The Allen Institute for Artificial Intelligence introduced an artificial intelligence (AI) system that successfully passed an eighth-grade multiple-choice science test, correctly answering over 90% of the questions, as well as scoring more than 80% on a 12th-grade test.
The Aristo system's milestone suggests understanding the language and logic that high school students are expected to possess is no longer outside AI's capabilities.
Aristo took standard exams written for students in New York schools, with questions including pictures and diagrams removed; some questions required simple information retrieval, while others required logical thinking.
Aristo was built atop Google's Bert, a language-model system that learned via guessing missing words in sentences. By feeding Bert a broad spectrum of question and answers, the researchers enabled it to answer similar questions by itself.
From The New York Times
View Full Article - May Require Paid Subscription
Abstracts Copyright © 2019 SmithBucklin, Washington, DC, USA