Scientist and "AI Weirdness" blog creator, Janelle Shane shares how the field of Artificial Intelligence developing and how is making our world a weirder place in her book "You Look Like a Thing and I Love You".
#NotAsSmartAsYouThink
Today's Artificial Intelligence is not like what you see in the movies. AI has to be taught by providing it with datasets and then given a goal to complete using what it learned from that data. This means sometimes it doesn't give a result like what you would expect. As we learn more about the human mind we come to realize that AI is less capable and less powerful than what a human can do. AI works best if you give it one simple task and provide a large diverse set of information to train it on, while humans can problem solve and process what they are experiencing. Janelle mentions that an AI has the brainpower of a worm. Sometimes what goes wrong with AI isn't its fault but rather human error.
#BecarefulWhatYouAskFor
Providing an AI with too general of a goal can get unexpected results since AI will do its best to complete what you told it to do while using the least amount of work. An example Janelle shares is how an AI was told to play Tetris and not lose. The AI let the pieces fall and then paused the game indefinitely when they got close to the top, therefore not losing. AI can also be led astray by human bias or even unexpected details that are in the data that was given to the AI. The AI doesn't know it did something wrong it is just working with what was given like one AI that got really good at identifying rulers since they were often in the background of photos from its dataset.
See the video below for a Ted Talk by Janelle Shane about Artificial Intelligence. Her book is also available in audiobook format from Ohio Digital Library/Libby.
