An image-based word prediction game has been developed by researchers by Department of Computational and Data Sciences at Indian Institute of Science. This computational model aims to play with the humans guessing the word for the picture drawn by the computer.
The fact that deep learning and immense progress in artificial intelligence has provided mankind with the tools to develop games which in some cases wins over human intelligence. Games like Double jeopardy, GO!, and online poker were already grabbing the attention of online gamers but now the introduction of AI based Pictionary game will fetch more online gaming audiences. IBM had developed the well-known chess playing game known as ‘Deep Blue’ in 1997, which defeated the then world champion Gary Kasparov, becoming the first machine to defeat a human.
This game is played based on guessing power of a human, preferably played in teams. Members of the team tries and guesses a word from the set of images drawn by another team member. As the game progresses the pictures get disclosed. Similarly, the Visual Question and Answering (VQA) technique is used in computers to create sense of its surroundings. In order to build the Pictionary game, VQA and games are merged together creating a model to play Pictionary.
A Pictionary style guessing job was created by the researchers known as Sketch Question Answering (Sketch-QA). In this style the humans guess the picture based on hand drawn images. The image is revealed one stroke at a time to the guessers, while prompting the users to guess every object after each stroke.
There are over 16624 guess sequences of 1108 subjects across 160 object categories, all included in a dataset named as GUESSWORD-160. Machine learning will be used to understand different decisions and collaborations made by humans gradually opening up the opportunity to imitate exactly like humans.
