Why is there no AU59?

In a perfect world, all 70 grades should be possible. I've wondered if some of the current grading ties to the old system in some way (ie, BU, choice, gem BU, AU, choice AU, ...). As others have said, grading is still highly subjective; even in our technology advanced world.

 

Log in or Sign up to hide this ad.

Actually, using AI autoprogramming may less satisfactory in this case. The machine will have to learn grading from someone. The subjectivity and inconsistencies of these folks will be transferred into the machine. Depending on the AI algorithms, the internal programming may be subject to drift in an unacceptable direction. The machine may have to be periodically retrained, and again it would pick-up the subjectivity and inconsistencies of its human trainers.

With explicit grading rules (i.e. fixed programming), multiple persons can review them and make periodic changes. Changes would be fewer and fewer as time went by. Drift would not occur.

Either way, transfer of human subjectivity would occur. In one case, it would be locked-in, but reviewable and subject to explicit change by humans. In the other, the inner workings would be constantly changing and unknown and unreadable by humans.

Cal

 

Once again, this is NOT true. The only human input that would be needed is to exclude obviously misgraded coins, perhaps by a panel of expert reviewers, and merely have the neural net "learn" from already graded coins with those manually-determined "clinkers" removed, both for overgrading and for undergrading. That is all that humans would be needed for. Eventually, with machine learning, the system would ALSO be able to identify OTHER misgraded coins NOT flagged by the human panel and report that, as well.

You'd darned well BETTER start believing in AI's value, because industry is about to jam it down your throat in MANY ways, if you like it or not.

 

I have often tried to make sense of the 70 point scale. If you draw a 7 in. by 7 in. box, then at the lower right corner as a pivot point, strike a radius from the lower left corner to the upper right corner, then draw 1 in. grid throughout the box, then label the grid from 10 to 70 from left to right and from bottom to top. Where the grid crosses the radius one grade stops and another starts. You now have a 70 point grading system which makes perfect sense. Notice that the upper right end of the grid has more room for higher grades of UNC. to fall into, with plenty of room for + grades, while the lower grades simply fall into their GD, VG, FINE, V FINE, E FINE, AU. categories. Clearly there is ample space for 60 thru 70 grades at the upper right end of the curve. If the scale is drawn with a straight line rather than a radius, the 70 point scale simply does not work for the upper end coins which most grading is all about. I've used this grading scale for some time now as it does make sense of it all. I believe the grading services adopted a very similar method of designation when 70 point grading scale began.

 

I think we'll continue to disagree on this. There is subjectivity in the current set of graded coins, and deciding what is and is not a clinker will be subjective. Speaking as someone has written code since the '70s and who still does, I have some understanding of AI and have a feel of where it may work and where it may not be the best choice. I've done programming with the AI languages Lisp long ago and a tiny bit more recently with Python. However, I mostly use Visual Basic, Power Basic, and Arduino C/C++. No need to preach to me about the value of AI, I'm convinced, but have some feel for its limitations.

Cal

 

Does anyone who is not a dealer actually believe we "need" 11 Mint State grades?

 

Anyone trying to make sense of the 1-70 scale in the current market is going to fail. In 1949 when Sheldon set up his pseudoscientific pricing system, a full red uncirculated early large cent was worth, at auction, about 70 times what a barely identifiable example of the same variety was worth. Sheldon's scale was a pricing scale, not a grading scale.

We have it now for all coins because dealers saw the opportunity to make more money by getting leverage over collectors by using a scale that was illogical and at the same time gave the false impression that grading was quantitative and so reliable.

 

I agree that most of the grading issues come in the MS states. I've seen some MS65s that had a lot of bag marks and some MS63s that had almost none. Or a MS67 that was so poorly struck that it looked like an AU55 (with mint lustre). These example are extremes, but when you start slicing up the grades so much there is bound to be issues like that.
When I go to shows I look for MS states coins that I feel of under graded. A well struck MS64 can look better and be cheaper than a poorly struck MS65 or MS66. I think (hope) most collectors do that, but I'm sure a lot don't.

 

many don't know how to tell the difference between an actual MS-65 and an MS-66, let alone one good for the grade and poor for the grade.

 

I’m not a dealer, and never have been one, and not ONLY do I believe we need all 11, I’m pretty close to totally convinced we need more, like below 60 in MS plus the + grades too. I go to shows and look for UNGRADED coins. It’s kinda my thing.

 

I too have been writing code since 1971, and the main thing that experience has taught me is that that experience is pretty useless at the cutting edge today. Yeah, it gives me a leg up writing old school stuff, but true machine learning is an “experiential” paradigm, not a programming one.

 

There seems to be NOTHING IN THIS WORLD that folks older than me can’t interpret as a scam to steal their money, right?

 

AU56

 

AU58 in the opinion of CAC. Grading Indians is a "black art" anyway, but from what I see here, I can't see why it would even be as low as AU58.

 

At least a 58 in their opinion, they may view it as MS but only they can answer that.

 

I'll admit I can't tell a 69 from a 70. In fact, if they WOULD ever go "machine learning" for grading, one of the first categories of "clinkers" they'd need to exclude manually are 70's that have obvious faults, a lot as milk spots, but not ALL that. Scanning ALL current 70 slabs and telling a computer system they're all perfect would create the stuff of a Star Trek episode.

 

It's all about money.

 

All that needs to be said about computer grading and AI grading is that the market does not want it. The ability has been around for decades and it has been tried before and failed. The market doesn't want purely technical grading.

That said AI is already in use at at least PCGS with counterfeit detection and pointing out abnormalities for closer inspection.

 

Sure, as long as we fall for the fallacy of calling PCGS the spokesmen for "the market", you're right. Why would they want anything that reduces the number of Beemers and Benzes in the parking lot?

Why should coin grading be exempt from the "disrupt everything" ethos of the digital revolution? Once someone demonstrates that AI grades more reliably than PCGS, they go obsolete.

 

qwertyuiop