- An A.I. called Aristo, developed by the Allen Institute, was able to correctly answer 90 percent of questions on a science exam designed for eighth graders.
- The success represents recent progress in the A.I. industry to develop systems that understand language.
- It doesn't mean computers are nearly as smart as eighth-graders, but it does suggest we might soon see some striking improvements in A.I.-based technology.
Could you score an ‘A’ on an eighth-grade science test? If so, you’re in the same league as Aristo, an artificial intelligence system whose remarkable language and logic skills highlight recent progress in the A.I. industry.
For context: Four years ago, some 700 computer scientists competed for $80,000 to develop an A.I. that could merely pass an eighth-grade science test. None scored higher than 60 percent. But now, thanks to improved “language models” driven by neural networks, systems like Aristo are becoming much better at predicting language and understanding how to apply it to solve logic-based tasks.
Aristo, as The New York Times notes, is built on a neural-network technology called Bert, developed by Google. Bert was instructed to “read” thousands of articles and books, through which it learned about the patterns and mechanics of language. Eventually, Bert was able to look at a sentence with a missing word and correctly guess what it was.
Similarly, Aristo, developed by the Seattle-based Allen Institute, “read” numerous questions and answers that might be found on multiple-choice exams. Over time, the AI was able to learn logical patterns. However, the system is designed only to interpret language, meaning it can answer multiple choice questions, but not those featuring an illustration or graph.
But it was able to answer logic-based questions like this:
Which change would most likely cause a decrease in the number of squirrels living in an area?
(1) a decrease in the number of predators
(2) a decrease in competition between the squirrels
(3) an increase in available food
(4) an increase in the number of forest fires
Aristo’s ability to correctly answer (4) represents a significant jump from previous A.I. systems, like AlphaGo. In 2015, the Google-developed AlphaGo became the first computer to beat a professional human Go player in a match without handicaps. But, impressive as that was, winning Go is a matter of learning and exploiting a fixed set of rules. In contrast, successfully learning and applying logic to answer questions about the world, as Aristo does, is another pursuit altogether.
Still, Aristo’s success doesn’t mean computers are anywhere near as smart as eighth graders, but it does suggest that we could soon see striking improvements in A.I.-based products, such as search engines.
“This has significant business consequences,” Oren Etzioni, a former University of Washington professor who oversees the Allen Institute, told The New York Times. “What I can say — with complete confidence — is you are going to see a whole new generation of products, some from start-ups, some from the big companies.”