Everyday we hear of a new company or product that has Artificial Intelligence (AI) in it. Cars with AI, calculators with AI, food kits with AI, cameras with AI, cell phones with AI, cups with AI, kitchens with AI, microwaves with AI, and the list keeps growing. What exactly are these companies adding? Most of the time, it just means a new feature has been added to a traditional product that uses an advanced computer algorithm. For example Google added the ability to extract text out of images on your camera. Basic algorithms have text extraction have existed for many years, but the accuracy has improved. Only recently have the algorithms improved to a level that is usable for daily use. The general public wouldn’t understand “text extraction using a 20 layer ResNet v3 network with skip connections”. Many of these companies are not using any AI at all, they say there is AI in their product to help get more funding or interest in their product, but really they just added 3rd grade level math problem slope of a line and called it AI.
So what is real AI? The goal of Artificial Intelligence is to build computer systems that mimic human level intelligence. So far, it has proven to be too hard to build. Researchers will often try to work on smaller specific problems (often called narrow AI or weak AI) such as voice recognition or better search.And because the public doesn’t really understand that, the term AI has been morphed into this marketing term that means “new product with incremental improvement of computer algorithms”. Sometime that incremental improvement is actually large enough to change lives such as when google first came out. The original technology was a scaled up version of frequency-inverse document frequency (TF/IDF), a machine learning algorithm for searching through large amounts of text combined with the algorithm to rank the importance of academic papers.
Artificial General Intelligence is the term researchers use for human level intelligence and there is no ambiguity of its definition unlike AI.
So next time you hear about a crazy new AI product, you can probably expect an existing product with a 5-10% increase in performance.