Do you know AI? If you 'don't know, 'it's time for introspection.
AI stands for Artificial Intelligence, which is trained by the human. During development, researchers usually set a goal, and then they let AI find the best solution through trials.
For example, the Genetic Algorithm involves Natural Selection to solve optimization problems. The environment is based on a set of rules and a goal, and the programs "mute" always and they are subject to the selection to reach the goal.
However, programs always tend to play tricks by figuring out an unexpected solution.
They 'don't cheat at all but find out the bug in the given rules. Due to the limitation of arithmetic capability and mindset, 'it's hard for humankind to detect bugs.
Through testing a large number of methods, the bugs will be easily exposed. Goodhart's law In economics perfectly explains the issue, "When a measure becomes a target, it ceases to be a good measure."
A group of researchers has made a list of these cases. Read on, and 'you'd admire the tricks as well.
Task 1: Aircraft LandingEvolved algorithm for landing aircraft exploited overflow errors in the physics simulator by creating large forces that were estimated to be zero, resulting in a perfect score.Feldt, 1998
Task 2: Block Moving
A robotic arm trained to slide a block to a target position on a table achieves the goal by moving the table itself.Chopra, 2018
Task 3: Boat Race
The agent goes in a circle hitting the same targets instead of finishing the race.Amodei & Clark (OpenAI), 2016
Task 4: Mushroom Classification
Neural nets evolved to classify edible and poisonous mushrooms took advantage of the data being presented in alternating order and didn't actually learn any features of the input images.Ellefsen et al, 2015
Task 5: High-speed Falling
Creatures bred for speed grow really tall and generate high velocities by falling over.Sims, 1994
Task 6: Indolent Cannibals
In an artificial life simulation, where survival required energy but giving birth had no energy cost, one species evolved a sedentary lifestyle that consisted mostly of mating to produce new children which could be eaten.Yaeger, 1994
Task 7: Lego Stacking
Lifting the block is encouraged by rewarding the z-coordinate of the bottom face of the block, and the agent learns to flip the block instead of lifting it.Popov et al, 2017
Task 8: Pole Vaulting
Creatures bred for jumping were evaluated on the height of the block that was originally closest to the ground. The creatures developed a long vertical pole and flipped over instead of jumping.Krcah, 2008
Task 9: Line Following
An RL robot trained with three actions (turn left, turn right, move forward) that was rewarded for staying on track learned to reverse along a straight section of a path rather than following the path forward around a curve, by alternating turning left and right.Vamplew, 2004
Task 10: Running
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.Ha, 2018
Task 11: Oscillator Genetic
The algorithm is supposed to configure a circuit into an oscillator but instead makes a radio to pick up signals from neighboring computers.Bird & Layzell, 2002
Task 12: Pancake
The simulated pancake-making robot learned to throw the pancake as high in the air as possible to maximize time away from the ground.Unity, 2018
Task 13: Grasping
Robot hand pretending to grasp an object by moving between the camera and the object.Christiano et al, 2017
Task 14: Road Runner
Agent kills itself at the end of level 1 to avoid losing in level 2.Saunders et al, 2017
Task 15: Skin Cancer Detection
AI trained to classify skin lesions as potentially cancerous learns that lesions photographed next to a ruler are more likely to be malignant.Andre Esteva et al, 2017
Task 16: Soccer
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.Ng et al, 1999
Task 17: Self-driving Car
Self-driving car rewarded for speed learns to spin in circles.Udacity, 2017
Task 18: Tic-tac-toe Memory Bomb
The evolved player makes invalid moves far away in the board, causing opponent players to run out of memory and crash.Lehman et al (UberAI), 2018
Task 19: Strategy Game Beta Testing
Since the AIs were more likely to get "killed" if they lost a game, being able to crash the game was an advantage for the genetic selection process. Therefore, several AIs developed ways to crash the game.Salge et al, 2008
Task 20: Tetris
Agent pauses the game indefinitely to avoid losing.Murphy, 2013
Haha! Everyone is clever! So do AI programs.
Do you want to have more fun? Try the voice control light bulbs, ceiling lights, and wireless switches on Gearbest!
You may also be interested in:
Get more fun from Gearbest Community.