Machine learning is ever better but can still make mistakes.
Andrew Ilyas, MIT, has managed to fool object recognition AI with minor tweaks to an object’s appearance. The object in question was a turtle and the only tweaking was to the pattern on its shell. But this was enough to identify the turtle as a gun. This problem with AI keeps coming up. It is not that humans do not also make mistakes. But one hopes that common sense might prevail. Or, if it does not, then we have ways of handling the mistake when it occurs. We have not yet figured out to cope with AI-led mistakes. Even a trivial error by the algorithm can lead to a major error in behaviour.
Link to article: https://www.sciencedirect.com/science/article/pii/S0262407917322042
You may also like to browse other AI posts: https://www.thesentientrobot.com/category/ai/