The Allen Institute for Artificial Intelligence (AI2) is proud to announce the world's first AI system that can play a Pictionary-style drawing and guessing game collaboratively with a human partner. Unlike automated players in board games like chess or Go, AI2's player communicates using pictures, phrases, and concepts to bring AI game-playing a step closer to the real world. For more information see the IDTechEx report on New Robotics and Drones 2018-2028.
AI2's "Iconary" game partners a human player with the AllenAI artificial intelligence in a Pictionary-like game where one player must use a limited set of icons to compose a picture representing a phrase, and the other player must attempt to guess the phrase. The AllenAI player combines advanced computer vision, language understanding, and common sense reasoning to make guesses based on the human player's drawing, as well as to produce its own complex scenes for the human player to guess. AllenAI has the capability to modify its drawing as well as its guesses in response to the human player's moves, guiding its partner towards the end goal. Iconary is the one of the first times an AI system is paired in a collaborative game with a human player instead of antagonistically working against them, as in many recent demonstrations of classic games played by an AI.
The phrases used in Iconary go beyond simple nouns—the players must select from a limited set of icons to convey scenes like "people jogging in the park" or "engineer riding a train", concepts that no single icon can represent. The limited vocabulary of icons forces players to create novel combinations of icons to communicate a phrase to their partner. The depiction below on the left shows a composition created by AllenAI to depict: "people jogging in the park", demonstrating its ability to depict park using elements such as trees and a swing set. The depiction on the right shows AllenAI's composition for the phrase: "pope waving his hand in the church", demonstrating its use of abstract symbols to represent motion (arrows).
AllenAI was trained on over 100k collaborative games between two human players, spanning over 75k unique phrases. The Iconary game tests the collaborative capabilities of AllenAI by having human players play together with it on new phrases the AI has never encountered before.
"Unlike other game-playing AIs like AlphaGo and DeepBlue, our AllenAI player is capable of both understanding and producing a nearly infinite combination of real-world scenarios" said Dr. Oren
Etzioni, CEO of AI2. "This is a compelling demonstration of machine common sense and the power of human-AI collaboration."
Source: Allen Institute for Artificial Intelligence
Top image: Pixabay
Learn more at the next leading event on the topic: Sensors Europe 2019 on 10 - 11 Apr 2019 at Estrel Convention Center, Berlin, Germany hosted by IDTechEx.