— Tim Rathjen (@SandCTim) November 23, 2016
I hadn’t even posted the final piece in my 3-part series about computerized grading when I was contacted with some incredible news. There is a coin dealer in northwestern Washington named Tim Rathjen, he runs The Stamp & Coin Place. He has also invented a machine. At first glance, the contraption appears to be a simple coin sorter. It isn’t until you look in the collection bins that you see the genius at work.
Called simply “The Machine” by its creators, the invention utilizes bright field digital imagery and coin recognition software to sort coins by type, year, mint, grade, and value.
How it works…
- Coins are held in a hopper that feeds them onto a conveyor belt.
- The conveyor transports each coin to the camera for imaging.
- The digital image is “fed through” the software.
- The coin moves past imaging, is aligned with the proper bin, and continues to its decided destination. Depending which software program is running (there are currently over 40) the machine determines what to look for and sorts the coins accordingly. Per Tim Rathjen’s notes, here is how they sort a bag of common wheat cents:
Bin 1 – 40’s P & D
Bin 2 – 40/50 S-mint
Bin 3 – red/bright
Bin 4 – memorial/Indian/Canadian
Bin 5 – high value $2 or more
Bin 6 – 30? It knows it’s in the 30’s but can’t make out the last digit of the date or mint mark.
Bin 7 – 30-S
Bin 8 – 30-D
Bin 9 – 30-P
Bin 10 – dark/cull
Bin 11 – 20’s
Bin 12 – teens
Bin 13 – 50’s P & D
5. The machine records how long it took the coin to travel from imaging to the final bin. At normal operating speed, the machine can process 3 coins per second. That’s 180 coins per minute | 10,800 coins per hour | 86,400 coins in eight hours. Put another way, the machine can accurately sort 17 full cent bags in a standard work day.
Bright Field vs. Dark Field
I mentioned bright field imaging above and wanted to clarify what it means. Bright field imaging is where a light source is shone directly down on the subject being photographed, in this case a coin. This is accomplished by means of a two-way mirror. The light source is actually situated above and to the side of the object, reflected down by the two-way mirror (situated directly between the camera and the object), and the camera snaps a photo through the two-way mirror of the well-lit object beneath.
Dark field imaging is where the light source is again above and to the side of the object, but no mirror is used. The light shines in on the object at an angle and brings into stark contrast the raised surfaces from the low surfaces of the object. Any astronomy nerds out there know that the best time to look at the moon is when it is partially illuminated because the viewer can really see the craters from the shadows they cast. That is basically what dark field imaging does, only it is over the entire surface of the coin.
Tough to visualize? It was for me. So, here’s an image to help illustrate.
It is the argument most often used against computerized grading. How can a computer assess eye appeal? It can’t possibly grasp the concept of beauty, right?
It doesn’t have to. I explain it more thoroughly in “The Road to Computerized Grading: Part 2,” but we can teach a computer program to “see” like a human. It is done in incremental steps. Think of it this way—to a computer all the world is made up of data, everything is just 1’s and 0’s. If the computer is told that a certain range of combinations of 1’s and 0’s are called “beautiful,” it will call all future similar combinations beautiful. For instance, it will see all toned coins as beautiful, which obviously isn’t the case. Some toned coins are downright ugly. When the computer points to these ugly toned coins and says, “Pretty!” we simply tell it, “No.” By this process of eliminating exceptions we slowly teach the computer to recognize eye appeal as a person would. The actual process is far more complicated than that, but, as always, I like to over-simplify for the purpose of illustration.
The Limits of The Machine 1.0
The Machine has limits. Right now it can only accurately grade coins up to XF. As you may have noticed in my header for this paragraph, I’m calling it “The Machine 1.0.” This means that it is merely the first incarnation of this breed of computerized grading machine. The fact that it can only handle the coins where differences in grade are pronounced (PO – AG – G – VG – F – VF – XF) shouldn’t overshadow the fact that the machine can grade coins. Anyone expecting perfection from a first functional attempt has clearly never created something from scratch. It is a process of iteration where every version can do more than the previous. I explain this further in “The Road to Computerized Grading: Part 3.”
In his book Zero to One, Peter Thiel talks about how technology means augmenting humans in our tasks, not replacing us. This is the future I predict for computerized grading. Even if a software program can be taught to accurately grade up through MS-70, would people really trust it? We’re social animals, we like to stick with our own. I think it is most likely that we’ll see TPGs with grading machines and people to verify the results.
However, couldn’t one knowledgeable numismatist purchase a machine and go into business for themselves? If all that is necessary is a quick verification of the computer grade, one person could get a lot done. Plus, the machine need not be more complicated than a good camera, light source, and software package, if the grader is processing submissions by hand. This would allow someone to professionally grade coins from home, with no need for a large facility. Or, if not as a business, a collector could do this for themselves.
It’s all about augmentation, not replacement, and what makes us more independent will win out over the things that hold us hostage. ❑
Kendall Bailey runs TheCoinBlog.net.