Most of us probably spent a non-insignificant part of our childhoods battling against AI. 

Even in a pre-ChatGPT world, figuring out how to outsmart the computer has always been a core part of gaming, whether it’s learning all the ins and outs of their behavior or figuring out the ideal cheese strats. AI enemies can be punishingly difficult sometimes, but they’re also designed to always be surmountable. Maybe you’ve wondered what it would be like to play against more sophisticated versions of these enemies later down the line, ones whose behavior isn’t quite so straightforward.

As it turns out, beefing up AI players never really became a focus in the gaming industry. However, it did become a key point of research and improvement in the world of chess — one that has forever changed the way the game is played.

From the Turk to Deep Blue: a brief history of machine chess

The idea of having machines play chess has been a long-held fascination among masters of the game, one that goes all the way back to 1770, when The Mechanical Turk, Wolfgang von Kempelen’s fascinating invention, went on worldwide tours. It was a contraption that gave the illusion of a strong chess-playing machine to the delight and surprise of audiences – and the frustration of illustrious opponents like Benjamin Franklin and even Napoleon.

It was, of course, a fake. The ingeniously constructed device contained a small nook where a human chess master hid and controlled its every move. The world would have to wait another two hundred years for genuine chess computers to emerge. Still, the idea that you could construct a machine that would beat humans at their own games, or even play it in a way mere mortals would never be able to think of, remained a source of fascination throughout the ages. After all, the beating heart of the question has remained the same since: is chess solely about calculation and storing every possible (counter) move? Or is there something creative, more human to it?

Even the sixth ever world champion, Mikhail Botvinnik, had an interest in the subject. With his background in electrical engineering and computer science, he developed rudimentary algorithms and wrote two books on the subject, and followed the growth of the machines with interest after his competitive retirement in 1970. However, he passed away just a few years before the match that would change the landscape of chess forever: the battle between Deep Blue and world champion Garry Kasparov.

A changing of the guard: how humans were left behind

While the Deep Blue versus Kasparov rematch is the big event in the history of human-versus-computer chess battles, they’d been happening long before that fateful 1997 affair. You’d have to go back to 1956 to find the first instance of a computer defeating a human, though said human was an amateur.

Deep Thought, a precursor of Deep Blue, was the first machine to defeat a grandmaster in a tournament setting, taking down Bent Larsen in 1988. Eight years later, IBM’s updated version, Deep Blue, squared off with world champion Garry Kasparov. The lesser-known first match in 1996 ended in a 4-2 win for the Russian grandmaster; however, the 1997 rematch ended dramatically as the machine clinched the match win in the sixth and final game.

Looking back, it’s worth explaining just how notable and shocking this result was at the time. Kasparov was very much at the height of his powers, a monster at the chessboard – and for the general public, this was the standout match of its kind. Nowadays, the idea of machine learning is well-known, and ordinary people experience its effects every day. However, 28 years ago, it was still primarily confined to the realm of science fiction for the general public.

In the decade after the Deep Blue match, there was still some wiggle room in human-versus-computer chess matches. When playing against the machines, top grandmasters actively steered positions away from tactical and calculation-based complications – where humans were already comfortably outmatched – and instead aimed for long-term strategic advantages and less concrete benefits, which machines still tended to overlook. Over time, the “cheese” required to defeat chess computers became increasingly extreme. Then, eventually, it became irrelevant.

The 2005 edition of the Man vs Machine World Team Championships marked the end of an era, with Ruslan Ponomariov’s victory over Fritz marking the last human win over a chess computer under regular competitive conditions. Since then, matches with material handicaps have been the only way in which top grandmasters can remain even marginally competitive against chess engines, and even then, their prospects are dim.

So, man fought the machines and ultimately lost. And yet, chess is more popular than ever. It begs the question: why are we playing a game that machines have already solved? And what happens when machines train humans?

Tools-assisted speedruns on the chessboard

Using computer assistance for preparation quickly became standard practice among the chess elite once it became clear that the engines could comfortably match and even outplay top grandmasters. Still, there were some growing pains.

For an infamous example of what could go wrong, let’s go back to the Kramnik-Lékó classical world championship match of 2004. After seven draws, champion (Kramnik, playing White) and challenger (Lékó, playing Black) went down a sharp line in the Marshall Attack, both aided by extensive computer preparation – but only one of them could be right. 

After the position devolved into a spectacular queen sacrifice line, both players looked at a specific continuation on move 25 with the aid of their team and their computers – but Kramnik’s team didn’t leave the computer running long enough, and moved on with their analysis elsewhere, thinking the position was fine for them.

Lékó and his team evaluated this critical line longer and realized that it was a winning position for Black. With a combination of strong computer preparation and great calculation over the board, he scored a critical win in the match, and it took incredible heroics from Kramnik to ultimately defend his title.

This story is a good illustration of how computer-assisted chess preparation works. Chess openings are, in gaming terms, like a map veto process. Ultimately, the game will proceed along the lines of previous computer preparation and existing opening theory, and at some point, one player will run out of road before the other. 

So, what happens then? First of all, the player who is still in preparation, or “prep,” will be able to play their moves quickly since they require no calculation, just a memory check. Meanwhile, their opponent will have to proceed carefully, knowing full well that they’re still playing against something a supercomputer spewed out. At the very least, this will save time on their clock or lead to a significant disadvantage should they err in their manual calculation.

The Kramnik-Lékó game also serves as a great illustration of how deep chess preparation can go at the highest level of play. Even twenty years ago, with much more limited resources, the players went down a line whose first fifteen moves were first seen in 1947, with both players’ computer preparation going for a further ten moves before the fateful divergence. And that’s just one line in one opening! You can imagine how much further the preparation goes these days.

Turns out you don’t have to imagine it at all. 

A new metagame: competitive chess in the computer era

While Botvinnik first started his computer chess experiments with simple sorting algorithms, the consistent growth in computing power led to a focus on brute, raw calculation in the chess engine world for a considerable time. A huge paradigm shift occurred in late 2017 when Google’s AI-focused DeepMind turned its attention to the world of chess, introducing AlphaZero, a neural network-based chess engine that crushed its competitors at the time. 

Despite searching through just a fraction of the positions, DeepMind used an almost human-like approach to identify which positions warranted deeper analysis. It played an aggressive, dynamic style of chess that had not been previously seen, exploiting the sorts of long-term advantages and intangibles that were once seen as the last edge humans had over the machines.

In the wake of this revolution, the team behind popular chess engine Stockfish also incorporated a neural network approach, making it significantly stronger. Since Stockfish is an open-source project, you can load up a chess engine in your browser or on your phone that would easily crush any world champion in traditional play. All this makes for a whole new world of chess.

This democratization of chess means that even a club-level player can rattle off an opening sequence of perfect moves, as long as they properly learn and memorize the specific lines. (And that’s even before we consider the fact that you also have much easier access to your opponents’ games, meaning you can directly target their opening preparation.) Of course, unless this nets an immediate game-winning advantage, the level of play will then massively drop as they have to figure out their own moves on their own terms, but this doesn’t change the fact that you have to survive your way out of the opening.

On the super-grandmaster level, rented cloud computers and a dedicated team of seconds push the preparation into unfathomable territories. Metagaming and trying to find odd ways to gain advantages runs rampant – like purposefully playing the third-best computer choice 20 moves down the line, still maintaining a playable position, and snatching away the advantage gained by the opponent’s preparation.

The prevalence of chess engines also completely destroyed the concept of an “opening repertoire,” or specialization, at the highest levels of play. Previous generations could afford to exclusively devote their study to specific openings and lines – again, think of map bans and map pools for your favorite shooter – but the sheer breadth and depth of engine-assisted preparation these days makes this impossible. Now, you need to know everything. Preparation of this level – and the necessity of consistently revisiting and revising your lines – is an incredible commitment.

And that’s just the opening. At the other end of the chess game – namely, the end – we’ve long had tablebases, where computers precalculate and fully map out all positions with a specific number of pieces. By 2012, we had a near-complete tablebase of all chess positions with seven or fewer pieces. An eighth piece adds so much complexity, however, that such a tablebase still has not been completed. It goes to show the challenge of fully calculating the possibilities of a chess game, and it serves as a reminder that the human element still plays a key part in the proceedings.

The ease of access to a near-objective evaluation of any given chess position has also completely upended the way chess games are broadcast and how commentary works. Almost all tournament streams now include the “eval bar” on the side, a visual tool representing the current evaluation of the position. Anytime a player makes an incorrect move, it shifts – prompting an immediate reaction from fans and commentators alike. 

Many modern chess pundits see this new development as a matter of accessibility. As Levy “GothamChess” Rozman put it in a 2024 tweet in support of the eval bar: “You should be able to watch chess at a loud bar and know ‘the score.’ You can watch higher-level coverage without it, but that should not be the norm.” Grandmaster Jon Ludvig Hammer voiced similar opinions, arguing that the eval bar “lets commentators focus energy on phrasing and explaining rather than calculating,” and that “it builds storylines when there's only one good move.” 

However, the debate as to whether the eval bar is a help or a hindrance is still ongoing. When you’re watching your favorite chess streamer, it’s not like you can see the standings or the computer evaluation in real time. Just like with any other stream, you watch the high-level game unfold from one person’s perspective, hoping to see good strats, superior moves, and ultimately, a win. This dynamic is part of why some still resist the non-stop inclusion of an eval bar on the live broadcast.

In the end, it’s not about forgoing or solely relying on the eval bar. It may be helpful, but it’s still just a tool. Clear, good commentary, and a nuanced understanding of high-level chess are key when it comes to a good chess broadcast in the computer era. It is still all about the humans, after all.

The human element, revisited

Now, looking back at old writings on the nature of chess, it’s almost quaint to see pros and philosophers insisting that there is an artistry to the royal game, an inherent element of imagination and humanity that means no machine will ever surpass us at it. 

And yet, there’s still a grain of truth to it. Your computer may be able to devise the perfect move for any situation, but multiple elements of competitive chess remain unmistakably human. When the time pressure is on and physical exhaustion is setting in, it’s hard to think like a computer – that’s what makes competitive chess so exciting to watch.

In the final round of Norway Chess 2025, deep into the match, with everything on the line and just a few seconds left on the clock, Team Liquid’s Fabiano Caruana couldn’t quite find the winning continuation against the reigning world champion, Gukesh Dommaraju. Having a bit more time to calculate than his opponent, who only had a few seconds and the small move-by-move increment to work with, he opted for 47. f4, a surprising pawn thrust that the computer immediately notated with two question marks. On paper, he’d thrown away the win. But Caruana knew full well what he was doing. The unexpected play flustered his opponent, clinching the win for him. 

“I got down to one minute and thought, ‘I’m not going to think anymore,’” he explained afterwards on the C-Squared podcast. “I played the move knowing that, very likely, it is a draw in some way, but also thinking he has five seconds, and it’s an unexpected move. It was a cheap trap […] but if he had 30 seconds, he wouldn’t fall into it.”

It doesn’t matter that computers will always be able to outthink us. If that were the only thing that mattered, competitive chess wouldn’t exist at all. On the biggest stages in the world, chess is the same as it has always been: two humans locked in head-to-head combat, trying to get the better of one another, trying to find their way out of a deep, dark forest. Regardless of what computers can tell you, that will never change.

‍