We may have just figured out how to give AI true human intelligence.
Scientists have achieved a stunning new breakthrough in the world of artificial intelligence — whether that’s a good thing for humanity is anyone’s guess at this point.
Researchers have been trying to help computers learn on their own much like we humans do, and they think they may have succeeded in new research published this week, according to a PC World report.
This team of researchers created the Bayesian Program Learning framework and have been using it to teach computers to both identify and reproduce handwritten characters using just one example. The BPL framework allows the researchers to explain the data provided to the algorithm, which is different from standard pattern recognition algorithms that base concepts on how pixels are configured.
This algorithm focuses more on representing concepts, allowing the computer to produce the letter it sees by constructing the code itself, and learn how different people draw the same letter. It can then also learn other alphabets by basing it off of previous concepts, such as learning the Greek alphabet more quickly through its knowledge of the Latina alphabet.
This breakthrough allowed the computers to pass what is known as a “visual Turing test.” The researchers had both humans and computers reproduce handwritten characters after being shown one example of each. At the end of the test, human judges couldn’t tell whether it was drawn by a computer or a human.
The paper, which was published in the journal Science by researchers at New York University, the University of Toronto, and MIT, shows that computers can be taught how to learn things in the same way humans do, the researchers said according to the report.
It’s certainly an exciting development for helping computers do tasks that humans currently do, although many influential people like Elon Musk and Stephen Hawking have warned about computers becoming too intelligent.