AI’s next frontier issue is identifying the meaning of “none”

In the early days of mankind, we knew that one apple plus one apple equals two apples. We start counting from the physical, and we didn’t introduce the concept of zero until later, or we might say the number of apples in an empty box. The concept of zero has a revolutionary effect on mathematics, and today we can use zero in mathematical operations freely. But for artificial intelligence (AI), zero or nothing remains an area that has not yet been explored, and it could be a breakthrough in AI.

AI and deep learning need to be able to recognize meaningless and use no reasoning. Deep learning algorithms such as deep neural networks (deep neural networks or DNNs) have traditionally been supervised training areas to identify specific classes of things. After a large amount of high-quality data training, deep learning algorithms can accurately classify known objects.

But when an unknown object appears, the problem comes. If an unknown object that does not exist in the training dataset is introduced, the DNN has to guess. DNN, for example, trained with a collection of pictures of apples and bananas, and when faced with an image of an orange, it can only guess and classify it as the closest fruit — the apple– because its world is made up of apples and bananas.

The concept of no or zero can be useful when training DNN. If DNN can classify objects as apples, bananas, “no or unknown,” the developer will know that new categories need to be added. But so far, there’s no easy way to train DNN to produce a “no” response that signals developers that it sees something out of the ordinary.

Add a Comment

Your email address will not be published. Required fields are marked *