We Have Just Taught Robots to Read Closed Books; What Are the Implications?
It has been long established that robots would soon learn how to "read" words. In fact, developments in optical character recognition have made "reading" a lot easier digitally. However, it appears robots have taken the next step forward -- reading closed books.
Researchers at MIT and Georgia Tech are developing special imaging and machine learning algorithms to read closed books. According to their recently published paper, their tech uses terahertz radiation or electromagnetic radiation between infrared light and microwaves to scan pages in a book and identify letters.
The camera will shoot radiation at the book and produce distinct signals indicating where there is a blank page or printed ink. The algorithm will eventually process the data to distinguish individual letters, and they have been doing so for just nine pages out of a detectable 20.
According to Inverse, researcher Barmak Heshmat explained that the signal from deeper pages is weaker. The article explained that the technology uses an algorithm smart enough to beat text-based captcha system -- before no captcha re-captcha -- that uses a "dictionary" of possible letters and determines words.
This helps make "the invisible visible" as the implications are thrilling for those who want to transfer literature into digital form.
Alireza Aghasi from MIT tells Inverse the implications of such a technology. Aghasi explained trying to read through a book using electromagnetic waves is extremely challenging as there might be overlapping letters. Regardless, the immediate application of such a technology is to extract content information from old and historical books, where touching them could cause a lot of damage.
The idea also allows the algorithm to take pieces out of pages, which are characters, and try to construct words that best fit its appearance, akin to a puzzle. Interestingly, this poses quite the threat to captcha technology as this means robots are steadily being able to identify characters that are previously unknown to them.
While implications are still being listed, given that this is more or less a proof of concept, future devices will have more accurate reading capabilities.