"Set formations and even opponent strategies provided." How far has AI come? We entrusted it with the task of selecting Son Heung-min for the Korea-South Africa match. [22nd Anniversary Special]

*This content was translated by AI.

Artificial intelligence (AI) no longer remains limited to tallying statistics after a match ends. It reads opponents' movements, predicts the next scene, and even suggests how to adjust players' positions to achieve desired outcomes. By comparing countless options faster than humans can, it also proposes more effective tactical alternatives. AI has now entered the realm of decision-making for coaches and managers. Football stands squarely in the middle of this massive transformation.

Currently, the field where AI is most widely utilized in soccer is player evaluation. Go Sang-gi of the University of Seoul told StarNews in an interview, "The most basic example is expected goals, or 'xG.' It uses machine learning models to calculate the probability that a shot will result in a goal in a specific situation." He added, "Recently, models have emerged that evaluate the value of all actions performed by players on the field, including passes, shots, and dribbles, as well as off-the-ball movements when they do not possess the ball."

However, AI is advancing rapidly. While it previously focused on organizing records of events that had already occurred, it now also predicts situations that have not yet happened.

Professor Ko stated, "Pass success rate can be calculated by tallying how many passes were attempted and how many of them succeeded. Existing data analysis analyzes the statistical characteristics inherent in actual data like this." He added, "In contrast, AI-based models predict even scenarios that have not occurred and, based on these predictions, assess the value of specific situations and plays."

The predictive capabilities of AI are expanding beyond player evaluation to include tactical design. A representative recent example is the Brazilian powerhouse Palmeiras. Google DeepMind introduced in June that Palmeiras became the first football club to fully integrate its football tactical assistance system, "TacticAI," into routine open-play analysis.

Initially, the tactical AI focused on a limited set-piece situation: corner kicks. Palmeiras expanded this to open play, which refers to general match situations where the ball is in constant motion, excluding set pieces.

AI marks each of the 22nd players on the field as a single point and analyzes their positions, distances, speeds, and movements by connecting them into a network of relationships. Based on this analysis, it simulates how the overall tactical structure changes when the position of a specific player is altered. According to Google DeepMind, the system can predict the ball's location up to eight seconds in advance.

For example, when the Palmeiras data analysis team advances a left-sided defender by several meters on the screen, AI calculates and displays how the remaining 21 players react and which spaces open up or close as a result. This allows for comparing various tactical formations in a virtual space before testing them on the actual field. Google DeepMind has evaluated this as "a huge leap forward in sports technology."

The starting point for Tactic AI was Liverpool, a prestigious club in the English Premier League (EPL). Google DeepMind conducted several years of AI research with Liverpool to analyze football and finally unveiled Tactic AI in 2024, specialized for corner kicks.

At the time, TacticAI was designed to answer three key questions in corner kick situations. The first was "what will happen given the current player positioning." The AI predicted which player was most likely to receive the ball first when the kicker took the kick, and how likely a shot would occur afterward.

The second was "how to understand what has already happened." The team identified past corner kick scenarios similar to the current situation and analyzed what outcomes resulted from similar tactics.

The third question was "What needs to be changed to create the desired outcome?" In a defensive situation, it proposed adjusting the positioning and movement of which defender to reduce the opponent's shooting probability.

Tactic AI did not provide a single answer. It generated multiple alternative scenarios and enabled coaching staff to compare the pros and cons of each.

The results were also impressive. Google DeepMind explained that TacticAI accurately predicted which player would first receive the ball during corner kicks and whether a shot would occur, and that the player positioning adjustments suggested by the AI closely resembled actual movements observed in real matches.

The research team also conducted a blind evaluation targeting football experts affiliated with Liverpool. They compared the two tactics without revealing which one was actually used in real matches and which one had been modified by TacticAI.

The results were surprising. Liverpool experts could not easily distinguish the corner kick placements created by TacticsAI from actual tactics used in real matches. Furthermore, 90% of evaluators preferred the tactics proposed by AI after modifications over the existing placements.

The scope of AI application extends beyond tactics alone. Spanish club Sevilla has developed a generative AI-based "Scout Advisor" in collaboration with IBM. The system allows the AI to search and summarize over 200,000 scouting reports and player data accumulated by the club, identifying potential transfer candidates that meet specific criteria.

South Korea's promising player Kim Min-jae, who transferred to the Scottish powerhouse Rangers, and Son Heung-min, captain of the South Korean national team playing for Los Angeles FC in Major League Soccer (MLS) in the United States, are also known to utilize artificial intelligence in areas such as managing training loads and detecting injury risks.

Then, can AI take on the roles of coaches and managers? If it is tasked with setting the starting lineup and formation, designing strategies to exploit opponents, and managing tactical shifts and substitution plans after goals are scored or conceded, will it produce answers different from those of human managers. Will AI, which processes more information faster than humans, make more rational decisions? Or will it reveal clear limitations in front of soccer, where countless variables intertwine?

It is assessed that it is technically feasible to recommend formation setups and strategies to counter opponents' tactics. Professor Ko stated, "AI can recommend which plays to make in specific game situations to maximize our team's scoring probability," adding, "Determining the optimal positioning of given players to achieve the best results is also a task AI handles well."

He added, "Since soccer is a sport played by humans, there will remain many areas where human intervention is necessary." However, he noted that the story changes when it comes to making efficient decisions or choosing options that increase the probability of success. He believes that if sufficient data is provided, the likelihood of humans outperforming AI in these analytical areas is low. These domains can be gradually replaced by AI at a rapid pace.

There are also cases of actual overseas research analyzing what formation a team should adopt to achieve the highest pressure success rate when an opponent builds up with a back three.

Professor Ko stated, "By utilizing AI, it will be possible to recommend our team's optimal formation tailored to the opponent's formation or suggest which actions are best in a given situation." He added, "Technically, this is already feasible today. However, how much trust is placed in AI results and whether they are actually utilized on the field remains an issue that requires more time."

It was also explained that general-purpose generative AIs such as ChatGPT, Claude, and Gemini could provide tactical recommendations at a certain level if they are provided with sufficient data, even though they were not created specifically for soccer.

Professor Ko said, "A soccer-specific AI predicts the next situation or recommends optimal decisions based on game data," and added, "If sufficient data is provided to generative AIs such as ChatGPT, Claude, or Gemini, they could analyze it and reach certain conclusions."

Simply asking an AI, "What tactic is good?" is not enough. He emphasized that "AI cannot make accurate judgments just by being asked without data, because the information itself needed for judgment is missing." He stressed that "what matters is how quickly and accurately data and player information generated in complex soccer matches can be digitized and input into AI."

He added, "If there is sufficient data on how players have performed for their respective teams, as well as their current physical and fitness conditions, it should be possible to recommend player selections and tactical strategies."

Accordingly, StarNews entrusted three generative AIs — ChatGPT, Claude, and Gemini — with the key pre-match decisions for the Group A third-round match of the 2026 FIFA World Cup between South Korea and South Africa. At that time, South Korea could have advanced to the Round of 32 with just a draw, but it lost 0-1 to South Africa and was eliminated in the group stage.

In the experiment, the focus was placed on minimizing the possibility that the AI would encounter actual matches and results during its learning process. The team names "Korea" and "South Africa" were changed to Team A and Team B, respectively. Player names were also anonymized as A1·A2 and B1·B2.

The three AIs were provided with completely identical data. The dataset included each player's number of A-match appearances and goals, primary and secondary positions, height, participation records in the first and second group stage matches, cumulative playing time, disciplinary warnings and injury status, team-level match metrics, and group stage rankings.

Both teams were also provided with the basic formations, offensive directions, defensive strategies, and passing and shooting metrics used in the first and second matches they had previously played. It was also announced that Team A would advance to the next round if the result was a draw or better, while Team B required a victory to proceed.

The items to be decided by AI were fixed to align with the tasks a real coach prepares before a game. The system was designed to provide guidance on the starting 11 and basic formation, build-up style, pressing height, priority areas for attack, countermeasures against key opposing players, tactical adjustments after scoring or conceding goals, substitution plans, and the match scenarios that require the most vigilance.

Each AI received the same question 10 times in separate conversations without influencing one another. The total of 30 responses were categorized according to pre-determined items such as formation, starting lineup, pressing height, attacking direction, and player-specific roles. By synthesizing the most frequently selected options across these categories, a common tactical plan shared by the AIs was constructed.

In the total of 30 responses, the most frequently recommended basic formation was 3-4-2-1. Kim Seung-gyu (FC Tokyo) was selected as the goalkeeper. The three-man defense consisted of Lee Han-beom (Club Brugge), Kim Min-jae (Bayern Munich), and Lee Ki-hyeok (Gangwon FC). Kim Moon-hwan (Daejeon Hana Citizen) and Lee Tae-seok (Austria Vienna) were chosen as the wing-backs on both sides. Baek Seung-ho (Birmingham City) and Hwang In-beom (FC Porto) formed the midfield. Lee Jae-sung (Mainz) and Son Heung-min (LAFC) took positions in the second line, while Cho Gue-sung (Midtjylland) was named as the striker at the forefront.

South Korea actually deployed a 3-4-2-1 formation against South Africa. However, there were clear differences in the player lineup. In the actual starting XI, Oh Hyun-gyu (Besiktas) led the front line, supported by Hwang Hee-chan (Wolverhampton) and Lee Kang-in (Atletico Madrid). Lee Tae-seok and Seol Young-woo (Augsburg) operated as wing-backs on both flanks, while Hwang In-beom and Baek Seung-ho anchored the midfield. The three center-backs and goalkeeper were Lee Ki-hyuk, Kim Min-jae, Lee Han-beom, and Kim Seung-gyu, matching AI's selection exactly.

The biggest difference was whether Son Heung-min and Lee Jae-sung would start. In the actual match, both players began on the bench, but the AI included them in the starting lineup after considering their experience, attacking proficiency, and ability to manage the game.

The AI selected center-backs Lee Han-beom and Kim Min-jae, who started both the first and second matches, as well as Lee Gi-hyeok, without changes. The reason given was: "Although their cumulative playing time has been high, there was a six-day recovery period until the South Africa match, and we judged that the potential loss of organizational cohesion from changing the entire three-back system could outweigh the physical burden."

A notable selection was the decision to exclude South Korea's representative team ace, Lee Kang-in, from the starting lineup. The AI reported that "when prioritizing defensive involvement and stability in the midfield over offensive creativity, Lee Jae-sung is deemed more suitable." However, some responses also suggested that if the goal is to aggressively attack the opponent's right flank, starting with Lee Kang-in could be more effective.

AI also sought to leverage the fact that South Africa frequently utilizes wing attacks. It recommended exploiting the space left behind when South African full-backs advance by having Son Heung-min and the wing-backs penetrate, followed by a rapid counter-attack immediately after the opponent loses possession. The AI also identified Baek Seung-ho as a key player. Analysis suggested that if Baek Seung-ho protects the area in front of the three-man defense and blocks the central passing lanes where South Africa's counterattacks begin, the team can come closer to achieving its goal of at least a draw.

In addition, the team presented pre-match scenarios detailing how to adjust defensive pressure and offensive intensity after scoring first, which players to substitute in to shift the attacking formation after conceding first, and what risks must be accepted if the score remains tied until the 60th minute of the second half. These plans also included details on substitute players, their timing for entry, and the game flow patterns that require the highest level of vigilance.

It goes beyond simply selecting a starting lineup and setting up a formation in a soccer game; it connects various options that a real coach prepares before a match through a consistent logic.

However, presenting a plausible tactical plan does not mean AI can completely replace the duties of coaches and managers. Professor Ko stated, "A manager is not merely someone who decides players' positions." He added, "After deploying a player in a specific position, the manager must directly communicate how to play, motivate the player, and persuade them with their own judgment."

He further pointed out that "AI, on the other hand, may be limited to making decisions and providing recommendations."

A player's personality and their relationships within the locker room are areas that AI finds difficult to judge easily. Professor Ko stated, "Whether a player actually has a certain personality and how well they can fit in with other players in the locker room still requires direct human verification," adding, "AI lacks sufficient data on these aspects. For now, humans and AI must work in a complementary manner."

The Irish Times also highlighted the Liverpool and Google DeepMind's Tactic AI project on the 14th, noting that one must distinguish between the potential of technology and its real-world performance.

According to media reports, the research team collected 9,693 corner kicks from two and a half seasons of the EPL. They recorded the height and weight of all players involved in each corner kick, along with their starting positions and movements, then developed TacticAI to process this data. Based on this vast dataset, the system identifies repetitive patterns and enables coaching staff to more quickly determine the factors that influence the success or failure of tactics.

The importance of corner kicks in modern soccer is growing steadily. In the 2025–2026 EPL season, 18% of all goals originated from corner kicks, a significant increase from 12% in the previous season. Goals scored from set pieces, excluding penalty kicks, have also surpassed one-quarter of the total. As set pieces emerge as a key tactical area that can determine match outcomes, opportunities for AI utilization are expanding accordingly.

However, Liverpool's actual performance fell short of expectations. Even in the season when they won the title under former manager Arne Slot, Liverpool's set-piece performance remained at the mid-table level of the EPL. In the 2025-2026 season, set-piece weaknesses were identified as a core issue behind the team's poor form.

Liverpool ultimately dismissed set-piece coach Aaron Briggs midway through the season. At that time, Liverpool had conceded 12 goals from set-pieces while scoring only three. Briggs later reflected, "The club made every effort to provide substantial support, and the analysis team offered active assistance."

Liverpool drew a line by stating that at the time of the project's launch, it was merely in a research phase exploring the potential of TacticalAI, and there was no decision to use it on match day. Nevertheless, The Irish Times predicted that "it is unlikely that Liverpool's coaching staff ignored AI analysis information."

Instead, The Irish Times emphasized that AI does not make all decisions independently; rather, humans review AI suggestions and make the final selection. The media explained, "For AI tools to function properly, humans must first provide industry-scale relevant data and ask appropriate questions." It further noted, "And if humans accept the proposals offered by AI, those ideas must be implementable in actual stadiums."

He added, "Liverpool may have had the most insightful data on set-piece issues in history," but also analyzed that "there may not have been a suitable human to implement those solutions on the field."

He also emphasized that "the relationship between sports and AI has long surpassed the stage of exploring each other's potential. While there may be no romance in between, this relationship will continue forever."

Google DeepMind also defined Tactic AI not as a system replacing human coaches, but as an "AI assistant" that enhances the decision-making capabilities of coaches and coaching staff.

Professor Ko also predicted that, realistically, it is highly likely that AI will establish itself as an 'advisor' to coaches. He stated, "Coaches are currently deliberating whether to use a three-back or four-back formation before the match. If AI determines that a specific tactic is superior, the coach can discuss the reasoning with the AI and gain a clearer understanding of their own ideas." He added, "This could also be utilized when explaining tactics to players and persuading them."

He added that in the future, coaches are more likely to evolve not by being completely replaced by AI, but by leveraging AI to better persuade players and develop superior tactics. He emphasized that it is desirable for humans to remain the final decision-makers while AI serves as a tool to support better decision-making.

How well AI is utilized could also affect the future competitiveness of teams and their coaches. Professor Ko noted that while the difference between teams using AI and those not using it may not be immediately apparent, "if better decisions are repeatedly accumulated, that gap will gradually widen."

For those preparing to enter the soccer industry, AI utilization capabilities are becoming a new competitive edge. Professor Ko stated, "It is important to quickly learn what tasks AI can perform in the soccer field," adding, "One must understand which parts of the work they wish to do can be assisted by AI. This is not a matter specific only to soccer but will apply to almost all professions moving forward."

He also emphasized that "in the future, regardless of one's profession, an era will come where people work alongside AI," and added that "whether one wishes to be scouted or become a leader, it is essential to clearly understand how AI can assist in their specific role."

<© STARNEWS. All rights reserved. No reproduction or redistribution allowed.>

*This content was translated by AI.