![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
|
#1
|
|||
|
|||
![]()
Title says it all. This has been talked about in the past, but do we have any confirmation PSA is now using AI to grade most cards?
I believe this to be true based on the amount of cards they are grading monthly these days. Look at the chart. This are PSA grading numbers since 2021. I just don’t see how humans could possibly grade all the cards per month. Sure, I think for the high dollar cards they get a human look, but for most, I think it is all AI. |
#2
|
|||
|
|||
![]()
I mean…seems entirely plausible to hire more people and tell them to work faster. If they can assign a grade in 30 seconds, then 20 seconds shouldn’t be that much harder.
I’m not convinced that the technology is really there to do much of the grading. Edited to add: I suspect AI is generally okay when it comes to creating text based on information that is available online or that the AI has been fed. But when it comes to visual stuff, I’m not convinced that it has any ability to really tell what it’s looking at.
__________________
Trying to wrap up my master mays set, with just a few left: 1968 American Oil left side 1971 Bazooka numbered complete panel Last edited by raulus; 09-16-2025 at 07:12 PM. |
#3
|
||||
|
||||
![]()
I really don't believe there is any AI grading in use by any reputable company. They do have more locations now, and that has to come with more grading staff, but doubtful there is anything more to it than that.
__________________
. Infuriating entitled old men since 2022...the eBay Authenticity Guarantee. #itouchmycards Last edited by CardPadre; 09-16-2025 at 06:43 PM. |
#4
|
||||
|
||||
![]()
Most of the cards are Pokemon, first of all. Next would be Ultra Modern. Very small percentage makes up vintage. They have over 200 graders in house just for Pokemon, if I heard Ryan Hoge correctly in his most recent interview.
__________________
( h @ $ e A n + l e y |
#5
|
|||
|
|||
![]()
Drew I tend to agree. I’m thinking most cards early 2000’s and up are getting graded by AI. Higher grading service levels I would assume are still all being graded by the hum eye.
|
#6
|
|||
|
|||
![]()
Even with 200, could they grade over 1.5 million cards with just humans? I don’t think so.
|
#7
|
||||
|
||||
![]()
Just imagine receivables unboxing process and time it takes for hundreds of thousands of cards/boxes per month…yes some of those millions of orders are bulk orders, but many are singles, too.
|
#8
|
|||
|
|||
![]()
How do they operate when it comes to the grading talent? Is it a 24 hour (or near) shift-operated operation or are graders working a normal 9-to-5 where the business puts in their 8-ish hours and shuts for the day?
__________________
‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾ ▪ Cubs 1800s-present HOF/stars/notables ▪ Cubs oversized type examples ▪ Cubs autographed cards ▪ |
#9
|
||||
|
||||
![]()
Zero chance.
They do not have assembly lines rolling cards through scanners and then having an AI agent do its thing. That is ludicrous. Humans are spending 5-10 seconds per card as usual and moving onto the next card. |
#10
|
|||
|
|||
![]()
1.6M cards per month. How many cards can a person grade in 1 day? If you assume 1 minute/card for 6 hours, that's 360 cards per grader per day. So that's 1,800/week, 7,500/month per grader. It's hard to see them any more efficient than that, and when you factor in unpacking, handling, packing, they probably aren't even that efficient.
1.6M cards / 7,500 would imply 213 graders. Even if I'm grossly overestimating efficiency, it seems like a manageable amount of people. Also, if you look at it in terms of revenue, say they get $20/card, if a guy is grading 7,500 cards per month, he/she is generating $150K in revenue. Seems about right. Does the 1.6M include reslabs? |
#11
|
|||
|
|||
![]()
They added two huge facilities since 2021, and have increased their workforce exponentially. The whole reason they did it was to handle the capacity backlog that you see here. They aren't using AI to grade, just built an insane amount of capacity over the last few years.
|
#12
|
||||
|
||||
![]() Quote:
I think this has a direct correlation with the so-called “stricter grading”. I have no proof, just the upper management experience to believe that they have to have a strong hourly goals guideline to push these numbers that causes half-assedness in your employees to meet impossible expectations. They likely are also tracked in assigned grades. The leadership undoubtedly understands the devaluation that excessive high grades causes to submission of specific cards. If they give away too many 10s or high grades in vintage the return on grade gambling submissions will wane. I could see coaching of graders occurring if they were to assign too high of a percentage of modern 10s or vintage high grades no matter what cards they have in their queue. There is no shortage of “Dilbert” based management in any corporation. Fear for job security with silly obscene goal standards is always more likely to cause poorer job performance and cutting corners than better output. Just my opinion based on corporate experience for wayyyyy to long.
__________________
- Justin D. Player collecting - Lance Parrish, Jim Davenport, John Norlander. Successful B/S/T with - Highstep74, Northviewcats, pencil1974, T2069bk, tjenkins, wilkiebaby11, baez578, Bocabirdman, maddux31, Leon, Just-Collect, bigfish, quinnsryche...and a whole bunch more, I stopped keeping track, lol. |
#13
|
|||
|
|||
![]() Quote:
|
#14
|
||||
|
||||
![]()
Do any former PSA graders (but no longer PSA graders) have a non-compete provision linked to their employment? If not, why don't some of them show up here on the forum and answer our questions about their (PSA) procedures used while they were employed.
I know the forum is populated with many self-proclaimed experts, but hearing the evidence straight from the horse's mouth might be enlightening.
__________________
RAUCOUS SPORTS CARD FORUM MEMBER AND MONSTER FATHER. GOOD FOR THE HOBBY AND THE FORUM WITH A VAULT IN AN UNDISCLOSED LOCATION FILLED WITH WORTHLESS NON-FUNGIBLES 274/1000 Monster Number |
#15
|
||||
|
||||
![]()
Just curious, would there be something wrong if AI technology did the grading? Think about it, how much worse can it be than it is now?
A company would have to get the technology consistent and have the ability to detect reprints and alterations. Also, there really is no "industry standard" for grading although the TPGs are fairly consistent in their criteria. . . .
__________________
fr3d c0wl3s - always looking for OJs and other 19th century stuff. PM or email me if you have something cool you're looking to find a new home for. |
#16
|
||||
|
||||
![]() Quote:
__________________
Be sure to subscribe to my YouTube Channel, The Stuff Of Greatness. New videos are uploaded every week... https://www.youtube.com/@tsogreatness/videos Last edited by perezfan; 09-18-2025 at 10:45 AM. |
#17
|
||||
|
||||
![]() Quote:
![]() What gets to me though is that droves of collectors keep sending their cards off to be glanced at for less than ten seconds by disinterested employees to whom the job is simple drudgery. ![]()
__________________
That government governs best that governs least. Last edited by Balticfox; 09-18-2025 at 11:21 AM. |
#18
|
||||
|
||||
![]() Quote:
![]()
__________________
That government governs best that governs least. Last edited by Balticfox; 09-18-2025 at 11:22 AM. |
#19
|
|||
|
|||
![]() Quote:
Sent from my SM-S906U using Tapatalk |
#20
|
|||
|
|||
![]() Quote:
Sent from my SM-S906U using Tapatalk |
#21
|
|||
|
|||
![]()
Quite honestly, you wouldn't need AI to grade...just known guidelines for what constitutes a knock down in grade and how to weigh those findings vs each other.
Those variables and that that tech is already here, but how you implement it is the issue. I imagine you would need an ability to either move the card under a camera and light source or use multiple lights to detect surface issues, but so much of the other stuff could be done with a simple image. AI would be good (if proficient) for ID'ing a card type and the player associated with it as compared to what it was submitted as...
__________________
‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾ ▪ Cubs 1800s-present HOF/stars/notables ▪ Cubs oversized type examples ▪ Cubs autographed cards ▪ |
#22
|
||||
|
||||
![]()
AI is not good enough right now to grade. Still needs humans, they just keep hiring graders which increases output.
|
#23
|
||||
|
||||
![]()
The more graders a TPG has, the greater the variability there will be on grades going out the door. You can train them all to the same standards, but there is still some subjectivity (from day-to-day with the same grader and from grader to grader on any given day) in the grading process, and that is what leads to variability.
I know they are not slabbing cards, but Dean's is using machines to do their grading and are pretty open about how it works. There is a pretty long explanation on it, including some videos here. If Dean's has been doing it since 2019, any reputable TPG could do it also. But the variability in grading is what drives re-submits and the push to ratchet-up a card's grade. And those re-submits equal $$. Imagine if a collector was certain that resubmitting a card would only result in the same grade again and again (unless Kurt gets involved).
__________________
Working Sets: Baseball- T206 SLers - Virginia League (-1) 1952 Topps - low numbers (-1) 1953 Topps (-60) 1954 Bowman (-3) 1964 Topps Giants auto'd (-2) |
#24
|
||||
|
||||
![]() Quote:
.
__________________
Leon Luckey www.luckeycards.com |
#25
|
|||
|
|||
![]()
This morning, I heard that AI can pass the CPA test in minutes. You don't believe AI can grade cards? Every video that I have watched on Pokemon cards come back as 10's and 9's.
https://www.cnbc.com/2025/09/24/ai-c...tes-study.html
__________________
Successful Transactions: camaro69, dhicks67, Dr Orange, Ed_Hutchinson, jingram058, LACardsGuy, perezfan |
#26
|
||||
|
||||
![]()
Dean, the link in your post takes me to an article that says AI can pass the CFA (Chartered Financial Analyst) Exam in minutes, not the CPA (Certified Public Accountant) Exam.
__________________
Seeking very scarce/rare cards for my Sam Rice master collection, e.g., E210 York Caramel Type 2 (upgrade), 1931 W502, W504 (upgrade), W572 sepia, W573, 1922 Haffner's Bread, 1922 Keating Candy, 1922 Witmor Candy Type 2 (vertical back), 1926 Sports Co. of Am. with ad & blank backs. Also 1917 Merchants Bakery & Weil Baking cards of WaJo. Also E222 cards of Lipe, Revelle & Ryan. |
#27
|
|||
|
|||
![]() Quote:
People still don't seem to understand that AI is really not the right term. For the most part what people are referring to as AI is simply a search engine on steroids that then reports back what it found in a small bite. It's not "thinking" or "intelligent" It's only as good as the data it's given to pull from, and it needs TONS of it to even approach accuracy and even then I think we can all point to plenty of examples where AI is so dumb that it basically just makes something up because it misinterprets, or relies on bad data. the need for MASSIVE amounts of data make AI grading a tough proposition. I won't say impossible because things advance so quickly it. There may be some break through in the next few years that circumvents the challenges I mentioned. Of course then Skynet is my worry, not card grading. ![]()
__________________
Check out https://www.thecollectorconnection.com Always looking for consignments 717.327.8915 We sell your less expensive pre-war cards individually instead of in bulk lots to make YOU the most money possible! and Facebook: https://www.facebook.com/thecollectorconnectionauctions Last edited by Aquarian Sports Cards; 09-24-2025 at 06:43 PM. |
#28
|
||||
|
||||
![]()
PSA has been using AI and scanning technologies to support graders since at least 2021 when they acquired Genamint. They have competitors like TAG that rely almost exclusively on scanning and AI. It's only effective on ultramodern and some modern. TCG surpassed sports cards by submitted volume some time ago.
|
#29
|
|||
|
|||
![]() Quote:
My apologies. I heard CPA and that is what stuck in my head. In the Agriculture world there are machines that scan and blow puffs of air to keep grains of rice deemed cosmetically unacceptable from going forward in the process. I believe someone out there can figure out a way to scan cards to speed up the grading time.
__________________
Successful Transactions: camaro69, dhicks67, Dr Orange, Ed_Hutchinson, jingram058, LACardsGuy, perezfan |
#30
|
||||
|
||||
![]() Quote:
__________________
( h @ $ e A n + l e y |
#31
|
|||
|
|||
![]() Quote:
__________________
Successful Transactions: camaro69, dhicks67, Dr Orange, Ed_Hutchinson, jingram058, LACardsGuy, perezfan |
#32
|
|||
|
|||
![]()
Thanks for the link to Travis' original post. I had forgetten about his analysis which brings up a number of good points
|
#33
|
|||
|
|||
![]() Quote:
![]() |
#34
|
||||
|
||||
![]() Quote:
In fact if you cut the connection of an AI bot to the internet, I'm sure I would have destroyed any such bot in a CFA exam back when I was young and had a mind like a razor. ![]()
__________________
That government governs best that governs least. Last edited by Balticfox; 09-30-2025 at 10:32 AM. |
#35
|
|||
|
|||
![]()
I'm sure that most any AI could also pass the CPA exam. You only need 75% on each of the 4 sections to pass, which is a pretty weak score, considering my lowest score was 98.
As someone mentioned above, the AI has the advantage of basically being open book, because it can refer to source materials with all the answers. The CPA exam does not require much (or any really) in the way of critical thinking and reasoning. The vast majority of the questions are multiple choice, and are easily identifiable even by a fairly dumb machine with ready access to the source material. In theory, any decent computer with access to the source material shouldn't take more than a few seconds to blaze through the questions and answer them correctly, and that's probably moving slow. But passing a text-based exam based on having access to the source material is a long ways away from actually being able to grade cards.
__________________
Trying to wrap up my master mays set, with just a few left: 1968 American Oil left side 1971 Bazooka numbered complete panel Last edited by raulus; 09-29-2025 at 04:51 PM. |
#36
|
||||
|
||||
![]()
I'm pretty sure that AI (even with all it's hoopla) has absolutely zero options for actually "looking" at a card in order to take a guess on the grade.
__________________
My Monster Progess Complete Set......: 238 / 520 : 45% HOF Cards..........: 009 / 076 : 12% Southern League.: 000 / 048 : 00% Minor League......: 055 / 086 : 41% Portrait Cards......: 077 / 180 : 43% Horizontal Cards.: 000 / 006 : 00% |
#37
|
||||
|
||||
![]()
PSA isn't doing this...yet...but TAG is already doing it on cards produced after 1989. They have computers, not humans, grading corners, edges, centering, surface, and dimensions following human authentication. Here are some screenshots showing how to read their grading reports. Nat Turner has indicated in interviews that PSA is investing in similar technologies and they already use AI to help their graders. That said, they bought Genamint and SGC...so they could just buy TAG too.
|
#38
|
|||
|
|||
![]()
Patent application per Paul Lesko tweet, sounds like using AI to me. By the way, if they can do certain things to be more efficient I am not opposed to it. They scan every card, so data has been built to have authentic cards in a system to help make decisions.
![]() Sent from my iPhone using Tapatalk
__________________
BST h2oya311, Jobu, Shoeless Moe, Bumpus Jones, Frankish, Shoeless Moe again, Maddux31, Billycards, sycks22, ballparks, VintageBen (for a friend), vpina87, JimmyC, scmavl, BigFanNY, Bliggity, bluespruce, powell_am Last edited by Schlesinj; 10-02-2025 at 05:36 AM. |
#39
|
|||
|
|||
![]()
Oversimplification here, but I would say yes.
https://www.net54baseball.com/showth...ghlight=August 1.66 million graded in August. Now, let's just say with August being a 31-day month you subtract four weekends which leaves 23 days. I know this falls under the old saying "If ifs and buts were candy and nuts...." but I will let Don Meredith rest in peace and if you want to listen to the song, it is up to you. Now for the math: 1,660,000 divided by 184 hours (23 eight-hour workdays) Equals 8,695.65 per hour Now, with AI there is no work limit. Quality control......hmmm. Time and weekends off at PSA? Thus, the calculations could be way off. |
#40
|
||||
|
||||
![]()
My take:
If today's AI can generate Tilly Norwood, then it surely can pick out a number between 1 and 10 to represent the condition of a card and tell if has been altered or not.
__________________
Working Sets: Baseball- T206 SLers - Virginia League (-1) 1952 Topps - low numbers (-1) 1953 Topps (-60) 1954 Bowman (-3) 1964 Topps Giants auto'd (-2) |
![]() |
|
|
![]() |
||||
Thread | Thread Starter | Forum | Replies | Last Post |
FS: Lot of 14 Low Grade M116 Sporting Life Cards *ALL CARDS HAVE BEEN SOLD* | marcdelpercio | Pre-WWII cards (E, D, M, etc..) B/S/T | 0 | 01-23-2024 02:38 PM |
FS: Group of 8 Lower Grade T207 Recruit Cards *ALL CARDS HAVE BEEN SOLD* | marcdelpercio | Tobacco (T) cards, except T206 B/S/T | 1 | 01-11-2024 12:06 PM |
Grade or not grade rare boxing cards | PWAN | Boxing / Wrestling Cards & Memorabilia Forum | 3 | 03-29-2022 07:40 AM |
WTB low grade Type Cards T209/T210/T211/T213, C46, T3 & candy cards 1900's-1919 | 25801wv | Tobacco (T) cards, except T206 B/S/T | 0 | 10-23-2012 11:43 AM |
% of Low Grade vs. High Grade Trading Cards | Archive | Net54baseball Vintage (WWII & Older) Baseball Cards & New Member Introductions | 24 | 12-09-2007 07:16 AM |