{"id":1351,"date":"2025-05-27T06:00:00","date_gmt":"2025-05-27T06:00:00","guid":{"rendered":"http:\/\/localhost:8080\/decoding-ai-how-ai-is-powering-the-next-scientific-revolution\/"},"modified":"2026-07-29T14:53:50","modified_gmt":"2026-07-29T05:53:50","slug":"decoding-ai-how-ai-is-powering-the-next-scientific-revolution","status":"publish","type":"post","link":"https:\/\/news.skhynix.com\/en\/decoding-ai-how-ai-is-powering-the-next-scientific-revolution\/","title":{"rendered":"Decoding AI: How AI Is Powering the Next Scientific Revolution"},"content":{"rendered":"
The following article is an edited and abridged version of the original Korean contribution from Dr. Sungju Kang, also known as the YouTuber \u201cHang-sung.\u201d A regular contributor to Korea\u2019s top science YouTube channel Unreal Science<\/em><\/a>, Dr. Kang reveals some of the most exciting scientific breakthroughs made possible by AI.\u00a0<\/em><\/div>\n
\"Examples<\/p>\n

\u25b2 Examples of various AI applications in daily life<\/p>\n<\/figure>\n

The pace of progress in AI these days is nothing short of astonishing. AI can now write stories, create illustrations, and even suggest pasta recipes\u2014but that\u2019s just the beginning. It\u2019s now reshaping the very paradigm of scientific research.<\/p>\n

This is largely due to the sheer volume and complexity of the data that modern science must process. Scientists deal with enormous datasets, hundreds of variables, and countless experimental conditions. While it\u2019s not impossible for humans to process them manually, doing so takes a significant amount of time and is often inefficient, making it difficult to produce timely results.<\/p>\n

That\u2019s where AI comes in. In today\u2019s labs, AI has become a crucial tool. It can quickly analyze vast amounts of data that would be difficult for humans to handle, suggest promising combinations, and build complex predictive models, allowing science to progress faster than ever.<\/p>\n

We\u2019re now living in a time where scientists ask the questions and AI helps point them toward the answers. AI is no longer just a calculator. It has become a true companion for scientists, helping design experiments, analyze data, and push research forward.<\/p>\n

\u00a0<\/p>\n

\n
\n
<\/div>\n

AlphaFold: AI That Decodes the Structure of Life<\/h3>\n
<\/div>\n<\/div>\n<\/div>\n
\"AlphaFold,<\/p>\n

\u25b2 AlphaFold, Google DeepMind\u2019s protein structure prediction AI<\/p>\n<\/figure>\n

Proteins are the fundamental building blocks of life. From maintaining the shape of cells to hormone secretion, immune responses, and metabolic control, nearly all biological processes depend on how proteins function. But here\u2019s the catch: a protein\u2019s function is determined by its three-dimensional structure. In other words, to understand what a protein does, you first have to understand how it\u2019s shaped. However, figuring out that shape through experiments is a highly complex task that can take months or even years. This is the challenge AlphaFold<\/a> set out to overcome.<\/p>\n

AlphaFold is an AI model co-developed by Google DeepMind and Professor David Baker\u2019s team at the University of Washington. You simply input the amino acid sequence of a protein and AlphaFold predicts its 3D structure. The first version, AlphaFold1, was considered a significant breakthrough as it showed that AI could be used to predict protein structures. However, as the model only achieved an accuracy rate of around 60%, it wasn\u2019t suitable for real-world research applications.<\/p>\n

\"Median<\/p>\n

\u25b2 Median accuracy in the free modeling category (\u00a9Google DeepMind)<\/p>\n<\/figure>\n

The next version, AlphaFold2, showed a dramatic leap in performance, achieving over 90% accuracy in its predictions to surpass even human researchers. This remarkable improvement was made possible by deep learning algorithms trained on massive datasets of known protein structures, supported by powerful computational infrastructure.<\/p>\n

However, what truly sets AlphaFold apart is its commitment to open science. DeepMind released over 200 million protein structure predictions to the public for free. Today, more than two million researchers worldwide are using this data. Its impact has been far-reaching, including speeding up the design of COVID-19 vaccines, unraveling the mechanisms behind complex diseases such as Alzheimer\u2019s, and accelerating the development of enzymes that can break down plastic. In short, AlphaFold has become one of the most widely used AI models in the scientific community.<\/p>\n

\u00a0<\/p>\n

\"2024<\/p>\n

\u25b2 2024 Nobel Prize in Chemistry laureates (\u00a9The Royal Swedish Academy of Sciences)<\/p>\n<\/figure>\n

In 2024, the developers of AlphaFold\u2014Demis Hassabis, John Jumper, and David Baker\u2014were jointly awarded the Nobel Prize in Chemistry. It marked the first time that this prestigious honor was presented to developers of an AI-powered technology. In May of the same year, AlphaFold3 was unveiled. This version featured capabilities which expanded beyond predicting protein structures to include interactions with other biomolecules such as DNA, RNA, antibodies, and ligands* <\/sup>. With this upgrade, AlphaFold became an even more powerful tool in the field of biosciences. However, unlike its predecessor, AlphaFold3 was released without open-source access to its code. This shift sparked debate within the scientific community over the role and value of open science.<\/p>\n

* <\/sup>Ligand: A molecule that forms a complex with a biomolecule to serve a biological purpose. In protein-ligand binding, the ligand attaches to a specific site on the protein and triggers a signal,\u00a0often causing the protein\u2019s structure to change in response.<\/p>\n

\u00a0<\/p>\n

\n
\n
<\/div>\n

GNoME: AI That Discovers Materials the World Has Never Seen<\/h3>\n
<\/div>\n<\/div>\n<\/div>\n
\"GNoME,<\/p>\n

\u25b2 GNoME, Google DeepMind\u2019s AI for predicting new materials<\/p>\n<\/figure>\n

AI is now venturing into uncharted territory in science, including the discovery of new materials. One of the most exciting areas it\u2019s transforming is materials science. Traditionally, developing new materials meant experimenting with thousands of elements in countless combinations of crystal structures and properties. It often took years of trial and error just to find a single promising compound. In many ways, it was like scientific craftsmanship but the situation is now changing fast. Today, AI is becoming central to materials research. One standout example is GNoME<\/a> (Graph Networks for Materials Exploration), an AI model released by Google DeepMind in late 2023.<\/p>\n

GNoME is designed to rapidly and accurately predict the potential of new, previously unknown materials. It learns from hundreds of thousands of data items on known crystal structures to identify patterns in atomic bonding and energetic stability. Then, it simulates possible combinations and automatically generates candidates that are likely to form stable solid materials.<\/p>\n

What makes GNoME especially powerful is that it goes beyond simple pattern matching. It incorporates physics-based modeling, including quantum mechanical calculation and the Voronoi algorithm* <\/sup>30, to produce more precise and realistic predictions.<\/p>\n

*<\/span> Voronoi algorithm:\u00a0A method used to\u00a0divide\u00a0a\u00a0space into regions\u00a0based on\u00a0the distance to\u00a0a\u00a0set of\u00a0given points, known as a seeds.<\/p>\n

\u00a0<\/p>\n

\n
\n
\n\t\"\"<\/a>
<\/div><\/div>\n<\/div>\n
\n\t\"\"<\/a>
<\/div><\/div>\n<\/div>\n<\/div>\n
<\/div>
<\/div>
<\/div><\/div><\/div>\n\n

\u25b2 Voronoi algorithm: A mathematical method for dividing the spatial distribution and boundaries between atoms (\u00a9Ovito)<\/p>\n

Quantum mechanical calculations help determine whether a particular combination of elements is physically viable by evaluating how atoms interact and how electrons behave within a given structure. Meanwhile, the Voronoi algorithm evaluates whether a stable crystal structure can be formed without atomic collisions based on the space each atom occupies and its distance distribution relative to neighboring atoms.<\/p>\n

By considering both microscopic quantum interactions and spatial geometry, GNoME\u2019s predictions go far beyond simple data-driven guesses, yielding results that are more physically grounded and more likely to succeed in real-world experiments.<\/p>\n

\"Number<\/p>\n

\u25b2 Number of stable new materials discovered through various methods (\u00a9Google DeepMind)<\/p>\n<\/figure>\n

\u00a0
\nThanks to its highly detailed, theory-based simulations, GNoME has generated over 2.5 million candidate materials. Of these, around 380,000 were evaluated highly likely to be experimentally stable. To put that in perspective, it would have taken thousands of scientists several decades to achieve the same result using traditional methods. This is a powerful example of just how dramatically AI is accelerating the pace of science.<\/p>\n

Today, GNoME is actively being used to search for next-generation materials in key industries such as batteries, semiconductors, superconductors, and energy storage. In areas where experimentation is especially difficult such as next-generation energy devices or quantum computing materials, researchers are increasingly turning to GNoME\u2019s predicted candidates to guide their work. In this way, GNoME is evolving beyond a simple computational tool, becoming an \u201cintelligent scientific partner\u201d that opens doors to undiscovered materials.<\/p>\n

\u00a0<\/p>\n

\n
\n
<\/div>\n

Chemprop: AI That Understands the Language of Molecules<\/h3>\n
<\/div>\n<\/div>\n<\/div>\n
\"GNoME,<\/p>\n

\u25b2 GNoME, Google DeepMiChemprop, an AI model for predicting molecular properties<\/p>\n<\/figure>\n

For a long time, chemistry has been seen as a branch of science guided by intuition and experience. Predicting the properties and reaction pathways of countless compounds requires both complex theoretical knowledge and years of experimental experience. Today, AI is reshaping that approach from the ground up. At the center of this transformation is a technology called Chemprop.<\/a><\/p>\n

\u00a0<\/p>\n

\"\"<\/p>\n

\u00a0<\/p>\n

\u25b2 Unlike traditional chemical models that rely on molecular formulas or structural diagrams, Chemprop represents molecules as graphs using Global Neural Networks (\u00a9Wojtuch, Agnieszka, et al. \u201cExtended study on atomic featurization in graph neural networks for molecular property prediction.\u201d Journal of Cheminformatics 15.1 (2023), p 81)<\/p>\n<\/figure>\n

\u00a0
\nChemprop was developed in 2019 by researchers at MIT as a molecular property prediction AI model. What sets Chemprop apart is how it sees molecules completely differently from traditional chemical models. Most conventional models represent molecules using atomic sequences or chemical formulas. Chemprop, however, takes a more nuanced approach. It views a molecule not as a flat formula, but as a graph, where atoms are nodes and chemical bonds are edges. It then uses a Graph Neural Network (GNN)* <\/sup> to learn how electrons move and interact within this molecular network. This method allows Chemprop to predict a molecule\u2019s physical and chemical properties with far greater precision than previous models.<\/p>\n

*<\/span> Graph Neural Network (GNN):\u00a0A type of deep learning model designed to analyze graph-structured data. In the case of molecules, nodes represent atoms and edges represent the bonds or relationships between them.<\/p>\n

\u00a0<\/p>\n

\"Nodes<\/p>\n

\u25b2 Nodes and edges illustrated using a water molecule<\/p>\n<\/figure>\n

\u00a0
\nChemprop can predict a wide range of molecular properties, including toxicity, solubility, stability, and biological activity. This helps researchers assess the performance and potential side effects of drug candidates in advance. In fact, major pharmaceutical companies such as Merck and Novartis have used Chemprop to shorten their drug development timelines by one to two years. As a result, Chemprop is quickly emerging as a powerful tool in pharmaceutical research.<\/p>\n

Beyond drug discovery, Chemprop is also contributing to sustainable chemistry. It\u2019s being used to predict the potential of chemical materials for biodegradable plastics and eco-friendly catalysts, thereby helping to reduce waste and environmental impact before any lab experiments take place. This has earned Chemprop recognition as a promising example of \u201cgreen AI.\u201d<\/p>\n

Most importantly, Chemprop is open source. Its accessibility and scalability allow university labs and startups around the world to build their own predictive models or customize tools for specific groups of compounds with just a few lines of code. In this way, Chemprop is fundamentally changing how we decode the chemical world.<\/p>\n

\u00a0<\/p>\n

\n
\n
<\/div>\n

HelioLinc3D: AI That Watches the Skies<\/h3>\n
<\/div>\n<\/div>\n<\/div>\n
\"HelioLinc3D,<\/p>\n

\u25b2 HelioLinc3D, an AI for asteroid detection<\/p>\n<\/figure>\n

\u00a0
\nBeyond life, material, and molecules, AI is now beginning to read the order of the cosmos. Developed by the University of Washington, HelioLinc3D<\/a> is an AI algorithm designed for asteroid detection which is setting a new standard in the field of astronomy.<\/p>\n

<\/figure>\n
\"Conventional<\/p>\n

\u25b2 Conventional asteroid detection involves tracking an object multiple times in a single night to estimate its orbit (\u00a9 JPL | NASA Center for Near-Earth Object Studies)<\/p>\n<\/figure>\n

\u00a0
\nConventional asteroid detection relies on a method called tracklet-based linking* <\/sup>, which requires the same object to be spotted at least four times over a single night. As a result, only bright and fast-moving asteroids could be reliably tracked, while slower or dimmer objects were often left out of the data altogether.
\n* <\/sup>Tracklet-based linking:\u00a0This technique connects multiple short observation segments,\u00a0called tracklets,\u00a0into a continuous path, enabling ongoing tracking even when objects are briefly lost or obscured.<\/p>\n

HelioLinc3D takes a completely different approach. It uses AI to connect fragmented observation data captured on different days from various locations to reconstruct a single, unified orbit.<\/p>\n

By analyzing millions of data points simultaneously, including an object\u2019s brightness, speed, direction, and position, HelioLinc3D identifies potential candidates and links them together into one cohesive celestial track. The algorithm\u2019s true innovation lies in its ability to merge fragmented data and match it to a 3D orbital model. Thanks to this, HelioLinc3D can detect objects that would likely have been missed by traditional systems.<\/p>\n

\u00a0<\/p>\n

\"Reconstructing<\/p>\n

\u25b2 Reconstructing a complete orbit from fragmented observations taken across multiple days (\u00a9ATLAS\/University of Hawaii Institute for Astronomy\/NASA)<\/p>\n<\/figure>\n

In 2023, HelioLinc3D made headlines when it successfully detected the asteroid \u201c2022 SF289\u201d that had gone unnoticed by conventional detection methods. The asteroid moved slowly, blended into the background starlight, and had observational data scattered across different dates and telescope locations. In other words, it was exactly the kind of object traditional systems would likely miss altogether.<\/p>\n

HelioLinc3D was able to connect those fragmented pieces of data, like assembling a cosmic puzzle, and reconstruct the asteroid\u2019s orbit. This breakthrough approach means that HelioLinc3D is set to play a key role in the upcoming LSST* <\/sup> project, one of the largest astronomical surveys ever conducted in the U.S. Specifically, HelioLinc3D will process and interpret the tens of terabytes (TB) of astronomical data collected by LSST each day.
\n* <\/sup>Legacy Survey of Space and Time\u00a0(LSST):\u00a0A massive 10-year project led by the Vera Rubin Observatory in the U.S., aiming to observe approximately 37 billion celestial objects and track their changes over time.<\/p>\n

\u00a0<\/p>\n

\n
\n
<\/div>\n

AI Is Becoming a True Partner in Science<\/h3>\n
<\/div>\n<\/div>\n<\/div>\n
\"AI<\/p>\n

\u25b2 AI is evolving from assistant to experiment designer, data analyst, and simulation predictor<\/p>\n<\/figure>\n

\u00a0
\nAt its core, science is about asking questions, observing the world, and making sense of what we find. Until recently, this entire process was led by humans. But now, AI is stepping in\u2014not just as a helper, but as a designer of experiments, an analyst of complex data, an explorer of the unknown, and even a watcher of the cosmos.<\/p>\n

AlphaFold is solving the puzzle of life. GNoME is redrawing the map of new materials. Chemprop is interpreting the language of molecules. Meanwhile, HelioLinc3D is tracing celestial bodies once hidden from view. Though these AI models are designed for very different fields, they are all moving toward one shared goal: advancing science. This progress is driven by companies such as SK hynix, which are continually innovating technologies to fuel advances in AI.<\/p>\n

\u00a0<\/p>\n

\"SK<\/p>\n

\u25b2 SK hynix is contributing to the development of AI technologies that are redefining the scientific paradigm<\/p>\n<\/figure>\n

\u00a0
\nOf course, AI isn\u2019t a magic wand. Human intuition, ethical judgment, and creativity remain essential to scientific discovery. One thing is clear though\u2014AI is changing the pace of scientific advancement and now science must move forward in partnership with AI. As these technologies progress, it\u2019s just as important to address questions around data bias, misuse, and the need for ethical safeguards.<\/p>\n

So, what kind of AI system will write the next chapter in science? It\u2019s a future we should all look forward to.<\/p>\n

Disclaimer:<\/em><\/strong>\u202fThe opinions expressed in this article are solely those of the author and do not necessarily reflect the official position of SK hynix.<\/em><\/p>\n

\u00a0<\/p>\n

\"\"<\/figure>\n","protected":false},"excerpt":{"rendered":"

The following article is an edited and abridged version of the original Korean contribution from Dr. Sungju Kang, also known as the YouTuber \u201cHang-sung.\u201d A regular contributor to Korea\u2019s top science YouTube channel Unreal Science, Dr. Kang reveals some of the most exciting scientific breakthroughs made possible by AI.\u00a0 The pace of progress in AI [\u2026]<\/p>\n","protected":false},"author":23,"featured_media":1335,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_migrated_source_id":18708,"footnotes":"","_members_access_role":[],"_members_access_error":""},"categories":[5],"tags":[12,466,455,467],"class_list":["post-1351","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tech-and-ai","tag-ai","tag-ai-and-science","tag-google-deepmind","tag-nobel-prize"],"acf":[],"_links":{"self":[{"href":"https:\/\/news.skhynix.com\/en\/wp-json\/wp\/v2\/posts\/1351","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/news.skhynix.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/news.skhynix.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/news.skhynix.com\/en\/wp-json\/wp\/v2\/users\/23"}],"replies":[{"embeddable":true,"href":"https:\/\/news.skhynix.com\/en\/wp-json\/wp\/v2\/comments?post=1351"}],"version-history":[{"count":5,"href":"https:\/\/news.skhynix.com\/en\/wp-json\/wp\/v2\/posts\/1351\/revisions"}],"predecessor-version":[{"id":11628,"href":"https:\/\/news.skhynix.com\/en\/wp-json\/wp\/v2\/posts\/1351\/revisions\/11628"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/news.skhynix.com\/en\/wp-json\/wp\/v2\/media\/1335"}],"wp:attachment":[{"href":"https:\/\/news.skhynix.com\/en\/wp-json\/wp\/v2\/media?parent=1351"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/news.skhynix.com\/en\/wp-json\/wp\/v2\/categories?post=1351"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/news.skhynix.com\/en\/wp-json\/wp\/v2\/tags?post=1351"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}