Grading coins and inherent variance of subjective differences

RollBuster

Part of the grading process is the subjective interpretation of the grader. Then I understand that a group consensus is reached. Many human variables can affect the results much have influence over the outcome. What is the thought of the introduction of artificial intelligence using an optical comparator? It would make for more consistency and exceptions can always be reviewed by a human.

yosclimber

Grading needs more than just consistency, i.e. low variance.
It needs accuracy, i.e. a low mean square error.

It has proven difficult so far to create grading / computer vision software that equals the accuracy of good human graders.

jmlanzaf

"Eye appeal" is not exactly objective nor definable with an "optical conparator".

Anyone who thinks AI would be more consistent doesn't fully understand AI. The surface of every coin is unique. Grading isn't based on simply counting marks. Where they are located matters as well as how obvious they are. AI would always give the same result for the same coin. But it wouldn't necessarily interpret two different 65s as being 65s.

JBK

If AI robots are going to buy and collect the coins then AI robots can grade them.

But where humans are involved, there are determinations and nuances and preferences that can't be programmed into a computer.

sellitstore

The problem isn't in the hardware end. Consistent, reliable scanning technology has existed for decades.

The problem is the subjective element. How everyone values the several different relevant factors varies greatly among collectors, and this can vary over time, too. When TPG grading started getting big in the late 1980s, a weakly struck coin could not grade gem. Today, there are many weakly struck coins in gem holders. The importance of strike has diminished in the eyes of the market. But I still won't accept a weakly struck coin at gem prices, as it's not a gem by my definition. With time and experience everyone develops their own unique grading criteria and preferences. And that's why grading will always have a considerable subjective element.

Grading, even TPG or scanner computerized, will always have a large subjective element. We simply don't and won't agree on what is important and what the proper weight to assign to each grading factors should be.

291fifth

Learn to grade. Set your own strict standards and stick to them. Forget about AI.

Manifest_Destiny

Aren't they already doing it?

https://forums.collectors.com/discussion/1042609/pcgs-launches-ai-powered-coin-grading-technology

spyglassdesign

@jmlanzaf said:
"Eye appeal" is not exactly objective nor definable with an "optical conparator".

Anyone who thinks AI would be more consistent doesn't fully understand AI. The surface of every coin is unique. Grading isn't based on simply counting marks. Where they are located matters as well as how obvious they are. AI would always give the same result for the same coin. But it wouldn't necessarily interpret two different 65s as being 65s.

Not only that, the lens only picks up what it can see... A human can tilt and twist the coin around in the light that might reveal other good or bad factors that may not show up in a scan or image. This is the hardest part for me personally when imaging a coin... Sometimes imperceptible flaws glare and vice versa. It's tough to make a coin truly look true to life, therefore it would be difficult if not impossible to have a true grading Ai... If a lens can't see it, or makes it too obvious, that will degrade the grading ability.

Cougar1978

Grading can be subjective. I like brilliant coins with super luster. Others may like tarnished or toned coins. One guy may tolerate black spots I don’t.

One Morgan dollar picked up in estate take it all or nothing deal a generic NGC MS63 the face looks MS65 but numerous small black spots both obv and reverse. From a marks pov plus it’s nice luster it looks higher grade. However overall - In that case the assigned TPG grade makes sense to me.

ricko

If AI reaches levels being discussed today, it is likely to become in general use. It is not there today (That I know of). I predict that because of the fact that if something can be measured (that covers all aspects, not just hits or scratches), then it can be defined and programmed, then assessed in totality, thereby the institution of AI grading. It will take time and technology growth. Cheers, RickO

yosclimber

@spyglassdesign said:

@jmlanzaf said:
"Eye appeal" is not exactly objective nor definable with an "optical conparator".

Anyone who thinks AI would be more consistent doesn't fully understand AI. The surface of every coin is unique. Grading isn't based on simply counting marks. Where they are located matters as well as how obvious they are. AI would always give the same result for the same coin. But it wouldn't necessarily interpret two different 65s as being 65s.

Not only that, the lens only picks up what it can see... A human can tilt and twist the coin around in the light that might reveal other good or bad factors that may not show up in a scan or image. This is the hardest part for me personally when imaging a coin... Sometimes imperceptible flaws glare and vice versa. It's tough to make a coin truly look true to life, therefore it would be difficult if not impossible to have a true grading Ai... If a lens can't see it, or makes it too obvious, that will degrade the grading ability.

This can be mostly handled by using multiple images with different lighting angles.

leothelyon

With 33 years in a series, I came to know/learned which coins/Jefferson nickels have the most details, the best strike, what could be expected from the earliest manufactured set of working hubs and dies, possibly from the very first set. Which coins (the PL ones) have the deepest mirrors to reflect/bounce back the greatest amount of luster/light into the eyes. Most collectors aren't savvy enough or even close to discerning the differences between a weak strike and a very early-die-state (VEDS) strike let alone having the ability to think/believe whether or not a paid 2nd opinion upholds a standard.
If they have the disposable funds, they can take it up with a grading company and think/believe that they will sort it all out for them. This has not always been the case and I won't go into that. If having the best coins in your collection is one's desire then I recommend they learn what to look for. Becoming/being savvy enough with a discerning eye on which coin grades higher takes years of study, even if it's just with one series. But one's accumulative grading skills can expand into any denomination and series once learned.

Leo

RollBuster

I really appreciate hearing everyone’s thoughts on this. It will be interesting to see if any developers take an interest in the challenge.

yosclimber

@RollBuster said:
I really appreciate hearing everyone’s thoughts on this. It will be interesting to see if any developers take an interest in the challenge.

It's been tried several times already.
How about some past threads on the subject:
2021 https://forums.collectors.com/discussion/1057670/watch-a-computer-grade-a-coin
2018 - 2020 https://forums.collectors.com/discussion/1042609/pcgs-launches-ai-powered-coin-grading-technology
2018 https://forums.collectors.com/discussion/1002282/thoughts-on-automated-coin-grading-ms66-or-ms67-or-au
2016 https://forums.collectors.com/discussion/958237/has-anyone-attempted-to-develop-computer-grading
2010 https://forums.collectors.com/discussion/767996/quot-the-big-one-quot-optical-or-laser-grading-of-coins
2005 https://forums.collectors.com/discussion/391282/whatever-happened-to-digital-computer-coin-grading
2005 https://forums.collectors.com/discussion/440322/pcgs-computer-grading-patent

RollBuster

Thank you

yspsales

@leothelyon said:
With 33 years in a series, I came to know/learned which coins/Jefferson nickels have the most details, the best strike, what could be expected from the earliest manufactured set of working hubs and dies, possibly from the very first set. Which coins (the PL ones) have the deepest mirrors to reflect/bounce back the greatest amount of luster/light into the eyes. Most collectors aren't savvy enough or even close to discerning the differences between a weak strike and a very early-die-state (VEDS) strike let alone having the ability to think/believe whether or not a paid 2nd opinion upholds a standard.
If they have the disposable funds, they can take it up with a grading company and think/believe that they will sort it all out for them. This has not always been the case and I won't go into that. If having the best coins in your collection is one's desire then I recommend they learn what to look for. Becoming/being savvy enough with a discerning eye on which coin grades higher takes years of study, even if it's just with one series. But one's accumulative grading skills can expand into any denomination and series once learned.

Leo

Nominate for Quote of 2023