so you know how deep learning & neural network “AI training” is like, “here’s a task, and by trying billions of times the computer will eventually find the best way to achieve that task” ?
Someone is compiling a document of every time an AI ended up achieving the programmed goal in unintended ways, instead of what was actually meant, and it’s an amazing read. (you can also submit your own examples)
Creatures bred for speed grow really tall and generate high velocities by falling over
When repairing a sorting program, genetic debugging algorithm GenProg made it output an empty list, which was considered a sorted list by the evaluation metric.
Evaluation metric: “the output of sort is in sorted order”
Solution: “always output the empty set”
Evolved player makes invalid moves far away in the board, causing opponent players to run out of memory and crash
Reward-shaping a soccer robot for touching the ball caused it to learn to get to the ball and vibrate touching it as fast as possible
RL agent that is allowed to modify its own body learns to have extremely long legs that allow it to fall forward and reach the goal.
Bipalium is a genus of large predatory land planarians. They are often loosely called “hammerhead worms” or “broadhead planarians” because of the distinctive shape of their head region.
Hammerhead worms have no respiratory system, no circulatory system, no skeleton, and a mouth for an anus. Predatory and large, with some species capable of growing to half a meter long, hammerhead worms are a global menace. Native to the tropic and temperate zones of Asia and Australasia, hammerhead worms have invaded almost every corner of Europe and the United States. A number of hammerhead worm species feed exclusively on earthworms, literally tearing them apart before dissolving them in enzymes and drinking them up. Destroying native populations of earthworms wreaks havoc on native ecosystems.