Options
Artificial intelligence, will it be grading coins in the future?
crazyhounddog
Posts: 13,909 ✭✭✭✭✭
Think it’ll ever happen down the road or do you think coins will forever continue to be graded by humans? Hopefully we will find a solid solution to this task that won’t waver.
Thoughts?
The bitterness of "Poor Quality" is remembered long after the sweetness of low price is forgotten.
2
Comments
I think AI will be doing it for someone (that is someone will try it).
How accurately - I don't know. Many issues involved.
For a major TPG. Would have to be proven first and there is an investment in that. 3rd party?
I believe it would be initially best suited for variety attribution and counterfeit detection.
Then maybe some modern 68, 69, 70 grading type of thing. But still have some issues to work through.
https://youtube.com/watch?v=2YNufnS_kf4 - Mama I'm coming home ...................................................................................................................................................................... RLJ 1958 - 2023
All of my intelligence is artificial and I could be grading coins in the future.
No.
It sure will.
Let's hope not.
I will be dead before that happens, so I am not worried. Besides, AI was designed by humans, so how much superior can it be?
I'd rather see pigs fly
Custom album maker and numismatic photographer, see my portfolio here: (http://www.donahuenumismatics.com/).
I doubt it. May be useful for auction descriptions though.
With AI creeping into many facets of many markets I wouldn't be surprised if NGC, CACG and our host haven't already employed or tested the viability of using AI.
I'd be ok with AI used as a sanity check or 3rd level QA but not as the primary decision maker.
USAF (Ret.) 1985 - 2005. E-4B Aircraft Maintenance Crew Chief and Contracting Officer.
My current Registry sets:
✓ Everyman Mint State Carson City Morgan Dollars (1878 – 1893)
✓ Everyman Mint State Lincoln Cents (1909 – 1958)
✓ Morgan Dollar GSA Hoard (1878 – 1891)
Absolutely!
The subjective manner in which they have been graded is ripe for AI. Especially moderns by the bushel.
Look no further than the wild range of guesses from forum experts when a coin is posted without grade.
Calling @treybenedict
Trey's app is already doing AI generated grades for toning. It's a short step to doing it for grades.
Yes for sure
Machine learning that is. You just feed millions of data points in and the pc does the rest. Should be right around the corner.
No doubt in my mind.
Dave
The infrastructure needed to built these sorts of models for grading is a much larger case.
To be able to attempt this, grading standards need to be very specific and set. Given changing grades, crackout game etc. It’s a large task, but doable.
My estimate is 1-2 years out for a good model, and a team with enough resources that won’t burn capital like most AI startups right now.
Edit: I think it’s most useful for modern coins as vintage is a more difficult game to play. (Low range of datasets for very rare coins)
Its very possible with the current state of the world and the implementation of AI, our coins wont really matter anymore!
How do you figure?
Yes - possibly a self grading holder. You will be able buy them at Walmart. The holder will give bid /ask price too.
Yes. And with AI grading, a new grading company will be created. And a new forum monitored by AI.
"That coin only grades a 69."
That's because it's not really intelligence. It isn't alive. It's just more computer code in a computer program.
I thought AI had already been implemented once and failed for coin grading. And isn't it used for card grading?
Consider the fact that PCGS had already tried to do a scan/sniffer grading procedure in the past (like over 10 years ago?) Already has THE largest coin photograph data set (has taken far more photos of our coins way beyond published true views) And, that each grader cost them over $100,000 then the question is not if - but when?
WS
By the time AI is capable it will not matter. It will have already rid itself of the pesky humans.
https://smallcopperguy.com
The question infers machine learning not AI.
But humans are only temporary caretakers anyway.
For coins, the bigger question is will AIs collect them
I'm not sure, it's pretty hard to capture luster, hairlines, and bagmarks from a photo. AI could also have trouble with badly-struck buffalo nickels too.
Type collector, mainly into Seated. Young Numismatist. Good BST transactions with: mirabela, OKCC, MICHAELDIXON
I believe it’ll happen. When? I have no idea but I’m sure it will be implemented in the near future.
Great responses👍
I hope so, I'm tired of seeing inconsistently graded holdered coins. Right now they use two graders and a finalizer to judge each coin. I could easily see them go to a computer grade reviewed by a human finalizer.
I 100% agree. We need solid grades that are not subject to errors when graded. The human element needs to be removed.
I could see it within the next 5-10 years.
Successful transactions with: robkool, Walkerguy21D, JimW, Bruce7789, massscrew, Jinx86, jonasdenenbergllc, Yorkshireman, bobsr, tommyrusty7, markelman1125, Kliao, DBSTrader2, SurfinxHI, ChrisH821, CoinHoarder, Bolo, MICHAELDIXON, bigtime36, JWP, 1960NYGiants
I think it’s entirely possible and has great potential but as mentioned by others, there would be a tremendous amount of work involved to assemble the infrastructure necessary to pull it off. I might think this could be a very useful application of AI if done correctly.
AI is already being used on some analysis of medical and radiographic images. It’s a work in progress but making headway.
It’s far from perfect but certainly has great potential moving forward- with skilled human oversight (hopefully).
Happy, humble, honored and proud recipient of the “You Suck” award 10/22/2014
I believe the hobby will see AI used to determine surface preservation and assign a preliminary grade, perhaps determine if a coin is genuine. However, a human will determine the final grade based on eye appeal and originality.
AI grading ? How about counterfeit detecting. I'd be happy if AI could change the oil , or take care of honey- do lists.
It was tried years ago and failed BUT it’s going to happen. The question isn’t if but when. AI has made scary advancements. It is now capable of doing thinks today that couldn’t be done ten years ago. The advancements have caused serious concerns among some folks that know a lot about the development of AI. It has a self learning characteristic that I think at some point will take over grading in the future.
Be very careful what you wish for or even what may be contemplated as possible...
I doubt AI will work for several reasons. Seems programming and creating the intel to begin with would be challenging. Not all coins within a series are created equal and for AI to grade effectively, it will have to be programmed to recognize differences- a 1904 Morgan in 65 is simply not as attractive as an 1881-s at the same grade level. And there are countless other examples. The investment to even consider this would likely end the endeavor before starting.
I do not see this as viable. And in 10 years I suspect I will be proven wrong.
Experience the World through Numismatics...it's more than you can imagine.
I agree wholeheartedly with this assessment!
Very well put coinkat!
Just in the past month, in various articles extolling the virtues of AI for analysis of images, I spotted numerous blatant errors in the identification and labeling of anatomical structures. The authors (non medical) were passing these along as great examples of what AI might accomplish but they weren’t catching the mistakes. They all had stars in their eyes!
Happy, humble, honored and proud recipient of the “You Suck” award 10/22/2014
Considering the grades I receive; the graders are already using their artificial intelligence.
Or, could it be my intelligence?
here is my best guess as how they should do it ..
Have a computer that is tied to fixed cameras on the coins at many angles and have the top tier graders (many different ones) grade tons of different coins all while being recorded and all the data goes into an AI program. Do this for a couple years with all that data from the best graders going into the AI and I think you may have something ...
Also have the graders talk to the AI as they are grading on why they graded each coin a certain grade.
Hmmmmmm, 69 or 70.........caculating.......
"When they can't find anything wrong with you, they create it!"
I've waited all night for that picture. Thank you!
Custom album maker and numismatic photographer, see my portfolio here: (http://www.donahuenumismatics.com/).
I think it’s inevitable. Which will be especially interesting because grading parameters will constantly evolve, grades will be more fluid, and slabs - the plastic itself- will be less important.
Doggedly collecting coins of the Central American Republic.
Visit the Society of US Pattern Collectors at USPatterns.com.
And AI will undoubtedly create a new grading system.
No doubt there will be a role for humans to play in monitoring, correcting and teaching the machine. But as time passes, the machine will need less and less supervision, and will eventually be allowed to decide for itself which coins should be run by the humans before assigning a final grade, and which coins can be finalized immediately.
Doggedly collecting coins of the Central American Republic.
Visit the Society of US Pattern Collectors at USPatterns.com.
Yes with a human "finalizer". Certainly necessary for rarities.
Like some (many) I think it could be used very soon for moderns, classics will take longer but will likely happen at some point.
My Collection of Old Holders
Never a slave to one plastic brand will I ever be.
Absolutely, not even sure humans (other than an occasions QC check) will be needed. It will just depend on the economics on how quick it will be fully developed ( or adapted ) for the hobby. Just thinking - If facial recognition can pull a face out of crowd 100 feet away now ... why can it not find an MS65 PL Morgan sliding down a shoot at the PCGS grading room table?
I think it's inevitable. I'm in science and machine learning algorithms for large datasets have been "a thing" for over a decade. So has automating certain repetitive human endeavors... the bigger the datasets, the better the outcomes. I can envision a near future where a coin comes in for grading and is videotaped at many different angles... and AI "grader" assigns the grade and a human "finalizer" does QC and a reality check. As good as AI is for so.e things, it still currently needs a reality check.
I teach a graduate course in Scientific and Technical Writing a d the writing samples turned out by AI (like Chat GPT) are impressive, but the references they use still need to be checked... some references were legit, but others were either from opinion pieces (not peer-reviewed) and the worst cases showed the reference actually was arguing against the point the AI was trying to make.
Successful BST transactions with: SilverEagles92; Ahrensdad; Smitty; GregHansen; Lablade; Mercury10c; copperflopper; whatsup; KISHU1; scrapman1077, crispy, canadanz, smallchange, robkool, Mission16, ranshdow, ibzman350, Fallguy, Collectorcoins, SurfinxHI, jwitten, Walkerguy21D, dsessom.
We all disagree amongst ourselves as to grades. Look at the differences of opinion on any of the guess the grade threads. Grading includes eye appeal and AI cannot as of yet determine that. Besides, eye appeal is in large part dependent upon the likes of the viewer. We will always have individual preferences and AI will never be able to satisfy everyone. I believe that AI could certainly assign a technical grade but that desirability of any technical grade MS coin will vary greatly from coin to coin. To the extent eye appeal plays a role in the grade assignment, and in my view it should and does, human input will always be a part of grading - or at least the interpretation of the validity of any assigned grade.
Imagine if you will,
An optically aided AI computer analyzing millions of morgan dollars, considering the strike, every known alteration, counterfeit, patterns and types of toning. additions of chemicals. applying loss of grade value to a myriad of marks. hairlines or any variations to the original. A possible AI report might be:
"This is 4256 TO 4269th striking of an 1886 dollar of VAM 13, after the die repolishing at striking 5902-5933, with striking detail value of 89%, retains 98% of It's original luster, bag marking consistent with grade 63.5 to 63.7 of known specimens, Surface contamination value of .0039, current value is $42.67 wholesale, current buy offers $41.50/sells $43.25" Probable service grades, PCGS 63, NGC 64, CAC 64, ANACS 63
With enough ACCURATE data, anything is possible. J.P.