Table of Contents
Artificial Intelligence Translates the Language of Animals
Kim Kyung-jin, Attorney at Law
The Story of AI Learning to Listen to Whales, Dolphins, Birds, and Bees
Twelve chapters on how AI listens to dolphins, sperm whales, humpback whales, birds, and bees to find rules in their sounds, and what this technology means for its risks and for animal rights. Written in simple sentences a child can read, with verified sources in every section.
Table of Contents
AI Deciphers Ancient Scripts
Kim Kyung-jin, Attorney at Law
Ancient Records Revived by AI
In twelve chapters, this book explains how AI revives records once unreadable, from burned scrolls and wooden slips buried in mud to broken clay tablets. It covers virtual unrolling at Herculaneum, virtual collation of oracle-bone texts, reading Silla wooden tablets, computational analysis of undeciphered scripts, and multispectral archives, with verified references for each chapter.
Table of Contents
Artificial Intelligence for New Materials Design and Rocket Propulsion Engineering
Kim Kyung-jin, Attorney at Law
AI Potentials, Self-Driving Laboratories, and Physics-Informed Machine Learning (PIML)
Ten chapters on how artificial intelligence is changing new materials and rocket propulsion: atomic simulation, generative models, self-driving labs, high-temperature alloys, metal 3D printing, combustion, cooling design, and engine diagnosis and control. Written without equations, with verified sources in every chapter.
AI Library
The Age of Autonomous Scientific Discovery
Kim Kyung-jin, Attorney at Law
AI Scientists and Self-Driving Labs
This book follows how AI scientists and self-driving labs are changing the way science generates and verifies claims. It covers literature-based discovery, natural-language protocols translated into robot commands, multi-agent research systems, closed-loop laboratories, materials search, the verification gap, chains of evidence, research harnesses, journal ethics, and legal responsibility.
AI Library
A New Era of Life Sciences Opened by Artificial Intelligence
Structural Proteomics, Genomic Foundation Models, Autonomous Laboratories, and Global Governance
Kim Kyung-jin, Attorney at Law
This book is a research volume compiled with artificial intelligence. A human selected the materials and structured the work, while AI models drafted the sentences and cross-checked the facts.
AI Library
The Double Structure of Digital Sovereignty
Europe’s Departure from Palantir and the Chains of American Big Tech
Kim Kyung-jin, Attorney at Law
This is a record of 2026, when European intelligence agencies and defense ministries began removing analytics tools from America’s Palantir. It covers the replacement decisions made by France’s General Directorate for Internal Security (DGSI), Germany’s Federal Office for the Protection of the Constitution (BfV), and the Netherlands Ministry of Defense; the incident in which US export controls severed an ally’s ac…
New English Edition
Artificial Intelligence in Horticulture
Kim Kyung-jin, Attorney at Law
Across five chapters and ten sections, this book examines computer vision for crop diagnosis, harvesting robots and autonomous field systems, smart greenhouses and digital twins, precision irrigation and supply-chain quality control, high-throughput phenotyping, and predictive breeding.
New English Edition
Artificial Intelligence in Food Crop Agriculture
Kim Kyung-jin, Attorney at Law
Across six chapters and eighteen sections, the book examines digital agricultural infrastructure, remote sensing, crop diagnosis, yield forecasting, precision irrigation, genomics, molecular breeding, agricultural robotics, climate-smart agriculture, and global food security.
New English Edition
The Future of Forestry and Agroforestry
Kim Kyung-jin, Attorney at Law
Driven by Artificial Intelligence and Digital Innovation
Across five chapters and fifteen sections, the book follows satellites, drones, LiDAR, digital twins, forest-specific language models, wildfire and pest forecasting, forestry robotics, agroforestry, timber traceability, and forest carbon markets.
New English Edition
Smart Livestock Farming: AI Enters the Barn
Kim Kyung-jin, Attorney at Law
Sensors listen, cameras watch, and artificial intelligence helps farmers decide.
Across five chapters and fifteen sections, the book follows precision livestock farming from animal health and reproduction to robotic milking, virtual fencing, digital twins, methane reduction, welfare, and data ownership.
Table of Contents
Han Dong-hoon, Busan Buk-gu Gap: A Record of the 100 Days Before and After the Election (Mar. 26-Jul. 3, 2026)
Kim Kyung-jin
Table of Contents and 13 sections
From March 26 to July 3, 2026, this record follows the spring after expulsion, the Busan Buk-gu Gap by-election, victory as an independent, and the first bill submitted in the National Assembly.
Table of Contents
Artificial Intelligence and Medicine
Kim Kyung-jin, Attorney at Law
AI in clinical care, hospitals, education, and research
AI in medical imaging, risk prediction, treatment planning, hospital operations, education, and research, with patient safety, privacy, and accountability.
[AI Library] Chapter 15. Criminal Justice and Educational AI
Artificial Intelligence on Trial
Part 4. Physical Safety and Sector-Specific AI Litigation
Chapter 15. Criminal Justice and Educational AI
Attorney Kyungjin Kim
A. Predictive Policing
(1) NAACP Chicago Lawsuit
Michael Williams was 65 years old. He had lived his entire life on Chicago's South Side, and everyone in the neighborhood knew him. One evening in 2020, he offered a young man he knew a ride home. A few blocks later, bullets came through the open window. The young man was hit.
Police arrived. They arrested Williams as a murder suspect. Their basis was a single piece of evidence: ShotSpotter.
ShotSpotter is an acoustic surveillance system that claims to detect gunfire. Microphones attached to streetlights and traffic signals capture sounds, an algorithm determines whether the sound is gunfire, and the location is transmitted to police. Chicago deployed the system across the South Side and West Side starting in 2017. The contract cost $9 million per year.
There was a problem. In the Williams case, ShotSpotter initially classified the sound as 'fireworks.' The location was marked more than a mile away from the scene. Yet police used this alert as the basis for charging Williams. There was no evidence that a bullet had been fired from inside the car. There were no witnesses. Only the algorithm's judgment existed.
Williams spent 11 months in Cook County Jail. A diabetic, he did not receive proper meals. He contracted COVID-19 twice. His hands developed a tremor that did not go away after his release. Prosecutors eventually dropped the charges, saying they could not vouch for the reliability of ShotSpotter evidence. In July 2022, the MacArthur Justice Center filed a class action lawsuit against the City of Chicago.
Williams, along with two other victims, Daniel Ortiz and Derrick Scruggs, were the plaintiffs. Ortiz had been handcuffed, searched, and arrested by officers responding to a ShotSpotter alert outside a laundromat. The drug possession charge was dismissed the next day.
The lawsuit made two central claims.
First, that a ShotSpotter alert alone does not establish the reasonable suspicion needed to justify a stop and search, making such actions a violation of the Fourth Amendment.
Second, that Chicago's decision to deploy ShotSpotter sensors exclusively in Black and Latino neighborhoods violated the Illinois Civil Rights Act.
The numbers supported these claims.
According to a 2021 report by the Chicago Office of Inspector General, 89 to 91 percent of ShotSpotter alerts ended without any evidence of gun-related crime being found. Roughly 100 false alerts occurred every day. Eighty percent of Chicago's Black residents and 65 percent of its Latino residents lived under ShotSpotter surveillance. Only 30 percent of white residents did.
In September 2024, Mayor Brandon Johnson terminated the ShotSpotter contract. He called the system 'a walkie-talkie on a pole.' The city council twice demanded he reverse the decision, but the mayor held firm.
In August 2025, the lawsuit was settled for $90,000. Something mattered more than the dollar amount. The City of Chicago agreed that 'a ShotSpotter alert alone does not constitute justification for stopping or searching a person near the alert location.' The principle was confirmed: an algorithm's judgment cannot substitute for a person's constitutional rights. There was an ironic coda. In the year after ShotSpotter was turned off, homicides in the affected areas dropped by approximately 32 percent. This was according to analysis by the University of Chicago Justice Project. When the technology that was supposed to reduce fear disappeared, fear itself diminished with it.
(2) ShotSpotter and the Surveillance Technology Controversy
ShotSpotter's problems were not unique to Chicago. As of 2024, the system was deployed in more than 150 cities across the United States. New York, Washington D.C., Denver, Miami. Each city faced problems similar to Chicago's.
The company claimed 97 percent accuracy. That number was deceptive. The method of calculating accuracy was the problem. When an alert went off and police responded, failing to find evidence of gunfire did not count as 'inaccurate.' An alert was classified as inaccurate only when an officer voluntarily filed an 'error report.' Officers almost never filed those reports.
A genuine test had never been conducted. No one had ever scientifically verified whether ShotSpotter could distinguish gunfire from firecrackers, a backfiring car, construction noise, or helicopter sounds.
In March 2024, former ShotSpotter employees Chris Edwards and Ginsie Robinson became whistleblowers. Edwards had worked at the company for over two years, managing a sensor upgrade project. In court filings, he claimed that significant portions of the system were 'broken, corroded, and unmaintained.' He questioned whether accurate data was being delivered to customers. He raised his concerns with supervisors. He was fired.
ShotSpotter (now renamed SoundThinking) countersued Edwards, alleging he had taken confidential documents. Edwards argued this was a Strategic Lawsuit Against Public Participation (SLAPP) designed to silence him. In January 2024, a judge denied Edwards's motion to dismiss and allowed the lawsuit to proceed.
The controversy continued outside the courtroom. In 2024, the Massachusetts Supreme Judicial Court questioned the reliability of ShotSpotter evidence in Commonwealth v. Rios. What standards must algorithm-generated data meet to be admitted as evidence in court? Must an expert witness be able to explain how the algorithm works? Must the defendant be given an opportunity to examine the algorithm? These questions remain unanswered. But one thing became clear. Predictive policing does not predict crime. It projects past biases into the future. More police are sent to Black neighborhoods, more arrests occur, and that data feeds back into the algorithm, sending even more police. Scholars call this a 'feedback loop.'
As of January 2025, Chicago is considering deploying a new gunshot detection system. A budget of $9 million has been allocated. Eight companies, including SoundThinking, have submitted bids. Before the ink on the settlement agreement had dried, the same technology was trying to return under a different name.
B. Risk Assessment Algorithms
(1) The COMPAS System and Racial Bias
Vernon Prater was white, 41 years old, and had a prior conviction for armed robbery. His criminal record was long. In 2014, he was arrested in Broward County, Florida, for yet another offense.
The COMPAS algorithm classified him as 'low risk,' meaning it predicted a low likelihood of reoffending within three years of release.
Brisha Borden was Black and 18 years old. Her record consisted only of minor juvenile offenses. Attempted bicycle theft. The same algorithm classified her as 'high risk.'
Two years later, Prater was arrested again on a theft charge carrying a potential eight-year sentence. Borden did not reoffend.
COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) is an algorithm developed in 1998 by a company called Northpointe.
It assigns defendants a recidivism risk score between 1 and 10. This score is used in bail, sentencing, and parole decisions. The system, or similar ones, is used in 46 states, including New York, Pennsylvania, Wisconsin, California, and Florida.
The problem is that nobody knows exactly how the algorithm works.
It takes 137 variables as input: number of prior arrests, age, employment status, education level, neighborhood of residence, whether family members have criminal records. How much weight is assigned to each variable is a trade secret.
Judges don't know. Defendants don't know. Defense attorneys don't know.
(2) The ProPublica Investigation
In May 2016, the investigative news organization ProPublica published an analysis of data from more than 7,000 defendants in Broward County. They tracked COMPAS scores against actual recidivism over a two-year period.
The results were alarming.
Overall accuracy was just 61 percent. Slightly better than a coin flip. What was worse was the pattern of errors.
Black defendants who did not reoffend were nonetheless classified as high risk (the false positive rate) 45 percent of the time. For white defendants, the figure was 23 percent. Nearly double. Conversely, the rate at which people who did reoffend were classified as low risk (the false negative rate) was higher for white defendants: 28 percent for Black defendants, 48 percent for white defendants.
The summary is this: the algorithm rated Black people as more dangerous than they actually were, and white people as less dangerous than they actually were.
Northpointe pushed back. Their argument went like this.
COMPAS does not use race as a direct variable. Black and white defendants with the same score reoffend at the same rate. This is a fairness criterion called 'calibration.' Our algorithm satisfies this criterion.
Both claims were factually correct. And they cannot both be true at the same time. Mathematicians call this the 'impossibility of fairness.' When two groups have different base rates of recidivism, satisfying both 'calibration' and 'equal error rates' simultaneously is mathematically impossible. Whichever definition you choose, you violate the other.
The problem ran deeper still.
COMPAS does not ask about race directly. But it asks about neighborhood of residence. It asks about employment status. It asks about education level. It asks whether family members have criminal records. In the United States, all of these variables correlate strongly with race. Black neighborhoods are patrolled more heavily, and more arrests occur there. This data is fed into the algorithm. The algorithm is not predicting the future; it is quantifying past inequality.
In 2016, the Wisconsin Supreme Court allowed the use of COMPAS in State v. Loomis. Conditions were attached. Judges must not determine sentences based on COMPAS scores alone, and they must inform defendants of the algorithm's limitations. In practice, nobody monitored whether these conditions were being followed.
In 2018, researchers at Dartmouth College ran a revealing experiment. They showed defendant information to ordinary people with no expertise in criminal records and asked them to predict recidivism. Individual accuracy was 63%. When multiple people's judgments were combined, it reached 67%. COMPAS scored 65%. A complex algorithm using 137 variables performed no better than the intuition of untrained volunteers.
There was an even more startling finding. The researchers achieved the same accuracy as COMPAS using only two variables: age and number of prior convictions. The remaining 135 variables added no predictive power whatsoever. The complexity served mystification, not accuracy.
In 2024, researchers at Williams College published findings showing that COMPAS use in Broward County had lowered overall incarceration rates but simultaneously widened the racial gap in incarceration. The algorithm distributed leniency to some and harshness to others. The basis for that distribution correlated with skin color.
The question remains. Can an algorithm trained on biased historical data produce a fair future? Or is it nothing more than a tool that automates inequality and deflects accountability?
C. Facial Recognition Regulation
(1) The San Francisco Ban Ordinance
On May 14, 2019, the San Francisco Board of Supervisors passed a resolution by a vote of 8 to 1. It was the first major U.S. city to ban the use of facial recognition technology by city government agencies, including the police.
The sponsor was Supervisor Aaron Peskin. He called it the 'Stop Secret Surveillance Ordinance.'
The core provisions were threefold.
First, all city government agencies, including the San Francisco Police Department, are prohibited from using facial recognition technology.
Second, the adoption of surveillance technology requires annual transparency reports and audits.
Third, a civilian oversight board monitors the use of such technology.
There were dissenting voices. Some civil liberties groups advocated a moratorium rather than an outright ban. Joel Engardio of Stop Crime SF told NPR: 'It shouldn't be used right now. The failure rate is too high. But instead of a permanent ban, we should leave the door open for when the technology improves.'
San Francisco's decision influenced other cities. Oakland followed. Boston followed. Portland joined. As of 2024, dozens of cities and several states have enacted ordinances restricting or banning facial recognition use.
Whether the ban ordinances were actually being obeyed was another matter. In May 2024, the Washington Post published a startling report. Police in jurisdictions where facial recognition was banned were circumventing the prohibition by requesting searches from neighboring jurisdictions. In July 2024, civil rights group Secure Justice filed a lawsuit alleging that the San Francisco Police Department (SFPD) had violated the ordinance. Documents they obtained showed at least 12 instances where SFPD requested facial recognition searches from other jurisdictions. That was far more than the 6 cases the department had acknowledged.
The limitations of ban ordinances became clear. Technology knows no borders. If one city bans it but the next city allows it, the prohibition is meaningless. Unless the federal government establishes uniform standards, regulation will always have holes.
(2) The Gender Shades Study
In 2018, MIT Media Lab doctoral student Joy Buolamwini and Google AI researcher Timnit Gebru published an academic paper.
It was titled 'Gender Shades.' The paper changed the history of facial recognition technology.
The researchers tested gender classification algorithms from IBM, Microsoft, and Chinese company Face++. They divided 1,270 facial images into four groups: lighter-skinned males, lighter-skinned females, darker-skinned males, and darker-skinned females.
The results were unambiguous. The error rate for lighter-skinned males was less than 1%.
The error rate for darker-skinned females was as high as 34.7%.
A 35-fold difference.
Across all algorithms, darker-skinned females showed the highest error rates.
The cause lay in the training data.
The researchers found that training datasets for commercial facial recognition systems consisted of 79.6% to 86.2% lighter-skinned faces. Algorithms recognize best the faces they have learned. Systems trained predominantly on white male faces by white male engineers recognized white males best.
In 2019, the National Institute of Standards and Technology (NIST) conducted an independent evaluation of 189 facial recognition algorithms. It confirmed the conclusions of Gender Shades. False positive rates (misidentifying one person as another) for Asian and African American faces were 10 to 100 times higher than for white faces. Women were misidentified more often than men. The study's impact was immediate.
IBM announced its withdrawal from the facial recognition market. Microsoft and Amazon halted sales of facial recognition technology to police. Buolamwini founded the Algorithmic Justice League to continue raising the issue of AI bias.
There were criticisms. The Security Industry Association argued in a 2024 report that Gender Shades had tested 'gender classification' algorithms, not 'facial recognition,' and that these are different technologies. They also contended that algorithms tested in 2017 do not reflect current technical capabilities. According to NIST's 2024 evaluation, the top 100 algorithms achieved accuracy above 99.5% across all racial groups.
But the central question persists. A 99.5% accuracy rate means a 0.5% error rate. If this technology is applied to millions of people, thousands are misidentified. And the question is who those thousands are. In the case of Kaleace Periman, represented by the Minnesota ACLU, an innocent Black man was arrested and detained due to a facial recognition error. The algorithm's 0.5% error becomes a 100% catastrophe in one person's life.
The Gender Shades study posed a question that went beyond its technical findings. Whose face is the default? Whose face did the algorithm learn as 'normal'? What price do those who fall outside that 'normal' pay?
D. Educational AI Disputes
(1) The Hingham High School Suspension Case
In December 2023, RNH, a senior at Hingham High School in Massachusetts, was working on a history class project. It was an entry for National History Day. He and a classmate formed a team.
RNH used AI during his research process. He got help structuring his topic, organizing materials, and drafting an outline. He did not have the AI write the paper itself. At least that was his claim.
Teacher Andrew Hoey saw it differently. He determined that he had found traces of AI use in RNH's submission. The problem was that Hingham High School's 2023-2024 student handbook contained no explicit policy on AI use. There was a provision prohibiting 'unauthorized use of technology during assessments,' but whether AI fell under this clause was unclear.
RNH received a zero on the project. He was required to resubmit a new project. On his second attempt, he received a D. An academic dishonesty record was placed in his file. He was excluded from membership in the National Honor Society.
RNH's parents were furious. Their son was a student with a GPA above 4.0, a perfect ACT score, and a 1520 SAT. Early decision season for college applications was approaching. An academic dishonesty record would be fatal for admission to elite universities.
In September 2024, the family filed suit in Plymouth Superior Court. The defendants were the Hingham School Committee, the superintendent, the principal, and the teacher. The case was soon transferred to Boston federal court. The core of the plaintiff's argument was this: because there was no explicit AI prohibition policy, the punishment was 'arbitrary and capricious.' The project instructions made no mention of prohibiting AI use. The student's constitutional rights had been violated.
In October 2024, Federal Magistrate Judge Paul G. Levenson denied the motion for a preliminary injunction. His ruling was direct. The school could 'reasonably conclude' that the student had violated academic integrity by 'indiscriminately copying AI text.' Federal courts do not intervene in school disciplinary decisions unless they are clearly unjust.
The preliminary injunction was denied, but the lawsuit itself continues.
The family's attorney stated plans to obtain additional evidence through the discovery process. Meanwhile, Hingham High School added explicit provisions regarding AI use to its 2024-2025 student handbook.
(2) Doe v. Yale University
In the spring of 2024, a student in Yale School of Management's Executive Master of Business Administration (EMBA) program submitted a final exam for the course 'Sourcing and Managing Funds.' It was 30 pages long. It was among the longest answers in the entire class.
The teaching assistant sensed something was off.
The sentences were "abnormally long and sophisticated in structure."
"Near-perfect punctuation and grammar."
The professor ran the answers through GPTZero, an AI detection tool.
The result flagged suspected AI use.
The student in question was Thierry Rignol, a French national and entrepreneur living in Texas who operated hotel and real estate businesses in Mexico. He had enrolled in the EMBA program in July 2023 and was set to graduate in May 2025.
On July 24, 2024, Rignol met with the associate dean and the dean of students. According to the lawsuit, the dean made "multiple attempts to coerce the plaintiff into making a false confession of an honor code violation."
Had he used tools like Grammarly? Had he consulted with other students or teaching assistants? Was there any confusion about the exam rules? Rignol answered no to every question.
The dean implied that "his F1 visa could be revoked and he could be deported," the lawsuit alleges. Rignol was not in the country on an F1 visa.
Throughout August, correspondence went back and forth with the honor committee. Rignol was formally notified of an honor code violation charge. An investigation and hearing followed. The outcome was a one-year suspension and an F grade in the course.
In February 2025, Rignol filed suit against Yale University in the U.S. District Court for the District of Connecticut. He initially proceeded under the pseudonym "John Doe" before his real name was later disclosed. His claims were layered. First, AI detection tools including GPTZero are unreliable. Yale's own department had acknowledged that "no artificial intelligence tool can reliably detect artificial intelligence use." Second, he is a non-native English speaker, and according to Stanford University research, AI detection tools show higher false positive rates for non-native speakers. This constitutes discrimination based on national origin. Third, the hearing process violated due process.
In May 2025, federal judge Sarah Russell denied Rignol's motion for a preliminary injunction. Rignol had asked to graduate alongside the class of 2025, but the judge ruled that he "failed to meet his burden of showing that the interruption to his studies until the start of the next semester (Fall 2025) and an F on his transcript constitute irreparable harm."
(3) K-12 AI Guidelines and State-by-State Legislation
The Hingham and Yale cases are part of a larger problem. When ChatGPT launched in November 2022, not a single U.S. state had an education policy on generative AI.
By April 2025, at least 28 states had issued official guidelines on AI in K-12 settings.
The pace of change was rapid.
Schools initially responded with bans. The New York City Department of Education blocked access to ChatGPT. Los Angeles, Seattle, and Baltimore followed.
But the bans did not last. Students bypassed them on personal devices. Teachers recognized the educational potential. By late 2024, most school districts had shifted from "ban AI" to "manage AI."
State approaches fall broadly into three categories.
First, states that require mandatory policy adoption. Ohio passed a law in summer 2024 requiring all public K-12 schools to develop, approve, and publish comprehensive AI policies by July 2026. Tennessee mandated that each school district establish and publicize an AI use policy beginning March 2024. As of August 2025, these are the only two states with a legal mandate.
Second, states that provide guidelines without enforcement. California, Massachusetts, Colorado, Washington, and most other states fall into this category. California formed an AI working group under Senate Bill 1288 of 2024 and published comprehensive guidelines. Educators, students, and industry experts participated. Louisiana introduced a four-tier AI integration framework (AI Prohibited, AI Assisted, AI Augmented, AI Empowered). Nevada developed the STELLAR principles (Security, Transparency, Empowerment, Learning, Leadership, Achievement, Responsible Use).
Third, states that have taken little action so far. In these states, individual school districts are crafting their own policies. There is no consistency, and students face different rules depending on which district they live in. Movement is happening at the federal level as well. In November 2024, the U.S. Department of Education published guidance describing 21 scenarios in which AI could violate students' civil rights. In April 2025, President Trump signed an executive order titled "Advancing Artificial Intelligence Education for American Youth." It emphasized AI literacy and proficiency, integrating AI into education, and teacher training.
According to an analysis by the Education Commission of the States, state-level discussions are converging around several common themes. AI literacy should be taught across all subjects, not confined to computer science. Investment in teacher professional development is needed. Equity must be ensured. Data privacy must be protected. Academic integrity policies must be clearly defined.
An unresolved question remains: the reliability of AI detection tools. Tools like Turnitin and GPTZero are imperfect. Both false positives (flagging human-written work as AI-generated) and false negatives (failing to detect actual AI use) occur. Error rates vary depending on whether the writer is a non-native speaker, on particular writing styles, or on subject matter.
A more fundamental question also persists. Where is the line between writing with AI and writing without it? Is using a spell checker acceptable? What about Grammarly? Asking ChatGPT for ideas? Asking it to outline your argument? Asking it to draft a paragraph? Where do you draw the line?
RNH in Hingham and Rignol at Yale were punished in a world that has no settled answer to these questions. However their lawsuits end, one thing is clear. Schools cannot ban AI. They cannot ignore it either. The only option is to find a way to live with it. What that way looks like, nobody yet knows.
Kim Kyung-jin
Attorney · Former Member of the National Assembly · AI Policy Researcher
© 2026 Kim Kyung-jin. All rights reserved.