What are the current impediments to consistent computer-aided coin grading?

RWB · December 6, 2009 9:25AM

This has been raised before, but was wondering if opinions had changed.

SilverEagles92 · December 6, 2009 9:27AM

Computers can't determine positive or negative eye appeal.

Coin grading is an art, not a science.

LanLord · December 6, 2009 9:29AM

<< Computers can't determine positive or negative eye appeal.

Coin grading is an art, not a science. >>

Neither can a lot of people

PerryHall · December 6, 2009 9:29AM

<< Computers can't determine positive or negative eye appeal.

Coin grading is an art, not a science. >>

Bingo!!!

LindeDad · December 6, 2009 9:35AM

<< 

<< Computers can't determine positive or negative eye appeal.

Coin grading is an art, not a science. >>

Bingo!!! >>

Why can't they?

SilverEagles92 · December 6, 2009 9:37AM

<< 

<< 

<< Computers can't determine positive or negative eye appeal.

Coin grading is an art, not a science. >>

Bingo!!! >>

Why can't they?

>>

Eye appeal is completely subjective, that's why.

Mitchell · December 6, 2009 9:48AM

When the computer buys the coin, then computer grading will have arrived. Until then, the human eye Mark I is the final arbiter on eye appeal.

tbarreca · December 6, 2009 9:53AM

One could imagine using empirical studies to determine analytically the parameters of "eye appeal" by hypothesizing an initial set and doing subsequent correlation studies using pretty standard statistical techniques. You can create an N-dimensional parameter space, and then potentially learn to compute optimal values on each parameter. Topologically, the space defined by planar connections between the points in the parameter space is the "envelope of optimal eye appeal." Subsequent empirical tests can be used to refine the model, and well-known techniques in computer learning can be applied to engender automatic, continuous improvement. This is the easy part.

The hard part is the computer vision problem. Computer vision is currently best at tasks that have potential defense applications, for example, tracking motion in satellite imaging. But the grading problem is very different from that kind of application. Suffice it to say that there are no easy algorithmic, quantitative approaches for determining visual elements like "luster." Evaluating toning is even more difficult, since complex phenomena like iridescence come into play. Iridescence cannot, for example, be determined from a single, static image, but requires motion imaging. Think about it. Don't you tilt a coin through a series of angles to evaluate toning?

The human brain performs these processes virtually instantaneously, and we remain very far from able to duplicate this level of human cognition in an automaton. Until we can, grading will remain what so many others here have said, an art rather than a science.

SkyMan · December 6, 2009 9:55AM

Speaking as someone who interprets satellite imagery, and BELIEVE ME, there has been a LOT of money, courtesy of the DoD and the American taxpayer, spent on using computers to analyze satellite imagery, there are associational things that the human eye picks out that a computer "scan" of an image can not. So even this highly important national security question of interpretation comes down to an art as opposed to a science.

Remember, at the data level the image is solely a set of numbers. The assorted areas of an image; for example for a satellite image it might be roads, rock outcroppings, forested areas and open fields laid over a 3D topography; while for a coin it might be a Morgan $1 with fields and devices with dings and rub and maybe a little toning lit by a predetermined layout of lighting; will all produce digital values in many different wavelengths. These MAY give a signature (or as called in my field a "fingerprint", I just didn't want to use that term when dealing with coins) that is reproducible, but absolutely different items, let's say the shadow caused by a nick, may produce a signature exactly like the devices on the Morgan on the SW quadrant of a given light source. So, for the computer it would APPEAR TO BE the same. Granted, you could use a date/mm Morgan die template for the start of the interpretation, but that would only be a very crude starting point and die wear and the resultant strike characteristics would lead to a "wrong" answer pretty quickly.

LindeDad · December 6, 2009 10:00AM

The title says computer aided grading if eye appeal is really a part of the technical grade of a coin then the grader would have to do that part.
But initial scan of the coin for authenticity and recording and surface condition or initial grade as a deference from original mint state could be determined by computer.
We are all, including the third party grading firms, worried about the counterfeits that are flooding the hobby but if high resolution scans of the coins are filed as part of the grading process follow on questions of whether a coin is real could be handled online by sending in a image of the coin.

But I sure have seem some rather ugly high grade coins and way to many over priced under grade coins that somebody liked the sunset colors of.

MrEureka · December 6, 2009 10:15AM

The biggest impediment may be the public's expectation of a grading guaranty. I say that because any worthy computer grading system will improve over time, and every improvement will create a new standard. In other words, a computer-based TPG cannot guaranty its grades.

tbarreca · December 6, 2009 11:23AM

<< ...Granted, you could use a date/mm Morgan die template for the start of the interpretation, but that would only be a very crude starting point and die wear and the resultant strike characteristics would lead to a "wrong" answer pretty quickly.... >>

Great post, Skyman, and right on the money in every particular. I cite the quote above because I think your subjunctive should be imperative, i.e., not only "could" you use such a template-based approach, you'd pretty much have to do so. In fact, it came down to assuming such templates in a first approximation of such a system. However, it turns out that the construction of the templates is a massive problem in and of itself!

Bear · December 6, 2009 11:27AM

I would hate to see the day when a machine can replace the human

sense of appreciation and perception of what a beautiful coin is and

should look like. That would prove too much of a loss, of what makes

human beings, so unique and special.

Quarternut · December 6, 2009 11:36AM

What are the current impediments to consistent computer-aided coin grading?

The fact that computers are still not as smart as humans...(at least some of the smarter humans!

)

QN

MrEureka · December 6, 2009 11:46AM

I would hate to see the day when a machine can replace the human

sense of appreciation and perception of what a beautiful coin is and

should look like. That would prove too much of a loss, of what makes

human beings, so unique and special.

Bear - To whatever extent this is true of a computer-based TPG, isn't it equally true of human-based TPGs?

RedTiger · December 6, 2009 12:34PM

Money is probably the biggest hurdle. Coin grading is a niche business, there is not much money in it. Consistent computer grading would shrink the pie even more with resubmits becoming a thing of the past.

To those that think it can't be done, think again. Have the computer learn the same way humans do, by getting a reference set or grading guide, and then "looking" at lots and lots of coins. By building on real experience, the computer will learn almost everything the human learns, and after time will be as good if not better than almost all humans. The approach used 20 years ago in the PCGS project was more brute force, and did not build upon the human experience.

I've written about the chess program analogy before. Chess was once seen as the "holy grail" of artificial intelligence. Many thought it would never happen because there are so many variables, so many possible board positions. What broke the "nut" was an approach that fed every board position from every grandmaster level chess game ever played. The computer was programmed to look for similar board position possibilities, and whether those positions led to a win, a loss or a draw. By building on human experience, and having a huge database of human expertise to draw upon, the computer was able to beat the human. The computer never gets tired, never gets sloppy. If fooled it will store that mistake away and likely never make that mistake again.

Chess was seen as an academic pursuit, so a lot of research and grant money got thrown at the projects. If similar resources were available for coin grading, it would be done in a relatively short period of time, and done relatively well. Again, eye appeal would be learned by drawing on human experience, and a huge database of images of graded coins.

However, there's little money in it. The entire combined revenue for all the grading companies is modest, and the two entrenched companies have a stranglehold on the market. Many industry players do not want consistent computer grading, because that would mean no more crack out games.

mrpotatoheadd · December 6, 2009 12:40PM

The problem I see with rating eye appeal by computer is that you can show people a coin and get some replies saying "Love it!" and others saying "That thing sucks eggs!". How is a computer going to choose between the two?

RedTiger · December 6, 2009 12:43PM

<< The problem I see with rating eye appeal by computer is that you can show people a coin and get some replies saying "Love it!" and others saying "That thing sucks eggs!". How is a computer going to choose between the two? >>

Same way pro graders learn it. By looking at thousands and thousands of coins. The computer will draw on a database of graded coins, that humans have already rated for eye appeal.

COALPORTER · December 6, 2009 12:48PM

The problem is someone would need to tell the computer what a MS65 is, what an AU58 is, etc. Computers are still "dumb" and can only do what they are told to do. This would still require a human's input to get started, so the idea is flawed from the start.

For low population coins, I don't think it would work too well, because by the time you "trained" the computer, you will have used up all the coins anyway (which have been graded by a human, so thinks won't work).

PS. If computer grading ever works, wouldn't it be funny if they made a Cray check a HP and give it a sticker (or not) ?

mrpotatoheadd · December 6, 2009 12:51PM

<< Same way pro graders learn it. By looking at thousands and thousands of coins. The computer will draw on a database of graded coins, that humans have already rated for eye appeal. >>

I'm not arguing that point. What I meant was that the computer doing the grading is only going to reflect the opinion of whoever set up the program. Not everybody agrees on eye appeal, so computer generated grades incorporating eye appeal may be more consistent overall, but are not going to help *you* much if *you* don't agree with the computer about what's attractive.

dpoole · December 6, 2009 12:51PM

I don't see why coins could not easily pass through an initial scanning to do technical grading, to assess for wear at the designated points, hits, etc., according to designated and objective criteria.

The less objective factors more dependent on massive numbers of examples, such as lustre and eye appeal, may be preserved for the artistic human grader to posit as a parallel opinon on the technical grade.

derryb · December 6, 2009 1:12PM

Well, considering most computer chips are coming in from China, any such program would probably never spot a counterfeit.

keets · December 6, 2009 1:30PM

things would be helped if there was just a set of standards for each individual grade which would assist in assigning a technical grade that any/all of the grading services actually follow, then "eye appeal" or however you want to term it could be factored in. if a coin is technically an MS64 it just makes no sense for it to be graded MS66 just because it has pretty tone. this has all been discussed before with threads on Market Grading and it seems clear that the issue will never be satisfactorily resolved for the benifit of the hobby and collectors. what happens is that a coin which is an MS64 gets graded MS66 because it's got "eye appeal" and then it gets priced at multiples of sheet because it's PQ.

it's madness, i tell you, sheer madness!!!

like the old saying goes, "I just play the game, I don't make the rules."

RedTiger · December 6, 2009 1:42PM

<< 

<< Same way pro graders learn it. By looking at thousands and thousands of coins. The computer will draw on a database of graded coins, that humans have already rated for eye appeal. >>

I'm not arguing that point. What I meant was that the computer doing the grading is only going to reflect the opinion of whoever set up the program. Not everybody agrees on eye appeal, so computer generated grades incorporating eye appeal may be more consistent overall, but are not going to help *you* much if *you* don't agree with the computer about what's attractive. >>

Humans disagree too. The advantage for the computer is that it will be 100% consistent and remember every coin that it has ever seen.

Again with the chess analogy, the top chess programs draw upon the human experiences by looking for winning positions from games that have already been played. By having a coin grading program draw upon the information in mass of already graded coins, it will learn from all that experience. That wasn't the approach used by the older programs and why the older programs could not beat the best human chess players. The humans could draw upon their experience while the computer had none. When the approach changed and the computer could draw upon the collective experience of the best human players from the last century of play, the computer became virtually unbeatable.

So it would be with coin grading. Now that there are millions of graded coins out there, a computer that can draw upon all that information, and also remember every single coin, and every single mistake it makes, will be almost certainly be better than the best humans. Copyrights, information rights, and money for the project would be the primary hurdles to that approach.

Coinosaurus · December 6, 2009 3:03PM

Given a zillion coin images and the human-assigned grades, a computer should be able to "learn" how to do it, and to track market-grading changes over time.

One of the current impediments is getting the computer to physically see everything the human does. The human holds the coin in his/her hands, tilts it to check out luster, etc. The information being transferred between coin and human in this process is FAR greater than what you get in a scan or even a good photo. Capturing this experience in digital form for thousands of coins is quite expensive.

Barry · December 7, 2009 12:08AM

Computer grading would provide consistency. It would eliminate the crack-out game, therefore it would be detrimental to both dealers and TPGs, eliminating the merry-go-round of the same coin being graded and sold over and over again.

Kinda like the oil companies suppressing the 100 mpg car

EagleEye · December 7, 2009 12:18AM

The only practical computer grading system is one that tracks the same coin. When a coin was graded once by a person, the image is stored. When the coin is sent in again, regardless of the holder, or lack thereof, it will get the same grade.

Of course, it might have gotten the wrong grade (by the human grader) to begin with.

Although consistency is desired in grading, one price = one grade is not. Imagine a dull morgan dollar with no marks getting the same grade as a beautiful flashy example. Are they equal?

Also I suspect many of the collectors out there play the cherry pick for quality game to augment their collecting budget. Take the fun out of getting a score, and you may lose interest in building that set.

Gemineye · December 7, 2009 2:13AM

<< What are the current impediments to consistent computer-aided coin grading?
>>

Once a coin has been "graded" by a machine it would then be entered into memory.There would be essentialy NO chance of a re-grade or an up-grade later to challenge the original "grade".There also would be no need for stickers either..!!!..for their grades are in-fallible and error free...!!!...Only humans mmakee errrorss....!!!...

messydesk · December 7, 2009 3:19AM

I stand behind what I said on May 7, 2005, here.

Walmann · December 7, 2009 3:35AM

The design of a coin is art, mass produced machine production of coins is not.

Not so sure if the art aspect of grading is the true impediment, or if perhaps the graders have a union?

tjc2120 · December 7, 2009 3:46AM

<< I would hate to see the day when a machine can replace the human

sense of appreciation and perception of what a beautiful coin is and

should look like. That would prove too much of a loss, of what makes

human beings, so unique and special. >>

I agree replacing a human's sence of appreciation and perception would be a loss but the ability to create a machine that can approximate that would only confirm in my mind what makes human beings so unique and special.

Nysoto · December 7, 2009 4:59AM

Computer aided technical grading would work well for coins of relatively consistent strike. For pre-1839 coins (and others), each die variety strikes differently, with each die pairing there are varying die stages and striking pressure. Is a flat spot wear or a soft strike? A computer could not make that determination as it would rely on human programming from variables that are not completely known.

The TPG's would always be against computer aided grading because it would be more difficult for them to build in subtle grade inflation over the years to increase re-submissions. They would be forced to be more honest, putting them into a difficult position that is out of their comfort zone.

Dropdaflag · December 7, 2009 5:11AM

I agree with Rick Snow. My last purchase from Rick was possible due to several local scores.

RWB · December 7, 2009 5:28AM

The simple solution is for collectors to simply quit asking the grading companies to hold their
hands and pat them on the head.
[Cladking May 7, 2005]

WaterSport · December 7, 2009 7:53AM

I think it comes down to two reasons grading even exist, to make money, and bragging rights.

WS

CHAMELEON · December 7, 2009 8:43AM

<< 

<< 

<< 

<< Computers can't determine positive or negative eye appeal.

Coin grading is an art, not a science. >>

Bingo!!! >>

Why can't they?

>>

Eye appeal is completely subjective, that's why. >>

What does eye appeal have to do with the grading of a coin? Isn't the grading system used to identify the cleanliness and strike of a coin? A coin grading MS-63 with vivid neon toning should still grade MS-63 if the toning were to turn completely black. I've found that a lot of times TPG's will grade a point or two too high for exceptional toning, but in reality the eye appeal of a coin should not add or subtract from the actual grade.

What are the current impediments to consistent computer-aided coin grading?

Comments

Leave a Comment