Options
Vardera AI
Steve27
Posts: 13,281 ✭✭✭
I saw them advertising in The Numismatist, is this the future of coin grading?
"It's far easier to fight for principles, than to live up to them." Adlai Stevenson
0
Comments
For sure it is another method of separating the collector from his cash.
Whether it is worthwhile or not, not sure.
Collectors just need to learn how to grade. Increasingly they appear to not want to.
Grading has been less import for my collecting so far as I have focused on coins that are rare enough so that grades are less important.
Of course, that may change in the future so it's interesting to follow these developments.
"We accept photos from any device—even low-quality smartphone images" means the grade is going to be even more subjective than getting consensus among collectors with coin in-hand. Then comes the question of authenticity.
This is not the future. To be very accurate it would need extremely high res images at many angles for each coin. This could be useful as a first line filter for an online auction house to catch bad submissions.
Rare-Change.com - Low listing fee
Newbie collectors are getting sucked in by all these coin grading apps -- none of them are worth anything.
There was a discussion on this and several astute observers noted that to be accurate would require imaging the coin in 3D with it tilting and rotating while the light sources did their thing to be even remotely accurate with a single coin and it would need a massive database of coins imaged in the same way and compared to multiple legitimately graded (PCGS/NGC/CAC/ANACS) coins with the same treatment. So at this point it is a practical impossibility right now, but nothing stays impossible forever.
That would be a simple and useful step -- just do an overlay of known varieties. But don't expect EBay to shell out any cash to such an effort, they don't even take counterfeit reports seriously when the counterfeit is beyond obvious.
He said auction houses. Ebay is not an auction house.
Further, it would be useless for counterfeit detection of anything made with transfer dies and might well misidentify well worn real coins as counterfeits. It might even call everything fake that isn't photographed at the same angle.
All comments reflect the opinion of the author, even when irrefutably accurate.
Try to expand your thinking and not live in the tiny box of your own current knowledge base.
It would be completely useless anyplace that has anyone with half a brain, which leaves the EBay AI. If EBay sellers can't make decent photos then they should be kicked until they can -- it is NOT difficult. Any half wit with a cellphone can shoot a decent coin shot, but then if you are trying to obscure defects or faux errors like so many EBay sellers, yeah that is an issue.
And no, there are MANY EBay counterfeits that are simple to identify with a quick overlay, if you have minimal Photoshop skills.
You miss the point. It's not bad or good photos, it's photos at different angles or with different lighting. I never said anything about bad photos.
And my point was not addressed to the easy fakes but to false positives and false negatives which could well outnumber true positives.
I also never said it would be useless, in general. I said it would be useless with transfer die counterfeits.
Nothing that your giant box said addresses the issue of the statistical validity of its use, which needs to be considered. If it can only find easy fakes but has a high rate of false positives and negatives, it is simply not usable.
I'm not sure what you are arguing about. Nothing you said addresses any issue i raised. And nothing you said is even inconsistent with what I said.
Your box is apparently as tiny as mine. Congratulations or sorry. 🤣.
All comments reflect the opinion of the author, even when irrefutably accurate.
You are funny, which is why I cannot block you.
Disagree. There is a much higher percentage of people who cannot take a decent coin picture than the percentage of people who can.
Coin Photography
No one's ever said they didn't want to block me before. I don't know how to respond.
All comments reflect the opinion of the author, even when irrefutably accurate.
P.S. I think we made two different points, by the way.
All comments reflect the opinion of the author, even when irrefutably accurate.
I'm thinking more broadly. Could PCGS go out, purchase it an AI, train it on their database of coin images, and then use it to replace the graders? I think they would still have to have a Finalizer to make sure the AI wasn't hallucinating.
They could do something like that. I don't know anything about this particular product. But the key would be to have uniform, quality images.
All comments reflect the opinion of the author, even when irrefutably accurate.
I doubt it. They would likely need images at different angles and light reflections. This AI doesn't revue in a similar way to a human. Plus many fakes would be hard to detect with only images.
Rare-Change.com - Low listing fee
You're back in the box.
Look at the macro. The marketplace. When Joe collector sends his MS65 that hd paid MS65 retail, in to get his AI grade and it comes out MS63 that is not acceptable. Now transfer that scenario too an iconic rarity, for instance the 69 Large Cent. There is little chance that will AI grade 69.
That is why I think it will only be acceptable on moderns and with a different scale. That would be marketable and market acceptable. .
What you say is true. No question about it.
The bigger problem is that a coin is worth much less when its not slabbed.
Good deals can still be bought when raw so that works in a person's favor if they learn to grade.
Student of numismatics and collector of Morgan dollars
Successful BST transactions with: Namvet Justindan Mattniss RWW olah_in_MA
Dantheman984 Toyz4geo SurfinxHI greencopper RWW bigjpst bretsan MWallace logger7 JWP BruceS bigjpst
JWP PROMETHIUS88
Lol. I don't think i was ever in or out of the box.
There is no reason to assume that an AI system would not agree with current grades. You're in a different box. 😆 Many of my colleagues think they can recognize AI writing. They can't, except when clumsily done. You can get the AI to write exactly like you do, including idiosyncrasies and errors. There's no reason that an AI can't be trained to mimic visual subjectivity as well.
Even if it doesn't agree with current grading, it will likely happen anyway.
I'm sure no one in 1965 thought there would be such a plethora of grades. There was no 65 vs 66 much less 65+. Look at EAC grades vs PCGS grades, they don't align at all. The market adjusts to changes.
If you come over to my sandbox for a minute, what would it matter if you had AI graded coins co-existing with legacy human- graded coins? The market knows how to deal with ANACS grades and NGC grades and PCGS and CACG grades simultaneously, along with EAC grades. And, of course, with and without stickers. When Joe collector sends his 65 and it comes out a 63, it wouldn't matter as long as it consistently came out 63. The market would adjust.
All comments reflect the opinion of the author, even when irrefutably accurate.
I will call your market adjustability and raise you this scenario:
Joe collector is po'd his 63 is worth thousands less than his 65. Why would he entertain TPGAI?
Because the AI 63 would be worth exactly what the Human 65 is worth. It's exactly the same as an EAC XF being worth the same as a PCGS 62.
Why does Joe not have a fit now when his 64 sells for less than a 63 CAC?
All comments reflect the opinion of the author, even when irrefutably accurate.
I don't think they have enough images for more than a handful of coins. And many of the images they do have are internal and may not be of sufficient quality for a grading AI.
One of the things that people miss when they hear about image classifiers (which, if you think about it, is what grading is: here's a bunch of MS65 images, here's MS64, etc., classify THIS new image)... is that it's often opaque how they operate.
There are cases where people have decompiled the classifier and found it to focus on seemingly unremarkable minutiae.
ANA 50+ year/Life Member (now "Emeritus")
Author: 3rd Edition of the SampleSlabs book, https://sampleslabs.info/
That's a fair point, that the perceived value of an AI grading system could end up greater than human. That is why I say it would have to be a different scale for marketability purposes, to help differentiate perceived value.
Classic coins will be very noisy with results though, I don't think that can be overcome. Too expensive to accumulate data currently, still will have a pushback from collectors holding significant value material. Don't want to risk the AI grade. Moderns will work for sure, surprised if that is not in conception stages already.
We AGREE.
Moderns are 90+% of their volume, so if that's all it did, it's still a game changer.
All comments reflect the opinion of the author, even when irrefutably accurate.