Computer grading requires human input

It's almost 2008 and there is still no "little black box" to grade coins, but some have tried. This weekend, while looking through some old clippings, I came across an article about "The Expert."

The subject of the article called to mind a joke I'd heard from a former associate, Charlie Hoskins. He used to say that the word "expert" came from two roots, "ex" meaning from or out of and "pert" a little drip under pressure. In this case, "The Expert" in the article was neither a drip nor a human. Read on.

About 20 years ago, one of the grading services developed an expensive gadget to grade coins called "The Expert." Apparently, it was a video camera surrounded by a bank of high intensity lights set up to record the surface of a coin. The coin's image was digitalized as each light flashed in sequence and the resulting image was stored in a computer.

This was one of the first attempts to develop a computer to grade coins. There were several other parties working to develop a computer grading system at the same time. I recall that Charlie Hoskins, director of The International Numismatic Society Authentication Bureau, was a consultant to one of these firms that eventually produced a product under the name "CompuGrade." Since I was not involved with this project, I have no idea how that system worked. That grading service lasted a few years; yet, I cannot be sure if a computer ever generated any of the grades on their slabs.

One thing I did know for sure, at that time, computers could not grade coins. I based my opinion on the problems one firm was having setting up a system to grade Morgan dollars that were toned, and on an interactive demonstration video of how "The Expert" worked that was set up at an American Numismatic Association convention. It didn't work for me.

More importantly, I understood the difficulty of detecting "doctored" and "problem" coins, even for an experienced numismatist. From what I could see at coin shows in the 1980s, there were just too many altered coins going into slabs while the "new" grading services were in their infancy. If some of the professionals who were programming the computers could be fooled so easily, what kind of results could we expect from a machine? I think the saying goes, "junk in, junk out."

Thankfully, that is not the case today. The major grading services have become one of the best safeguards against "doctored" coins in the marketplace. So where does that leave us? I believe there is a place for "computer grading." Who can say how much better the results will be now with the vastly improved hardware and software? Perhaps a different approach, one that may have worked even 20 years ago, may be tried.

What are the most important aspects of any grading system? As has always been the case, it should be simple, precise and consistent. In an ideal system, a coin would receive the same grade no matter how many times it was submitted as long as its condition of preservation remained unchanged. That is what computer grading was to provide to numismatics. Grade it, put it in a holder and that would be the end. This was possible 20 years ago and is still possible today except it would not be the "true" computer grading as envisioned long ago.

Grading is still too complicated a task for a machine. What if "The Expert" had been set up at the beginning to record the imperfections on a coin, to fingerprint it the way they do for diamonds and then to store that information so that it could be pulled-up from a database as a "match" if the coin was ever seen again?

With this system, the computer would not be needed to actually grade the coin; yet it would eliminate some of the inconsistencies we find with human graders. Human experts can grade the coin. Then the results, plus a scan of the coin can be stored in a data base. Voila, computer grading!

Each time the coin was submitted, it would be returned with the same grade. End of the grading problem.

Why didn't they take this approach 20 years ago?