Fritz 20: When a Chess Engine Learns to Be a Teacher

At a chess club near the Han River in Da Nang, I sat behind an eleven-year-old boy and watched him move the mouse. He dragged a knight back to its old square, ran the analysis, saw the evaluation bar slide from +0.3 to -0.9, then put the mouse down and asked the boy beside him about homework. No anger, no sigh, no second look. Only a small red question mark beside the move, and a child deciding that the answer was not meant for him.

I thought about that question mark all afternoon, and for many days afterwards. It is the product of three decades of computer chess industry, and also that industry's confession of failure. The machine spoke. The human did not understand. The space between those two sentences is where Fritz 20 has just been placed, with the publisher's own introduction from ChessBase: a personal trainer, a toughest opponent, a strongest ally.

According to ChessBase's product description, Fritz 20 targets both players taking their first serious steps in chess training and players already competing at tournament level. The promise sits in three adjectives: more efficiently, more intelligently, more individually. I read those three words and asked myself why they appear now rather than ten years ago, when chess software was still a luxury inside national team training rooms.

The answer is historical. Fritz has been present in chess since the mid-1990s, when a program developed by Frans Morsch and Mathias Feist, running on a personal computer, won the World Computer Chess Championship in Hong Kong in 2026, ahead of systems powered by supercomputers. Thirty years later people still tell that story like a fairy tale about the underdog. In October 2026, in Bahrain, Deep Fritz drew 4-4 with Vladimir Kramnik in the match known as Brains in Bahrain. In November 2026, in New York, X3D Fritz drew 3-3 with Garry Kasparov. In December 2026, in Bonn, Deep Fritz beat Vladimir Kramnik 4-2.

Those matches belong to another era, when machine and human were staged facing each other like two fighters. Sports culture is a graveyard of names, kept alive so their stories can play one more round. Fritz, Junior, Hiarcs, Rybka, then Stockfish and Leela Chess Zero: every name was once an empire, and every empire was replaced by something simpler, cheaper, stronger.

The turning point came with neural networks. Since engines using neural networks and NNUE techniques took over the rating lists, top-level chess strength has become a free commodity. Stockfish is open source. Leela Chess Zero is a community project. A mid-range phone in Da Nang today holds a player stronger than every world champion combined. If a program's value lay in its strength, that market would have died long ago.

Fritz 20 survives because it sells something else. The engine race has shifted from which machine calculates deeper to which machine teaches more clearly, and Fritz 20 is built around the second question. This matters because most Vietnamese players still judge software by a single measure: analysis depth.

A program like Fritz 20 can offer three tiers of training, and they differ enormously in difficulty. The first tier is the sparring partner: a machine that can lower or raise its strength, play at human level, even imitate a specific player. The second tier is diagnosis: it flags mistakes, classifies blunders, measures positions. The third tier is curriculum: it knows where a student is weak, why, and what exercise comes next. The first tier is so easy that every free app has it. The second is the standard of the age. Only the third deserves the word coach.

And the third tier is the hardest, because it demands that software understand something it does not possess: human slowness.

Consider the boy at the club again. When the machine said -0.9, it delivered a conclusion calculated from a thirty-move chain of assumptions. The boy needed a reason and received a verdict. The distance between a correct move and a human who understands why it is correct is the entire frontier of machine-based chess teaching. A human coach would reset the pieces, recall a similar old game, point at the pawn structure and say: your knight move was not bad, it simply arrived too early. A machine cannot say that unless someone programs it to know that explanation sometimes matters more than evaluation.

Based on my experience watching games at grassroots and youth chess events, most players use engines as referees rather than teachers. They play, switch on analysis, see where they went wrong, nod, and start a new game. That loop runs thousands of times and leaves very little behind. It resembles running the same flat route every day: there is sweat, there is a feeling of having trained, but there is no progress.

I have kept two notebooks since my early years in the trade, one for data and one for feeling. The data notebook records names, Elo, openings, results. The other records what cannot be measured: a young player rubbing his eyes after being pulled back to equality, an old man sitting alone after the clock ran out, the sound of pieces on a wooden board in a room so quiet you could hear breathing. The first time I sat in front of a microphone I mispronounced three names and was told I sounded like someone reading a newspaper aloud. Mistakes in front of a microphone taught me that perfection is only a draft of sincerity. The same lesson applies to software: a good training program must accept that learners need emotion to remember, and need to be caught at the exact moment they are lost.

In Vietnam this story has an extra layer. Vietnamese chess has Le Quang Liem, world blitz champion in 2026 in Khanty-Mansiysk, Nguyen Ngoc Truong Son with his SEA Games medals, Pham Le Thao Nguyen among the region's leading women. Beneath that summit lies a thin ecosystem: a few large centres in Hanoi, Ho Chi Minh City and Da Nang, some grassroots clubs, and many families improvising on their own. In such an ecosystem, software costing a few dozen euros equals half a month's tuition at a chess class. It can be a bargain, and it can also be an expensive way to convince yourself your child is being trained professionally.

In Da Nang I know a coach who teaches more than forty students a week. He copies every student's game into a notebook by hand, marking three kinds of error in red pencil: calculation, planning, time management. Software that could do that job would save him dozens of hours a month, and those hours could go into what machines cannot do: sitting beside a child who is losing composure. I still advise three things to anyone starting to learn with an engine. Analyse only your own games, because analysing other people's games is entertainment. Write down your own reasoning before switching the machine on, because if you switch it on first there is nothing left to compare. And keep one notebook without a machine, so that what belongs to human beings stays with human beings.

I still like walking past the chess boards set up on Saigon pavements and the xiangqi boards under trees in Da Nang parks. There is no evaluation bar there, no red question mark. Only a loser who must stand up and give away the seat, and a winner who must remember how he just won. The silence of those boards teaches what software cannot: how to endure an opponent's gaze, how to hold back an impulsive move because you know your opponent is waiting for exactly that move.

What I want to say against the wind concerns a collective belief. For thirty years the chess world believed stronger tools would produce stronger players. History does not confirm that belief. The number of grandmasters has grown, but an individual's rate of progress is not proportional to the power of the tool in their hands. The bottleneck of modern chess is not the depth of the machine but the attention span of the human. A child can sit in front of the strongest Stockfish for ten hours without gaining a single Elo point, if during those ten hours no idea genuinely convinces him.

The second blind spot lies in the most advertised feature: personalisation. A program that always adjusts difficulty to the learner keeps the learner around longer, but also keeps the learner inside a comfort zone. Progress in chess comes from games in which you are suffocated, misunderstood, punished for one small slip. A machine can create pressure, but that pressure is sterile. There is no one to be embarrassed in front of, no one to apologise to, and therefore nothing to remember.

The third blind spot is professional. Chess increasingly resembles a sport of contracts, sponsors and protected images. Young players are packaged as brands before they fully understand a king-and-pawn endgame. They learn how to answer interviews before they learn how to speak honestly about their own games. In that environment, software becomes a kind of uniform: it creates a feeling of professionalism, while what gets neglected are very basic skills such as endgame calculation, passive defence, or simply the patience to sit still.

Here is a comparison I keep using. People worship a goalkeeper's distribution, to the point of forgetting that what sustains a goalkeeper's career is basic reflex. Chess is the same: people worship engines, to the point of forgetting that what sustains a player is the ability to calculate without a hint.

So where can Fritz 20 genuinely help a Vietnamese learner? I think it lies in the program admitting something the mighty engines do not: strength is not the same as understanding. If the software agrees to play the teacher, to speak slowly, to ask questions, it fills the biggest gap in learning chess alone. If it merely becomes stronger, it repeats the exact mistake the chess world has made for thirty years.

Sport does not promise victory; it only promises a pulse willing to continue. In chess, that pulse is the moment a learner discovers why his own move was wrong, before the machine says it. A successful training program is not one that makes the learner dependent on it, but one that gradually makes the learner no longer need it.

I returned to the club near the Han River on another evening. The eleven-year-old was still there, still moving the mouse. But this time, when the red question mark appeared, he did not turn away. He dragged the timeline back, put both hands on the table, and began telling the boy beside him about his knight, slowly, clumsily, like a person teaching himself.