AI2 Creates World’s First AI to Play a Pictionary-like Game

Source: coffee/pixabay

There are many forms of smarts. You may know of people who are intelligent, or “book smart,” but show poor judgement. Common sense is the ability to make sensible decisions with sound judgement on practical matters that is common to nearly everyone. For example, when you see an empty restaurant among busy eateries bustling with patrons, common sense tells you not to eat there. Can a machine gain human-level common sense? Earlier this week, researchers at the Allen Institute for Artificial Intelligence (AI2) announced the world’s first artificial intelligence (AI) system that is able to collaboratively play an interactive online game inspired by Pictionary called “Iconary” with humans. Dr. Oren Etzioni, CEO of AI2, says it’s “a compelling demonstration of machine common sense.”

Computer algorithms are step-by-step procedures used to solve a problem. The human brain has an analogous algorithmic aspect. According to Keith Stanovich, Professor Emeritus of Applied Psychology and Human Development at the University of Toronto, and author of the book What Intelligence Tests Miss: The Psychology of Rational Thought, standard IQ exams measure the efficiency of the “algorithmic mind” that processes data versus the “reflective mind” that considers goals and beliefs systems. Common sense is distinctly different from intelligence, which is the capacity to comprehend, learn and reason.

The “smarts” of machine learning is the capability for computer algorithms to “learn” from data without explicit programming, and is measured by the accuracy of the results. For example, advances in speech/voice recognition, and image processing are due to improved pattern recognition by deep learning, a machine-learning method with more than two artificial neural network layers. Currently, computers can run algorithms efficiently, and are smart from this perspective. However, artificial intelligence woefully lacks common sense. The AI2 researchers believe that their AI system, AllenAI, addresses this gap.

AllenAI incorporates common sense reasoning, language comprehension, and computer vision to create drawings and make predictions on the human player’s drawing. AllenAI was trained using more than 75,000 unique phrases and over 100,000 collaborative games with human players.

Dr. Etzioni says, “Unlike other game-playing AI's like AlphaGo and DeepBlue, our AllenAI player is capable of both understanding and producing a nearly infinite combination of real-world scenarios.”

Iconary was created by AI2’s Perceptual Reasoning & Interaction Research (PRIOR) group—team members include Ani Kembhavi, Jordi Salvador, Dustin Schwenk, Eric Kolve, Alvaro Herrasti, Sachin Mehta, Sam Skjonsberg, Aaron Sarnat, Carissa Schoenick, Jonghyun Choi, Hannaneh Hajishirzi, and Ali Farhadi.

The game of Iconary is similar to Pictionary—both games use pictures to represent concepts and phrases for the players to guess. In Iconary, AllenAI is the online player that teams up with the human player. Phrases range from easy concepts such as “repairing a car,” “woman drinking juice,” and “smelling the coffee,” to more difficult ones like “releasing a bird in the desert,” “celebrating a win in the locker room,” and “helicopter landing on a ship in the ocean.”

Dr. Etzioni considers common sense and common knowledge as the “holy grail for the entire field of artificial intelligence.” With the release of Iconary powered by AllenAI, AI2 is advancing science and humanity one step closer toward artificial general intelligence.

"Common sense is the genius of humanity." —Johann Wolfgang von Goethe

Copyright © 2019 Cami Rosso All rights reserved.