在2018年世界国际象棋冠军赛中，Magnus Carlsen 与 Fabiano Caruana 的12局古典赛共历时超过50小时，全部收敛为和局，这是自1886年以来冠军赛首次；文中将其视为最高水准「近乎完美对局」的征兆。作者追溯到1925年 José Raúl Capablanca 的忧虑：高手越来越接近「随时和局」。当时大师级古典局平局率长期约为50%，而人工智慧出现后，这一趋势加速。1997年 Garry Kasparov 在 IBM Deep Blue 下败北，2006年桌上型电脑即可跑出可压制最强人类的棋程式；目前 Stockfish 评分为3653，几乎高出 Magnus Carlsen 的人类高峰约800分。

开局研究因引擎普及而更可量化，人人能看到同样的最优判断，职业棋手虽仍记住前10到20步，但可利用的「最佳方案」逐渐重叠，使精英对局更易陷入和局。Carlsen 的对策是避开过度定式化：在2021年防卫战中遭遇一场8小时对局与7次和局后，他放弃第四次卫冕，并更投入 rapid、blitz 与 freestyle（起始局面随机化）。即便如此，他仍于2025年夺冠 Norway Chess，并同时持有三种赛制的冠军，且在长局仍偏少和局，主要因实力明显优势；反而是对手把「先求和」视为务实选择。

2024年国际象棋 Candidates 赛事显示新世代理念转向：18岁的 Praggnanandhaa Rameshbabu 在 Ruy Lopez 采用引擎判为亚优的著法，且此路线据称25年未见其对手用法，结果迫使对手脱离开局理论并最终取胜。Peter Doggers 指出，现代开局已不再稳定追求可量化的明显优势，而是靠出奇招造成对手认知负担；Hikaru Nakamura 在同届赛事同样采用少见变化并取得和局。核心转变是：最佳著法不等于最好对局结果，能迫使对手启用棋感与临场判断，往往比仅依赖引擎更有效。Jan Gustafsson 所谓「按空格键」式依赖，会在偏离理想谱例时失效。AlphaZero与 Stockfish透过自我对弈改善机率决策，但大型语言模型在象棋上仍可能出错或幻觉；因此，随著决策更外包给AI，真正的上限会回到人类的理解、临场适应与心理博弈上。

In 2018 the World Chess Championship’s 12-game classical match between Magnus Carlsen and Fabiano Caruana lasted over 50 hours and ended in 12 draws, the first all-draw result in the event’s history since 1886. The article frames this as a “perfect play” era, echoing José Raúl Capablanca’s 1925 warning that elite chess seemed to be drifting toward draw-by-default outcomes. Since then, the draw rate among masters in classical chess had long hovered around 50%. AI accelerated this trend: Kasparov lost to IBM’s Deep Blue in 1997, and by 2006 a home desktop engine could beat top human players. Today Stockfish is rated 3653, almost 800 points above Magnus Carlsen’s best human level.

With universal engine access, opening preparation became standardized; players often memorize the first 10–20 moves but now share the same move-selection logic, producing narrower, more drawish top-level variation trees. Carlsen’s response was to step outside classical norms. After the 2021 title defense that included one 8-hour game and seven draws, he declined a fourth defense, then focused more on rapid and blitz and on freestyle chess, where starting positions are randomized. He still won Norway Chess in 2025 and holds world titles in all three formats, yet in long-form play his own draw rate stays relatively low because of superior strength, so challengers often treat drawing as safer than pressing for a win.

The next generation has started exploiting AI’s blind spot: engines optimize for best-line consistency, not fallible human resistance. In the 2024 Candidates Tournament, 18-year-old Praggnanandhaa Rameshbabu played an opening move in the Ruy Lopez judged inferior by engines and unseen in that line for 25 years, then won when his opponent was forced out of book; Hikaru Nakamura used a similarly unconventional variation and drew. The trend shifts from “find the best move” to “choose the move that forces opponent thought.” Jan Gustafsson’s “spacebar” critique captures this: repeated engine queries are useless if a position changes. AlphaZero/Stockfish improve through self-play probabilities; LLMs still struggle at chess and may hallucinate despite fluent explanations. As decisions are outsourced to AI, understanding becomes rarer, so top players win by surprise, psychology, context, and adaptive judgment.