MYTH - AI Can Understand Humans

One of the greatest misunderstandings about AI is that AI actually “understands”.

AI’s “understanding” is very shallow. Machine learning uses massive amounts of data to optimize decision engines that have superhuman accuracy within one domain.

Given a massive amount of data in a particular domain, deep learning can be used to optimize single objective functions, such as “win Go,” “minimize loan default rate,” or “maximize speech recognition accuracy.”

However, these are far from “understanding”. Today’s AI is just just mathematical equations that can optimize these objective functions. For example, AI can recognize faces in a video, but cannot understand the nuances of the scene. AI can recognize speech and understand it in a given domain, but it is far from human’s broad comprehension of speech.

What are the Risks of AI

AI is Limited in it's Ability to Understand

AI has minimal ability to reason across domains, learn from minimum examples, plan in complex domains, apply common sense. And AI has no self-awareness, emotions or feelings. In these areas, AI is not even comparable to a three-year old.

The above concepts are not intuitive, which is why many people are easily misled. AI Superpowers explains this in detail using layman’s language, so that you can assess for themselves the potential and limitations of AI.

Buy the Book