Pictionary may provide AI with common sense

AI could transition from patterns to common sense through an after-dinner game, researchers say.

Top End turns on fish-counting AI to avoid crocs and sharks Australia's Northern Territory is a very nice place to do a fish census if the crocodiles and sharks don't get you first.

Pictionary is a game more commonly associated with after-dinner family time than next-generation technologies and artificial intelligence (AI), but if researchers prove correct, this old-school game could give AI programs a modicum of common sense.

As reported by MIT Technology Review, researchers at the Allen Institute for AI (Ai2) believe that Pictionary, in which players draw an image to convey a word or phrase, could bridge the gap between the use of algorithms, machine learning (ML), cognitive computing and what we know as "common sense."

Perhaps difficult to define, we generally consider common sense to be the application of information and decision-making in deference to our immediate environment and the world around us.

AI can compute and perform tasks defined by its programming at great speed, but this does not mean that it can make the connections between objects and people or be able to reason based on real-world knowledge.

An example would be asking AI the pin-and-carrot question: if you stick a pin into a carrot, does the carrot have a hole or the pin?

The answer is obvious to us, but not so for today's AI models.

CNET: Google Doodle welcomes Chinese New Year with AI game

This lack of common sense is hampering a number of AI applications today, such as chatbots and live voice assistants, as they cannot translate or understand queries beyond the most simple, stripped-down questions.

However, the Ai2 team believes that training AI through Pictionary may result in the holy grail, common sense-equipped AI, becoming a reality.

In order to test the theory, the researchers have created an online version of the game, dubbed Iconary, which connects AI and human players, both of which are tasked with guessing the meaning behind an image.

Each picture is converted into clip-art icons and the AI then pulls phrases and words from a database in an effort to 'guess' the correct answer. If needed, the AI can keep guessing and request additional information to find the right answer. In total, the database contains 1,200 icons, 75,000 phrases, and 20,000 words.

TechRepublic: The 10 most in-demand AI jobs in the world

It is hoped that as the AI learns, it will develop its own kind of common sense by forming connections between different abstract concepts -- such as how a PC, keyboard, and mouse go together, for example, or the general link between a tea bag and a mug.

Microsoft co-founder Paul Allen is the creator of the non-profit lab and has recently invested a further $125 million into the organization's projects.

See also: Fairness in AI, StarCraft Edition

"When I founded AI2, I wanted to expand the capabilities of artificial intelligence through high-impact research," said Allen. "Early in AI research, there was a great deal of focus on common sense, but that work stalled. AI still lacks what most 10-year-olds possess: ordinary common sense. We want to jumpstart that research to achieve major breakthroughs in the field."

Previous and related coverage