The Evolution of Chess AI

Noctie represents a significant advance in chess AI technology. To better understand it, let's take a look at the history of chess AI.

Chess has long been considered the ultimate test for artificial intelligence. The game's simple rules but huge strategic depth makes it an ideal proving ground for AI..

Over time, chess AIs have evolved from simple rule-based systems to complex neural networks that can outplay the strongest human grandmasters. Let's take a look!

The Early Days

The first chess programs arrived in the 1950s, but were extremely limited.

These initial programs selected moves based on hard-coded rules and simple evaluation functions - basically checklists for assessing positions. While they could play legal chess, their abilities were rudimentary compared to any human player.

Brute Force vs. Human Play

In the 1960s and 70s, chess programs began employing more sophisticated searching techniques like minimax and alpha-beta pruning to look deeper into possible move sequences.

This brute force approach of calculating as many positions as possible became the dominant paradigm. Programs relied less on encoding chess knowledge and more on sheer computation - considering millions of positions per second.

The competing school of thought favored a more "humanlike" approach, using chess heuristics and pattern recognition to select candidate moves, which could then be search more selectively and deeply. It was argued this would lead to more positionally sound, insightful play. But the brute force searchers racked up better results.

Rise of the Super Computers

By the 1980s, chess engines running on supercomputers could play at master level. But there was still a big question - could a computer defeat the World Champion?

In 1989, World Champion Garry Kasparov easily beat the Deep Thought computer 2-0. But progress was accelerating. Purpose-built chess supercomputers like Intel's Deep Blue were on the horizon.

In 1997, after a pitched battle, Deep Blue famously prevailed over Kasparov 3.5-2.5, marking a historic milestone. For the first time, a computer had beaten the world's strongest human player under standard match conditions.

The Advent of Neural Networks

Entering the 21st century, top chess engines were leagues ahead of human players in terms of brute force calculation. But they still didn't "understand" chess in a human sense - they were just extremely fast at looking ahead.

That began to change with the rise of neural networks and machine learning. AlphaZero, published in December 2017 incorporated deep neural nets to both evaluate positions and select move candidates. It was soon replicated in the shape of the open-source Leela Chess Zero or Lc0.

Critically, these neural nets weren't pre-programmed by developers - they learned chess just by playing millions of games against themselves. The resulting play style felt uncannily humanlike.

Neural net engines started a fierce competition with calculation-based engines like Stockfish, taking turns winning computer competitions. Eventually, Stockfish relented and incorporated a simplified neural net based on Lc0 as its evaluation function, regaining the #1 spot.

Humanlike AIs for Training

While engines like AlphaZero brought a more humanlike positional understanding to chess AI, they were still focused on playing the objectively strongest moves.

For chess learners, playing against such an AI isn't necessarily productive. Even if you limit its strength, the AI makes very un-humanlike decisions - simplifying where a human would press, hanging pieces or playing odd moves that rely on tactical shots no human would consider.

In 2023, we built Noctie, a new chess AI that takes a fundamentally different approach.

Rather than aiming for maximum strength, our goal is to emulate human play at a wide range of skill levels. We want our AI to be a realistic sparring partner - one that makes the types of plans, decisions, and mistakes that a human of a given rating would make.

To achieve this, we trained our neural network on a vast database of over 1 billion human games. By studying this huge corpus of human decisions, our AI developed a deep understanding of human chess patterns.

It knows what types of moves players at different ratings tend to look for, what strategic ideas they favor, and what oversights they're prone to making. With this knowledge, Noctie can accurately reproduce human-style play from beginner to grandmaster level.

When set to a 1200 rating, it doesn't just play weaker moves - it plays the kind of moves an actual 1200 player would be attracted to. It falls for the same tricks, misses the same tactics, and struggles with the same endgames.

Conversely, when operating at GM level, it shows the refined positional judgment and calculation you'd expect from a strong human player.

Importantly, Noctie's human emulation extends beyond just move choices. It also matches the time management and general "rhythm" of human play at different levels. So you won't see it instantly playing book moves in the opening like a traditional chess engine. It takes a realistic amount of time to think in different stages of the game.

Noctie and the Future of Chess Learning

Tools like Noctie represent a new paradigm in chess AI - one centered on understanding and emulating human play at a granular level.

For learners, this opens up possibilities. Imagine having a tireless training partner that's always ready to play at your level. One that can offer insightful tips during your games, and provide bespoke puzzles based on your weaknesses.

Better yet, imagine playing games in the opening of your choice , with your opponent picking amongst the exact distribution of moves that your human opponents will!

That's the chess learning revolution that Noctie aims to kickstart. By harnessing cutting-edge AI to enhance the age-old tradition of chess improvement, we're working toward a future where every aspiring player has an ever-present coach and companion.

At Noctie, we're excited to be pioneering this approach - and we can't wait to see the new heights that chess enthusiasts everywhere will reach with it!