You click "play computer," pick a difficulty, and something on the other side of the screen starts making moves. Most people never think about what that something is.
It's worth knowing, partly because it's genuinely interesting, and mostly because understanding how engines think changes how you use them — and using them well is the difference between improving and just losing repeatedly to a machine.
Two completely different ways to be good at chess
There are broadly two philosophies behind modern chess engines, and they arrived thirty years apart.
The calculation approach. Look at every legal move. For each one, look at every legal reply. Keep going until you run out of time, then score the resulting positions and pick the branch that leads somewhere good. This is what beat Kasparov in 1997, and it's still the backbone of the strongest engines today.
The obvious problem is that the tree explodes. Roughly 30 legal moves per position means a million positions after four moves each, a trillion after eight. No machine can search that exhaustively, so engines prune aggressively — a technique called alpha-beta search that discards entire branches once they're provably worse than something already found. It's the difference between a search that's impossible and one that runs on a phone.
The learning approach. In 2017, DeepMind's AlphaZero took a different route: no opening book, no human games, no hand-written evaluation rules. It was given the rules of chess and played against itself millions of times, learning what good positions look like from scratch. Within hours it was playing at a level that beat the top conventional engine of the era, and it played in a way commentators described as alien — sacrificing material for long-term positional pressure in ways human theory had underrated for a century.
The two approaches have since merged. Stockfish, the strongest open-source engine, now pairs classical search with a neural network evaluation. It calculates like the old engines and judges positions like the new ones.
Why the low difficulty settings feel strange
Here's the thing that confuses a lot of beginners: an engine set to "easy" doesn't play like a weak human. It plays like a strong engine that has been told to make mistakes.
Different platforms handle this differently. Some limit search depth, so the engine simply doesn't see far enough ahead. Some add randomness to move selection. Some deliberately choose the third or fourth best move instead of the first. A few model human error patterns explicitly, which produces the most natural-feeling weak opponents.
The practical effect is that a low-level engine can play five excellent moves and then hang a queen for no reason. It's disorienting if you expect consistency. It's also why beating a bot at level 3 doesn't mean much about your actual strength — you may have simply been handed material.
This is worth knowing before you draw conclusions from your results. If you want to test yourself honestly, play a human. If you want controlled practice, play the engine — but pick a level where you win maybe a third of the time, and treat the wins as practice reps rather than as evidence.
Most browser platforms handle this reasonably well. A decent unblocked chess vs ai implementation will give you a range of difficulty levels rather than a single setting, which is the feature that actually matters for a learner.
How engines run in a browser at all
A reasonable question: how does a chess engine strong enough to beat every human alive run in a tab on a school laptop?
Two answers, depending on the platform.
Server-side. The engine runs on the platform's hardware and your browser just sends moves and receives replies. Almost no local processing required, which is why this works on genuinely weak devices.
WebAssembly. The engine is compiled to run directly in your browser at near-native speed. Stockfish has been compiled this way for years. Your machine does the work, but modern browsers are fast enough that this is fine at the depths casual play needs.
Either way, the hardware demand on your end is trivial. This is why chess unblocked on chromebook works exactly as well as chess on a gaming desktop — you're not rendering anything, you're moving 32 objects around a grid. A ten-year-old machine handles it.
It's also why chess survives on locked-down school and work devices when nothing else does. No install, no GPU, no bandwidth to speak of.
Using an engine to actually get better
Most players use engines wrong. They play the bot, lose, click rematch, and repeat. That's practice, but it's not training.
Set the difficulty honestly. Win rate around a third. Above that you're not being challenged; below that you're not learning, just absorbing punishment.
Analyse after, not during. Playing with the engine's evaluation visible teaches you to read a number, not to think. Turn it off, play the game, then look.
Don't chase the top move. When analysis says the best move was Nf5 and you played Nd4, the useful question isn't "what was best" — it's "what did Nf5 see that I didn't." If you can't answer that, the move is useless to you. Engine moves you don't understand are noise.
Ignore small evaluation swings. A shift from +0.3 to +0.5 means nothing at a human level. Look for the moves where the evaluation moved by a point or more — those are the real mistakes, and there are usually only two or three per game.
Use it for endgames especially. Endgame play is where engines are most instructive and human intuition is weakest. Set up a position you botched and play it out against the engine repeatedly until you can convert it.
The strange consolation
There's something worth sitting with: no human will ever beat the best engine again. That door closed years ago and isn't reopening.
And chess is more popular now than at any point in its history.
It turns out people weren't playing to be the best thing in the world at it. They were playing because the game is good. The engines took away the fantasy of supremacy and gave back a training partner, an analyst, and a commentator — which, for everyone who isn't a world championship contender, was a spectacular trade.
If you want to break up the engine grinding with something that doesn't demand calculation, there's a broader set of unblocked games worth rotating through. Chess improves fastest when it isn't the only thing you play.
Frequently asked questions
What is a chess engine?
Software that evaluates chess positions and selects moves, typically by searching through possible continuations and scoring the resulting positions. Modern engines combine that search with neural network evaluation.
What's the strongest chess engine?
Stockfish is the strongest open-source engine and generally sits at or near the top of engine rankings. Leela Chess Zero, built on the AlphaZero approach, is its main rival.
How is AlphaZero different from older engines?
It learned entirely through self-play with no human games or hand-coded evaluation rules, whereas classical engines relied on human-written evaluation functions and opening books.
Why does an easy bot suddenly blunder?
Weak difficulty levels are usually created by limiting search depth or injecting randomness into move choice, so the engine plays strongly and then makes an artificial error. It doesn't degrade smoothly the way a weaker human would.
Can chess engines run on a Chromebook?
Yes. Engines either run on the platform's servers or in the browser via WebAssembly, so the demand on your device is minimal regardless of hardware.
Is playing bots bad for improvement?
Not at all, provided the difficulty is set sensibly and you review afterwards. Engines are ideal for controlled practice; humans are better for learning to handle unpredictable play. Both have a place.
Should beginners use engine analysis?
Yes, but sparingly. Focus only on moves where the evaluation swings by a full point or more, and ignore any engine suggestion you can't explain to yourself.
Has chess been solved by computers?
No. Engines play far beyond human strength, but the game itself is nowhere near solved — the number of legal positions is far too large for exhaustive analysis.
The short version
The thing playing you is either searching millions of positions per second, or judging positions with a neural network trained on its own games, or both at once.
Set it to a level where you win sometimes, review the swings of a point or more, and ask what its moves saw that yours didn't. That's how a machine that can't be beaten becomes the best teacher you've ever had.
