Hacking a brilliant machine is not a distant dream. Microsoft and Alibaba’s research showed that robots will replace humans in the area of employment and other aspects of life. The two companies came up with the research that states their artificial intelligence beats human’s brains in reading comprehension.
The test recalled as the Stanford Question Answering Dataset (SQuAD), was made to instruct AI to respond to questions about some Wikipedia articles. The image-recognition software which has been embedded in commercial photo apps, these systems makes us believe that machines have become exceedingly proficient of reproducing human perception, discerning images or sounds and swiftly reading text passages and throwing back answers with human precision.
However, the machine experts who configure deep-learning networks and other AI systems are discovering just how delicate their creations are by hammering down to see if these machines really know anything. Stress-testing software could be installed into a self-drive car would be vital to prohibit errors that could cause disastrous accidents.
Anish Athalye, a Massachusetts Institute of Technology graduate student who researches AI said that in some areas neural nets are surprisingly super humans, they beat human accomplishment. But they possess these strange properties that lead us to believe that we can trick them very easily.
Athalye and other students at MIT inclusively have written articles known as LabSix, indicated that they could generate a grave learning system, one instructed to identify objects from numerous examples like making it think that the skier was a dog and turtle was a rifle.