A Self-driving car with Artificial Intelligence met with an accident by accelerating at the stop sign rather than slowing down. As the accident report there were four rectangles had been stuck on the stop sign. This misleads AI to read stop sign as speed limit 45.
Researchers also said how to fool an AI system like misleading facial recognition AIs by sticking printed patterns on glasses or hats and also have tricked speech recognition AI by sound of phantom phrases with patterns of white noise in the audio.
They were only some examples how to break pattern matching made by Deep Learning Networks .Deep Learning Networks has been proved successful in classifying objects of he realtime world like cat,dog,plane,man,woman,etc.. , but if small changes made in the inputs of the Deep Learning Network can easily spoil an AI. So the misleading or changed input can also lead to a misclassification of an object through a Deep Learning Network. Input is all that matter to a Deep Learning Network if its correct classification is correct or opposite or misleading.
Deep Learning is being highly used nowadays in real world rather than labs like self-driving car, mapping crime and health diagnosis. If the pixels are maliciously added to the input of the cancer detection can lead to wrong diagnosis of cancer. According to this hackers can also use this weakness to takeover the self-driving car and misleading it for its own purposes.
To correct this researchers have worked alot to find what is the reason for Deep Neural Network fails. François Chollet, an AI engineer at Google in Mountain View, California said that, there is no fix for the fundamental brittleness of Deep Neural Network. To remove this flows he and other researchers suggest to improve pattern matching Deep Neural Networks by adding extra functionalities like Artificial Intelligence exploring world themselves, can write their own codes and retain memory. This type of system will be the next story in Artificial Intelligence research.
Check Out My Related Articles:
Check out my News Website —-> The Hindustan Times