Symbol grounding problem: CNC machines and a zebra

When I first saw CNC machines working at our factory, what came up to my mind was the image of a zebra, funnily enough. To be more precise, it was a sense of superiority to AI that I had when first learning that AI couldn’t even recognize a zebra. It’s called a “symbol grounding problem.” This was one of the biggest problems AI had. I don’t know if it’s fortunate or unfortunate, but 3D CG data had already solved the problem. Accordingly, my sense of superiority soon turned out to awe for AI, though.

What is a symbol grounding problem?

Professor Stevan Harnad named it as above. It is sometimes called “a zebra problem” from the example of a zebra he made. With its installed cameras, AI can process visual information. The problem is it couldn’t recognize a zebra because the verification data were character-string data. Let me explain it in more detail. A zebra is defined in the data “a horse with stripes.” Even if we’ve never seen a zebra before, we can recognize a zebra only with the definition. On the other hand, AI can’t because the above definition is just a set of 17 characters for AI. The core of the problem is: AI can store and process a lot of data, much more and faster than us, but none of them makes sense to AI.

Innovation made by 3D data

Now, the image processing algorithm of AI verifies visual information with its stored 3D data, and recognizes a zebra from its shape and surface pattern. AI can’t understand the meaning of any data the same as before, though. This innovation has also improved our furniture production. CNC machines work according to numerical data. We had to input each of them in the past. Now, computer calculates it out from 3D data created by scanning prototypes as shown in the above image. However, our production doesn’t get much easier because there’s no margin for error in placing parts in the CNC machine, for example. Expert skills are still required to handle CNC machines.

---

---