I was chatting to my housemate yesterday and made a random connection between the implicit problem of evolutionary search algorithms and the story of the magic soup stone.
Iâ€™ll start with the story of the magic soup stone, in case youâ€™re not familiar. The story as I first heard it goes like this. A tramp comes across a house in the woods and having not eaten all day goes knocking at the door. The lady of the house expresses great sympathy but says she has no food either. The tramp considers this then produces from his pocket a smooth, ordinary-looking pebble. He tells the woman that this is a special stone that he keeps for emergencies which is capable of producing a very tasty soup all by itself. The woman is intrigued. She invites him in and puts a pot of water on to boil. The tramp adds the stone and starts stirring.
After a while he tastes it and declares it is coming along fine, but wonders if the lady has maybe a pinch of salt just to bring it to perfection. The woman obliges him and he continues stirring. After another short while he tastes it again. He claims it is some of the best soup the stone has ever produced. His only regret it that it lack texture. Perhaps the lady has a few cabbage leaves lying around. She does, they add and he keeps on solemnly stirring. Things go on like this with turnips, carrots and meat being added to the pot. Finally, the tramp announces the soup is ready and they serve up with a few loaves and some wine. The lady is so pleased with the result that she insists the tramp takes her bed while she sleeps downstairs. In the morning the tramp gives her the stone in return for her hospitality before setting on his way.
So, whatâ€™s this to do with Evolutionary Algorithms? Well, they are often touted as a good all-round search algorithm which can easily be slapped onto many optimisation problems without any special customisation. However, as the gleefully apt No Free Lunch theorem demonstrates, the more domain knowledge you can stir into the pot in terms of representation and search operators the better your search is going to work.
I like this analogy. It nicely encapsulates a point that often goes unacknowledged in EA research.