For several years, AI systems have been approaching or exceeding human performance on many benchmark tasks. The benchmarks are usually tasks in which the AI system are Classifiers. That is the AI system receives an input data item and attempts to produce an output that is as close as possible to a specified classification target.
Recently, AI systems in another category have shown remarkable progress and now, on several tasks, can produce creative output that seems to be produced by an intelligent human. These systems are called Generators.
However, in spite of the ability to generate human-like output and to approach or exceed human performance on benchmark classification tasks, both types of systems lack common sense.
These systems are all based on a technology called "deep learning." More specifically, they use deep neural networks and are trained by a process called "gradient descent."
However, a classifier trained by gradient descend is inherently vulnerable to making a mistake that no sensible entity would make. For example, a deep neural network trained to recognize images various objects, including cats and cheese can be easily fooled to think a picture of a cat is a piece of cheese. If a mouse ever made such a mistake, we would not merely say that it lacks common sense, we would that its decision makes no sense at all. It is not a mistake that any real mouse would ever make.
Just a few years ago, the best generator systems were able to produce some images that looked real but also produced other images that looked ridiculous, like a horse with a head at each end or a dog with six legs. The text generators of a few years ago produced text that at first glance appeared to make sense, but that degraded after a few paragraphs. However, the quality for both tasks has made amazing progress. The best systems can now produce output that is hard to distinguish from art or text from that produced by humans.
However, these systems do not claim to have common sense. The leading text generator not only admits, it repeatedly asserts, that it cannot distinguish true statements from false statements and will apologize if you disagree or object to a statement that it makes.
Despite its apology and apparent modesty, it will repeatedly maintain a false statement. It will even continue to maintain and repeat two statements that directly contradict each other. If you point out that the statements contradict each other, it will apologize but then continue to make both statements. Again, not only does this violate common sense, it fails to follow the simple logic that should be easy for a computer.