the reaction of some members of the scientific community would have been more moderate if the Chemistry Nobel Prize had not been computation-focused as well.

Here, the debate is more focused on the real contribution of AI to the field, giving some credit to the “Guess the Nobel got hit by AI hype” criticism.

This is because neural networks like AlphaFold, predicting the 3D configuration of protein, were built on top of a massive treasure trove of data built over decades of practical experiments. This includes especially the  Protein Data Bank, a freely available repository of more than 200,000 protein structures.

These structures were determined experimentally by thousands of human researchers over decades, using advanced methods (often themselves Nobel Prize-winning discoveries) like  X-ray crystallography, cryo-electron microscopy, etc.

“I don’t think AlphaFold involves any radical change in the underlying science that wasn’t already in place.

It’s just how it was put together and conceived in such a seamless way that allowed AlphaFold to reach those heights.”

David Jones – Bioinformatician at University College London, who collaborated with DeepMind on the first version of AlphaFold

AlphaFold의 영향력은 어느 정도인가?

Criticizing AlphaFold as “only building onto previous research” can maybe be seen as a little off the mark. After all, it is a very common pattern that a Nobel Prize-winning discovery is often built on 3-4 other previous Nobel Prize-winning discoveries.

More importantly, it is key to determine if AlphaFold is a true breakthrough, making the previously impossible suddenly achievable.

우리는 이전에 AlphaFold가 기존 데이터를 단순히 정제하는 것이 아니라 새로운 분자와 약물을 발견할 수 있는지 조사한 과학 논문에 대해 논의했습니다..

And it seems to be the case:

Researchers determined that the proportion of compounds that altered protein activity for each of the models was around 50% and 20% for the sigma-2 receptor and 5-HT2A receptors, respectively. A result greater than 5% is exceptional.

Considering that drug discovery is very much like trying to manually find a needle in a haystack, a 50% success rate of “guessing” where the needle is from the first try is indeed exceptional.

So here too, this year’s Nobel Prize in Chemistry might not fit the bill when it comes to “pure” science and progress in fundamental understanding. But it seems to match well the state goal of the Nobel Prize: “to give a prize to those who, during the preceding year, have conferred the greatest benefit to humankind”.

논란 마무리

We can agree that as AI becomes a key tool in most scientific disciplines, we should not see every Nobel Prize rewarding mostly AI moving forward. In the same way, we do not regularly reward “basic” computing (or, for that matter, chemistry or metallurgy) because researchers use computers (and chemicals and metals) on a daily basis.

However, it seems that the reaction against this year’s Nobel Prize attribution was a little exaggerated, maybe illustrating the frustration of scientists to see so much media attention sucked up by AI only despite impressive scientific progress in most fields.

AI가 실제로 할 수 있는 일은 무엇인가?

In any case, AI is likely to be a truly transformative technology, and 인류 역사에서 증기 기관, 전신, 내연 기관, 최초의 컴퓨터와 같은 이전 기술들에 버금가는 진정한 이정표가 될 것입니다..

Here is a short list of the potential applications of AI:

• 신약 발견 및 생물학 데이터 분석.
• 진단 및 자동화된 의료 치료, 수술 포함.
• 로봇공학, 가정용 로봇부터 자동화 생산까지.
• 자율주행 자동차 및 물류.
• 맞춤형 학습 및 지식 확산.
• 에너지, 재료 과학, 나노기술 등에서 새로운 물질이 발견되고 있습니다.
• 광자학, 양자 컴퓨팅 등을 포함한 새로운 컴퓨팅 방법.
• 우주 탐사, 자동화된 외계 식민지부터 소행성 채굴까지.

그 맥락에서 AI에 대한 “과대광고”는 전적으로 정당합니다.

하지만 대부분의 새로운 중요한 기술과 마찬가지로 초기 혁신을 실용적인 사용 사례로 전환하는 과정은 기대보다 다소 오래 걸릴 수 있습니다. 예를 들어 1970년대의 컴퓨터는 인터넷 혁명 수십 년 전, 그리고 AI보다 훨씬 이전에 등장했으며, 그 영향력은 누구도 예상하지 못한 규모일 가능성이 높습니다.