AI in AR | Can You Trust It? What AI is and isn't.
AI - Can You Trust It?
Depending on who you talk to, AI is best thing to ever happen, a great asset to business and mankind, or an incredible danger that will take your job and ultimately destroy the world. Which do we believe? How do we sort through the media-lead frenzy of hype and sensationalism? How can we determine if it has the ability to help us perform day-to-day business tasks?
What AI Is
The first step is to truly understand what AI is today. The type of AI garnering all the headlines is called deep learning. It is truly powerful and has a lot of promise. Unlike conventional computer programs which deal in predetermined specific ways with a specific set of events, the power of deep learning is its ability to achieve a desired set of results without the need to program or even know how it happens. Very simply, we configure the deep learning model, define the outcomes we’re looking for and, using a large number of examples, train the system to reliably give the correct result. The current hype arises from the ability of this type of AI to perform many “intelligent” tasks as well as humans.
How AI Works
The important thing to understand is how this really works. AI, in its current form and stage of evolution, is able to produce “intelligent” behavior, but it’s not really intelligent at all, except in a very artificial sense of the term. Deep learning networks learn by statistical analysis. Their neural nets are driven by optimization algorithms structured to minimize the difference between their answers and desired outcomes as they “learn” example by example. In short, today’s state-of-the-art AI is nothing more than a statistical inference engine.
What AI Isn’t
So, if AI produces intelligent behavior and undeniably learns, even if it all reduces to statistics, why do I say it isn’t truly intelligent? Because it has no common sense or understanding of what it does. This is easiest to illustrate by example. An AI can be trained to recognize the difference between photographs of dogs and cats and score as well as a human. But the AI can be easily fooled. A knowledgeable person can configure what appears to be a random collection of dots that the AI will positively identify as either a cat or a dog. A human, or a system possessing true intelligence, would never make that kind of mistake. We’ll explore the implications of this in later posts.