The World's Most Advanced Automatic Chess Problem Composer

Development of the computer program, Chesthetica, started in 2006 as part of my doctoral research in computational aesthetics, a subfield of artificial intelligence (AI). It incorporated a computational aesthetics model to evaluate aesthetics or beauty in chess move sequences in a way that correlated positively with human assessment. This means that many thousands of chess problems or puzzles could be ranked automatically in terms of typically perceived aesthetics but in a very short period of time compared to humans doing the work. After completing my PhD in 2009, Chesthetica was further developed so it could also compose original chess problems, but this work was just preliminary and only a few were generated in 2010 (posted online a few years later).

While computers could already beat world chess champions, they could not compose even functional chess problems entirely on their own. This piqued my interest in computational creativity; also a subfield of AI. In particular, to a new approach I called, the 'Digital Synaptic Neural Substrate' (DSNS). Named so because it uses fragments of information from other domains (e.g., photos, paintings, music, regular chess game sequences) and combines them into 'strings' (a standard 'substrate') that allows recombination of information from seemingly unrelated domains. This is analogous to how a human brain (synaptic connections between the neurons etc.) might get inspired from being exposed to various creative objects as stimuli. Despite having the word 'neural' in it, the DSNS has nothing to do with machine or deep learning (or artificial neural networks); this is a common misconception.

Since 2014, Chesthetica has been using this approach to compose thousands of aesthetically-pleasing original chess problems entirely on its own with no limit in sight. In effect, it is doing the equivalent of thousands of hours of human creative labor. The aesthetics model is used optionally as a filter whereas the DSNS is what serves as the 'spark' of creativity. It remains to be tested if the DSNS could also be as effective in other domains where creativity matters. In 2020, a new type of machine learning (which also does not use artificial neural networks) was implemented to automatically select generated compositions based on personal user preference. These puzzles can be used by human players and even composers to train, to learn from, or purely for entertainment. They also likely help keep your brain healthy and mind sharp. You can find more resources on these topics here and the links below.


The Digital Synaptic Neural Substrate: A New Approach to Computational CreativityChesthetica's Book of Chess Constructs, Volume 1Chesthetica's Book of Chess Constructs, Volume 2
Chesthetica's Book of Chess Constructs, Volume 3Chesthetica's Book of Chess Constructs, Volume 4A Novel Machine Learning Method for Preference Identification: A Chess Case StudyChesthetica's Book of Chess Constructs, Volume 5


Amazon Facebook Instagram Minds Reddit
Steemit Twitter WordPress YouTube 1 YouTube 2

WordPress Feed


Medium, 29 October 2021: Is This the Most Complex Forced Mate Chess Problem Ever?
ChessBase News, 26 December 2020: Chess is Beauty, Can You See?
ChessBase News, 10 February 2020: What Do People Really Find Attractive in Chess Problems?
Medium, 6 November 2019: What Makes a Good Chess Construct?
ChessBase News, 22 March 2018: Challenging Studies, Automatically
ChessBase News, 2 October 2017: Four Knights vs Queen Challenge - Part 2
ChessBase News, 8 September 2017: Four Knights vs Queen Challenge - Part 1
ChessBase News, 7 April 2017: Chesthetica Composes Custom Mates
ChessBase News, 28 August 2016: Machine Creativity: What it is and What it Isn't
ChessBase News, 16 June 2016: Can Computers Compose Artistic Problems? - Part 2
ChessBase News, 14 June 2016: Can Computers Compose Artistic Problems? - Part 1
ChessBase News, 31 May 2016: Cheshetica: Studies and a Decade in Development
ChessBase News, 7 Sep 2015: Cheshetica Composes Longer Mates!
ChessBase News, 31 May 2015: Celebrating 300 Machine Generated Problems
ChessBase News, 6 Feb 2015: Computer Generated Chess Problems for Everyone
ChessBase News, 7 Nov 2014: A Machine that Composes Chess Problems
ChessBase News, 15 Dec 2012: A Computer Program to Identify Beauty in Problems and Studies


Bitcoin: 1MRArfVin3C921jy2vsv6irBbi6rzjWXMK
Ethereum: 0x26478f6444b2428569d2fbb23185238e8C671146

Contact: ayzed at yahoo dot com OR dr.azlan.iqbal at gmail dot com

Copyright 2006-Present Azlan Iqbal. All rights reserved.