The question of whether artificial intelligence can truly outperform humans in understanding the World Cup was brought to the forefront by Bloomberg in a recent report on the AI prediction craze in China.
The report detailed a "World Cup Prediction Man vs. Machine" contest organized by LENOVO GROUP, a FIFA official technology partner, and China Mobile's streaming platform Migu. The event pitted twelve major AI models against tens of millions of Chinese viewers, aiming to answer whether AI could defeat humans in football forecasting and which model would come out on top.
According to the Bloomberg report, as of June 29th, the AI camp had a prediction accuracy rate of 62%, leading the human camp's 55%. While these figures might suggest a temporary lead for AI, the report cautioned against overinterpreting the results.
Bloomberg interviewed Abraham Wyner, a professor of sports analytics at the Wharton School of the University of Pennsylvania. Wyner noted that from a statistical perspective, the sample size of 104 World Cup matches is too small to determine if any model possesses a genuine advantage. He also suggested that in sports prediction, the edge offered by AI over simpler statistical models might be limited.
The report further stated that high-quality data for AI training is scarce for the World Cup. Final team rosters are often confirmed just weeks before the tournament begins, and moments of individual brilliance from players are factors not easily quantified for AI consideration.
This tension—AI's temporary lead without definitive conclusions—is precisely what creates the news value of the contest organized by LENOVO GROUP and Migu.
Public information shows the "World Cup Prediction Man vs. Machine" contest was convened by Lenovo's Tianxi AI Super Agent, bringing together major domestic AI models like DeepSeek, Kimi, Baidu's ERNIE, Alibaba's Tongyi Qianwen, and China Mobile's Jiutian to form a prediction alliance for all matches of the 2026 FIFA World Cup.
Bloomberg's report framed the competition as a new battle of "brainpower vs. computing power," reminiscent of the Go match a decade ago between DeepMind's AlphaGo and Lee Sedol. However, football is not like Go. Go has clearer rules and a finite board state, while football is rife with randomness: injuries, weather, red cards, goalkeeper form, a deflection, a penalty, or even psychological pressure in the final minutes can change the outcome.
This inherent unpredictability makes forecasting football more challenging for AI than many structured problems. Models can analyze past records, player form, odds fluctuations, weather, and tactics, but they struggle to fully account for on-pitch randomness. A representative from the Alibaba project applying the Tongyi Qianwen model to World Cup predictions, cited in the Bloomberg report, noted that football's greatest charm lies in its unpredictability; anyone claiming 100% prediction accuracy is likely a fraud, not an AI.
This provides another layer of interpretation for the contest: its goal is not to present AI as a new oracle, but to place AI within the real uncertainty of sporting events, subjecting model capabilities to public testing.
The drama of this man-machine battle isn't solely from the AI side. As the tournament progressed, a highly viral example emerged from the human camp: an ordinary user from Chongqing Pengshui correctly predicted 31 out of 32 teams advancing, becoming one of the very few among tens of millions of participants to come close to a perfect score. Compared to the twelve AIs and various expert guests, this achievement by an ordinary fan makes the question of whether AI "understands" the World Cup more intriguing.
While Bloomberg focused on how Chinese AI companies are turning the World Cup into a new competitive arena, the aspect of the contest that truly sparked discussion in China was its placement of AI, experts, and ordinary people on the same playing field. AI can leverage agents, analyze data, and synthesize odds and weather; ordinary fans might also deliver astonishing answers based on long-term viewing experience, intuition, and an understanding of team character.
This transforms the "man vs. machine" contest from a one-dimensional tech showcase into an open examination: AI, human viewers, and commentator guests all make predictions under the same rules, with the actual match results serving as the final grader.
LENOVO GROUP and Migu have further turned this prediction contest into a program. According to public reports, "Man vs. Machine: Who is the World Cup Prophet?" began airing on Migu Video on June 24th, featuring 20 live episodes. On the show, guests and AI jointly predict match outcomes and scores, with each episode revealing a public prediction accuracy ranking and providing post-match analysis.
Bloomberg also noted that LENOVO GROUP and Migu didn't just organize a competition but also turned the concept into a reality show. The report described how the program displayed logos of the twelve AI models and featured guests including football commentators, internet influencers, and stand-up comedians. This translates the competition among AIs from internal model leaderboards within tech circles into sports entertainment content that ordinary viewers can watch, participate in, and discuss.
Citing information from LENOVO GROUP, Bloomberg reported that as of June 25th, over 20 million people had participated in the "World Cup Prediction Man vs. Machine" contest; that number has now reportedly approached 30 million.
This scale indicates that World Cup prediction provides a low-barrier entry point for the public to understand AI. Previously, user comprehension of AI often came through model launch events, technical rankings, parameter scales, and benchmark scores. In World Cup prediction, the judgment criteria are much simpler: what was predicted before the match, and was it correct afterward? Every upset, every draw, every collective misprediction makes the capabilities and limitations of AI clearer.
The value of the LENOVO GROUP and Migu "man vs. machine" contest lies not in simply declaring that AI has beaten humans, nor in proving that a particular model definitively understands football better. Its value lies in repeatedly testing AI capabilities within a real, uncertain, and emotionally charged public scenario.
AI may not yet understand football better than humans. Expert views cited by Bloomberg remind us that 104 matches are insufficient to prove a model's decisive advantage. Precisely for this reason, this contest remains worthy of ongoing observation. It does not provide a final answer but lays the question bare: when AI enters a field filled with randomness, emotion, and human experience, how far can it truly go?
The World Cup continues, and so does the man-machine battle, with each match presenting the next test.
Comments