Called Project Alexandria, Allen’s basing the research effort out of his Allen Institute for Artificial Intelligence (AI2) in Seattle. It will first seek to produce standard measurements to determine the common sense abilities of AI models. Once it’s possible to calculate whether something has common sense, researchers will be better placed to start figuring out how to teach it.
Project Alexandria will be looking at ways to crowdsource common sense knowledge from individuals. By collecting common sense reactions “at an unprecedented scale,” Alexandria will aim to develop a dataset sufficiently comprehensive to train an AI model. The system will work by aggregating data from images, text and human responses, folding all three into a unified repository.
The vision is that AI applications could then access this repository to approach problems from a common sense perspective. This could help to accelerate AI models that operate a wide variety of different workloads. Common sense might help boost performance in areas including robotics, machine leaning and computer vision by providing a shortcut to the output of a program.
READ NEXT: Samsung Galaxy S9+ claims the highest DxOMark camera score ever
Instead of computing the most efficient way to tackle a scenario, the AI could look first for a common sense response in the Alexandria repository. Successfully imbuing new AI models with an understanding of common sense would enable the technology to better replicate human thought patterns.
“Common sense represents one of the most fundamental and difficult problems for AI today – while AI has advanced dramatically in the past decade, there are still many things it cannot do, such as unstructured problem-solving or managing unanticipated situations,” said AI2.
“Project Alexandria will integrate and build upon the knowledge being developed in AI2’s other projects, including machine reading and reasoning (Aristo), natural language understanding (Euclid), and computer vision (Plato), to create a new unified and extensive common sense knowledge source.”
Allen’s $125 million investment in the project will be used to initiate Alexandria’s research. The New York Times reports the funding will triple AI2’s funding over the next three years. The additional financial resources will also be allocated across AI2’s other projects, including work to develop integrations with the common sense repository. AI2 was founded by Allen in 2013 as a non-profit dedicated to furthering research in artificial intelligence.
