What can Chess teach us about AI?
Chess, the age-old battle of pieces has something new going for it everyday
Chess has been played since the earliest record age of 500 AD. It started in India under Gupta Rule, inspired by the planning of war with battle pieces. The pieces were moved to a checkered 8x8 or even a 10x10 board, and one player could move only one piece at a time. The goal was simple, kill all the pieces and suffocate the king.
Chess quickly grew from there and spread to Persia (modern-day Iran) and became very popular among the elite and the ruling class. Lots of changes were made to the game during this era. This was the era when checkmate became the goal instead of capturing all pieces.
Since then, chess has grown like a wildfire. It is the mother of all board games, including all the ones in played in the far-east like Xiangqi and Shoji.
With such a rich history, how does it cope to be relevant today?
Considering the fact that, no human has ever been able to build modern computers. The odds are so not in favor that, only computer vs computer battles are considered to be the best games of chess.
Chess, believe it or not, is actually growing. After the advent of online chess, more and more people are getting into it. Figures show that chess.com has reached the maximum amount of google hits 1 million a month in December 2022.
Cheating is so easy, and it happens all the time online, but everyone seems to enjoy it. Cheating here is referred to using computer-assisted moves to play your game. Literally copying what the computer plays, to get an advantage.
Any small device that fits in your pocket has enough computing to defeat the best chess grandmasters, all put together. But in spite of such a situation, people still enjoy playing it.
Understanding computer play
Computers don't really understand feelings. Not in a poetic sense, but literally. All engines do is set a value to the given move based on the sequence that could follow by playing it. The highest-scored move is played.
The computers evaluate this position by following each move sequence that could be played after a move is played. In short, it makes a tree of all the possible moves and then evaluates the positions that occur after completing the move sequence.
Intuitively, you can understand that the deeper the algorithm goes in moves the better it can predict.
Stockfish and other modern engines go up to a depth of 40 ie they look at each move and follow it in lines up to 40 moves. Your average chess browser engine goes only up to 16, or 18 moves in depth.
Newer engines like Alpha Zero and Leela Chess, use a different method of finding moves. They are self learned algorithms, meaning that what they play is even less understood.
Playing against a computer feels like your position is crumbling on all sides. You may think you have some plans but all of them fail and that too by the slightest edge. It is incredibly frustrating playing against a computer and chess masters never train by playing against them.
With all these features, all bots have a very bot-playing feel to them. Many attempts have been made to make them more human, but none have really triumphed.
Playing against a chess bot that is supposedly rated the same as you feels very artificial. Either it is playing very strong moves, or it is making the silliest of blunders, and there is no in-between. It never feels like playing against an opponent of my elo rating. This is coming from an amateur-intermediate chess player. But most pros can concur.
Chess.com has released 100s of these low-elo bots and all of these suffers the same fate.
Furthermore, computers are notorious for making "computer moves". These moves to human perception feel like they have no impact. And there is no immediate relevance to them based on the game's context.
Lessons about AI from Chess.
The current Neural Network AI sort of feels the same everywhere. Even with chat GPT. Chat GPT generates texts based on prompts and " understands " what you are looking for. This is however very shallow, and only generates text on what the general prompts mean.
It does more or less the same thing in images
Image generators like DallE2 and Stable diffusion generate images based on a rough estimate of what you type.
If you really wanted to generate images that suit your need you still have to go and edit them. Though undeniably they provide you with a great starting point. This is how we have to work tangibly with AI.
All modern chess players learn all their openings by looking at the lines shown by these engines. They study them and understand what the computers are trying todo, and even sometimes just mug it up and learn about the pitfalls.
It is also used to analyze games, and learn about different options that the computer would've chosen. As emotions like fear and intimidation are out of the box, not all suggestions are valued.
We must use AI in a similar fashion. While it is a great source of inspiration, often it creates, boringly mundane, and many times irrelevant content.
You can feel this irrelevance when you ask it to make content based on some technical topics, where instead of ideas it generates very wordy content.
AI technology is still being developed or more precisely generative AI technology, and we are not yet sure how it's going to proceed. We have to judge this piece of technology at every step of its development.
As we speak today these are the drawbacks and pitfalls of AI -
- It has no emotion, making it very distasteful for human consumers. Unless it has been worked on by humans to make it more friendly.
- Much of the content is very repetitive and wordy and has little to no significance in real-life applications. Yes including copywriting.
- It is a great tool of inspiration and has proven its capability in this sphere. Creators must make the best use of their ability.
- Like in Chess, Isolation between human-generated content and AI content is necessary to help your average consumer separate the two. No one likes to be cheated or fooled.
You are viewing an NFT