What Happened
On February 27, MIT Technology Review published a reported feature examining how DeepMind's AlphaGo has restructured professional Go in the decade since its 2016 victory over Lee Sedol. The reporting, based in Seoul at the Korea Baduk Association, found that AI tools have become mandatory for professional play, and that the nature of preparation and competition has shifted fundamentally as a result.
Players now train by replicating AI-recommended moves as closely as possible rather than developing personal strategic philosophies. Opening sequences have converged across the professional field: the distinct opening styles that historically differentiated top players have largely been replaced by AI-recommended lines that most competitors have memorized. South Korean professional Ke Jie publicly lamented that watching the same opening sequences recycled endlessly has reduced his enjoyment of professional play.
The creative competition has migrated to the middle game, where AI analysis provides less deterministic guidance and raw calculation advantage is harder to express through pure pattern replication. The endgame remains relatively conventional.
On the access side, AI has democratized elite training methods that were previously gatekept by geography and connections to top coaches. More female professionals have climbed the rankings, a change attributed in part to AI tools providing high-quality training feedback regardless of institutional affiliation.
Why It Matters
Go is an unusually clean case study for AI's effect on creative human domains. The game has precise, measurable outcomes, a 2,500-year tradition of documented strategic philosophy, and a professional culture that explicitly theorizes about the relationship between moves and underlying principles. The changes observable in Go are likely to appear in other domains that combine structured expertise with creative judgment.
The pattern identified is specific: AI adoption collapses individual style variation in the structured, pattern-recognition-heavy phases of a domain while expertise concentrates in the phases where AI analysis provides less certainty. In Go that's the middle game. In knowledge work it is likely to be judgment calls that require integrating contextual information the AI doesn't have access to or can't weight appropriately.
For AI productivity tools specifically, the Go case offers a concrete picture of what "AI-assisted expertise" looks like at maturity: practitioners use AI to establish competent baselines efficiently, then differentiate through judgment applied in the uncertain middle phases where AI guidance runs out.
Our Take
The accessibility argument - that AI has helped players from smaller Go nations and improved pathways for female professionals - is worth taking seriously as a counterpoint to pure displacement narratives. High-quality training feedback that was previously exclusive to players with access to top Korean or Chinese coaches is now available to anyone with an internet connection. That democratization has observable effects.
The Ke Jie complaint about repetitive openings raises a structural question about what makes competition valuable. If the optimal opening is known and universally used, the opening phase is no longer a test of strategic originality - it's a recitation. Whether competition remains meaningful under those conditions depends on how much weight you place on the phases where AI analysis doesn't dominate. For Go specifically, the middle game is still genuinely contested. For other domains, the answer isn't yet clear.