Although Google's Neural Machine Translation system is able to produce translations that can sometimes match the accuracy of human translators, artificial intelligence (AI) might never be able to learn and completely understand human language.
Google notes its system still makes significant errors that human translators would never make, and struggles to translate sentences within the broader context of a paragraph or page.
To recreate human thinking and speech, AI must be able to mimic human learning, which goes beyond compiling data; the brain processes information through emotional and social filters and commits new information to memory to create and break patterns.
Researchers try to duplicate this process in machines using neural networks, which can now recognize objects, animals, or faces. However, recognizing words and phrases is much more difficult for current neural networks.
According to Fei-Fei Li, director of Stanford University's Artificial Intelligence Lab, AI researchers will need to integrate emotional and social understanding, abstraction, and creativity, as well as raw information to create a machine with a humanlike grasp of language.
In its 100 Year Study of AI, a Stanford panel agreed a future inhabited by superhuman robots is unlikely to be possible.
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found