@MrEureka said:
The most intriguing part about AI grading is that we will be told up front that the standards will be constantly evolving. Scary in a way, but also liberating in a way.
it won't work though. 100 MS 65 Morgan Dollars all have different elements of deviation from perfection.
Similarly, that is why the self driving car is virtually bust. Too many factors to process.
There are self driving cars on the road now. And they will continue to improve. If you used the internet in 1980, it looked nothing like it did in 1990 much less today.
You are an educator. Can your position be replaced successfully by AI?
If not now, how many more years?
Ironically, i told my class this morning that some of my function could already be replaced. I used ChatGPT to ask specific content questions that we had discussed this week. The result was a well structured description of what we had talked about in class. Now, my replaceability goes down as the complexity of the material goes up... for now.
What really changes, ultimately, is the necessary skills for a student. For example, there is little need to either teach or test on facts that your cell phone can provide. There is, however, a need to properly query the AI to get good responses as well as critical evaluation skills required to assess the response.
I beg my students to use AI for homework and learning. I insist they use it to help write papers. I implore them to consider how their career will be defined by their ability to use AI.
But AI controls what is learned. We can't post here using examples of the way Google and the press block or change actual facts in politics, Covid, etc. Thus, it can change grading in any way it wants. And who needs AI computer grading it there will be an evolving lack of standardization? Give me PCGS!
That actually doesn't change anything that i said. Try to not use the internet and see how you perform relative to your peers.
The application developer also has the ability to use the AI as an API and therefore has control over how the LLM is used and what information it receives. So, for example, PCGS AI would largely be in the hands of PCGS not open AI. If standards charged, it would be because the TPGS wanted the change.
Personally, i don't think AI grading will be any better than human grading. It might be cheaper and faster. It might be equal in precision. But it has the same basic problems that limits human grading, primarily the uniqueness of each coin.
I don’t necessarily disagree with the sentiment of your comment, but you didn’t mention anything about the potentially remarkable level of consistency that would be possible with AI.
@MrEureka said:
The most intriguing part about AI grading is that we will be told up front that the standards will be constantly evolving. Scary in a way, but also liberating in a way.
it won't work though. 100 MS 65 Morgan Dollars all have different elements of deviation from perfection.
Similarly, that is why the self driving car is virtually bust. Too many factors to process.
There are self driving cars on the road now. And they will continue to improve. If you used the internet in 1980, it looked nothing like it did in 1990 much less today.
You are an educator. Can your position be replaced successfully by AI?
If not now, how many more years?
Ironically, i told my class this morning that some of my function could already be replaced. I used ChatGPT to ask specific content questions that we had discussed this week. The result was a well structured description of what we had talked about in class. Now, my replaceability goes down as the complexity of the material goes up... for now.
What really changes, ultimately, is the necessary skills for a student. For example, there is little need to either teach or test on facts that your cell phone can provide. There is, however, a need to properly query the AI to get good responses as well as critical evaluation skills required to assess the response.
I beg my students to use AI for homework and learning. I insist they use it to help write papers. I implore them to consider how their career will be defined by their ability to use AI.
But AI controls what is learned. We can't post here using examples of the way Google and the press block or change actual facts in politics, Covid, etc. Thus, it can change grading in any way it wants. And who needs AI computer grading it there will be an evolving lack of standardization? Give me PCGS!
That actually doesn't change anything that i said. Try to not use the internet and see how you perform relative to your peers.
The application developer also has the ability to use the AI as an API and therefore has control over how the LLM is used and what information it receives. So, for example, PCGS AI would largely be in the hands of PCGS not open AI. If standards charged, it would be because the TPGS wanted the change.
Personally, i don't think AI grading will be any better than human grading. It might be cheaper and faster. It might be equal in precision. But it has the same basic problems that limits human grading, primarily the uniqueness of each coin.
I don’t necessarily disagree with the sentiment of your comment, but you didn’t mention anything about the potentially remarkable level of consistency that would be possible with AI.
That's because I'm not sure. Lol. It depends on the AI and it depends on what you mean by consistency. AI has come to be shorthand for LLMs which is really only one kind of AI. it's long been possible to create an algorithm that spits out the same number for the same coin. The problem has been to get consistency between different coins of the same grade. The LLMs have a different logic that incorporates probability. I don't know how consistent they would be. I'm not saying they won't be. I just don't know.
That said, I'm incredibly bullish that AIwill prove to be very disruptive Technology. And the Nobel committee was equally bullish.
@Married2Coins said:
Consistency suggests an unchanging standard that eliminates change. Therefore consistency in grading does not exist. Has it ever?
Of course not. But even though AI will constantly improve its skills and therefore keep changing its standard, the degree of those changes should continually decrease until the standard becomes close to fixed, and it’s grading nearly perfectly consistent. For better or worse.
Andy Lustig
Doggedly collecting coins of the Central American Republic.
Visit the Society of US Pattern Collectors at USPatterns.com.
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
@Married2Coins said:
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
I don't think it's quite that simple. It is more likely that PCGS has its AI, NGC has its AI, etc. So you'll likely still have variation between TPGS's. But you may have greater consistency within the TPGS, at least until or unless the TPGS "evolves" its protocols.
The developers of the AI models create the basic functionality that individual app developers call for their specific application. As such, it is unlikely you will see one global AI coin standard.
@MrEureka said:
The most intriguing part about AI grading is that we will be told up front that the standards will be constantly evolving. Scary in a way, but also liberating in a way.
it won't work though. 100 MS 65 Morgan Dollars all have different elements of deviation from perfection.
Similarly, that is why the self driving car is virtually bust. Too many factors to process.
There are self driving cars on the road now. And they will continue to improve. If you used the internet in 1980, it looked nothing like it did in 1990 much less today.
You are an educator. Can your position be replaced successfully by AI?
If not now, how many more years?
Ironically, i told my class this morning that some of my function could already be replaced. I used ChatGPT to ask specific content questions that we had discussed this week. The result was a well structured description of what we had talked about in class. Now, my replaceability goes down as the complexity of the material goes up... for now.
What really changes, ultimately, is the necessary skills for a student. For example, there is little need to either teach or test on facts that your cell phone can provide. There is, however, a need to properly query the AI to get good responses as well as critical evaluation skills required to assess the response.
I beg my students to use AI for homework and learning. I insist they use it to help write papers. I implore them to consider how their career will be defined by their ability to use AI.
But AI controls what is learned. We can't post here using examples of the way Google and the press block or change actual facts in politics, Covid, etc. Thus, it can change grading in any way it wants. And who needs AI computer grading it there will be an evolving lack of standardization? Give me PCGS!
That actually doesn't change anything that i said. Try to not use the internet and see how you perform relative to your peers.
The application developer also has the ability to use the AI as an API and therefore has control over how the LLM is used and what information it receives. So, for example, PCGS AI would largely be in the hands of PCGS not open AI. If standards charged, it would be because the TPGS wanted the change.
Personally, i don't think AI grading will be any better than human grading. It might be cheaper and faster. It might be equal in precision. But it has the same basic problems that limits human grading, primarily the uniqueness of each coin.
I don’t necessarily disagree with the sentiment of your comment, but you didn’t mention anything about the potentially remarkable level of consistency that would be possible with AI.
That's because I'm not sure. Lol. It depends on the AI and it depends on what you mean by consistency. AI has come to be shorthand for LLMs which is really only one kind of AI. it's long been possible to create an algorithm that spits out the same number for the same coin. The problem has been to get consistency between different coins of the same grade. The LLMs have a different logic that incorporates probability. I don't know how consistent they would be. I'm not saying they won't be. I just don't know.
That said, I'm incredibly bullish that AIwill prove to be very disruptive Technology. And the Nobel committee was equally bullish.
What I mean by “consistency” is a system that incorporates a table or harness for the coin to tilt and rotate in front of a camera lens. HD pics and videos that let the camera see the coin like we do, depth mapping or lidar or something like that to measure wear. IDK the technical parts, you’re the scientist here 🤷♂️.
As the data pool grows and as the machine learns from any human intervention, it makes adjustments on its own. After enough time, the data pool is robust and the machine is dialed in.
Not to mention the software would be able to recognize a coin on a repeat submission, identifying based on contact marks and what have you. One would be able to circumvent that with some creativity, but for the most part it would work well with classic coins.
@MrEureka said:
The most intriguing part about AI grading is that we will be told up front that the standards will be constantly evolving. Scary in a way, but also liberating in a way.
it won't work though. 100 MS 65 Morgan Dollars all have different elements of deviation from perfection.
Similarly, that is why the self driving car is virtually bust. Too many factors to process.
There are self driving cars on the road now. And they will continue to improve. If you used the internet in 1980, it looked nothing like it did in 1990 much less today.
You are an educator. Can your position be replaced successfully by AI?
If not now, how many more years?
Ironically, i told my class this morning that some of my function could already be replaced. I used ChatGPT to ask specific content questions that we had discussed this week. The result was a well structured description of what we had talked about in class. Now, my replaceability goes down as the complexity of the material goes up... for now.
What really changes, ultimately, is the necessary skills for a student. For example, there is little need to either teach or test on facts that your cell phone can provide. There is, however, a need to properly query the AI to get good responses as well as critical evaluation skills required to assess the response.
I beg my students to use AI for homework and learning. I insist they use it to help write papers. I implore them to consider how their career will be defined by their ability to use AI.
But AI controls what is learned. We can't post here using examples of the way Google and the press block or change actual facts in politics, Covid, etc. Thus, it can change grading in any way it wants. And who needs AI computer grading it there will be an evolving lack of standardization? Give me PCGS!
That actually doesn't change anything that i said. Try to not use the internet and see how you perform relative to your peers.
The application developer also has the ability to use the AI as an API and therefore has control over how the LLM is used and what information it receives. So, for example, PCGS AI would largely be in the hands of PCGS not open AI. If standards charged, it would be because the TPGS wanted the change.
Personally, i don't think AI grading will be any better than human grading. It might be cheaper and faster. It might be equal in precision. But it has the same basic problems that limits human grading, primarily the uniqueness of each coin.
I don’t necessarily disagree with the sentiment of your comment, but you didn’t mention anything about the potentially remarkable level of consistency that would be possible with AI.
That's because I'm not sure. Lol. It depends on the AI and it depends on what you mean by consistency. AI has come to be shorthand for LLMs which is really only one kind of AI. it's long been possible to create an algorithm that spits out the same number for the same coin. The problem has been to get consistency between different coins of the same grade. The LLMs have a different logic that incorporates probability. I don't know how consistent they would be. I'm not saying they won't be. I just don't know.
That said, I'm incredibly bullish that AIwill prove to be very disruptive Technology. And the Nobel committee was equally bullish.
What I mean by “consistency” is a system that incorporates a table or harness for the coin to tilt and rotate in front of a camera lens. HD pics and videos that let the camera see the coin like we do, depth mapping or lidar or something like that to measure wear. IDK the technical parts, you’re the scientist here 🤷♂️.
As the data pool grows and as the machine learns from any human intervention, it makes adjustments on its own. After enough time, the data pool is robust and the machine is dialed in.
Not to mention the software would be able to recognize a coin on a repeat submission, identifying based on contact marks and what have you. One would be able to circumvent that with some creativity, but for the most part it would work well with classic coins.
The advantage of recognizing resubmissions would not be to avoid changing the grade, because changing standards will be part of the system design. The advantage will be recognizing changes in the coin itself, which could mean a coin has been doctored or that it’s a counterfeit. In some cases, those items will be kicked back to humans for further study.
Andy Lustig
Doggedly collecting coins of the Central American Republic.
Visit the Society of US Pattern Collectors at USPatterns.com.
Great discussion, everyone. I have some thoughts to share after I've had time to consider where all this is heading.
I shared my thoughts on last week's The E-Sylum, I would append my thoughts on where AI grading is headed below:
"In short, AI is now just two steps away from surpassing human graders. The next step is AI being able to analyze a live video feed. The final step is integrating live video into a robotics platform. The AI would be able to analyze a coin in real-time. Like a human, it can dynamically rotate a coin to see all angles. But beyond a human, it can:
Adjust to any lighting conditions.
Quickly snap from 2x to 20x lens to inspect specific spots.
Switch to different viewing modes like LiDAR for 3D scanning.
X-rays are used to determine the composition of counterfeits and identify them.
And so much more
At some point the narrative will shift from "AI can now mimic a human grader" to "AI is better than a human grader in every way".
I will note that my underlying assumption is AI using technical grading standards. Market grading is a whole other beast."
@1madman said:
Humor me please and run Dan’s beaver coin through the AI analysis. I’m curious to know what it thinks the grade is.
@CaptHenway said:
As the first TPG Professional Grader, I would expect to see a lot more sample grading before you say that the system works. We did 3+ months of testing before we started accepting coins.
Agreed! This was just an initial test to sanity check that an AI could even identify and analyze a coin through video. Significantly more testing must be done across coin series and different grades. The key point I am making is that AI made a major step in mimicking the conditions of a human grader.
Also, what were T-Rex's like back then?
@CRHer700 said:
Also, didn't Compugrade go under for a reason?
For those interested, here is an online post discussing the history of Compugrade. I've heard that they also struggled with differentiating between wear and a weak strike.
@leothelyon said:
If there is one area of coin grading the cgc's have failed to master for over 25+ years is determining/calculating the strength/completion of the strike/details implementing that criteria, a must/important factor to receive a MS66 and higher grade.
I'm still building up my knowledge of computer vision but I wonder if something like LiDAR can be used to detect strike strength?
@PeakRarities said:
Readers will look back at this thread 20 years ago and laugh. It’s not a matter of “if”, but “when”. If the goal is only to match the “accuracyne” of human graders, ai technology's is already more than capable of doing so.
The issue is that when the technology is applied with the proper data sets, it would expose the current system for what it is. To be abundantly clear, I use the word “expose” not because it’s some big conspiracy, just that the system is inherently flawed, as most of us know. The first respectable entity to implement an AI system will state, authoritatively, that “Our system is the arbiter of truth, and thus, only our grade is the coins true and everlasting grade.” 🤴🏼
This presents a bit of a conundrum for PCGS, with respect to optics. They would effectively have to invalidate the system that served as the foundational bedrock, upon which their entire business model was built. They’ll have to be thoughtful and delicate in their approach.
I've been thinking a lot about the consequences for the TPGs with the emergence of AI grading. It's may impact their business model in many ways. But I've also considered whether the TPGs are even incentivized to adopt AI grading.
I believe that, eventually, the TPGs will use AI to help increase efficiencies in areas like counterfeit detection. But overall, I don't see them replacing human graders anytime soon.
There are a few reasons why I think this way:
The TPGs value is not grading. They have billion-dollar valuations because of their reputation and the liquidity they provide to the marketplace. They make it easier for dealers to buy and sell coins. It just so happens that this liquidity is provided through their grading services. It doesn't matter that imperfect human graders are used. It works and works well today. There's no need for them to rock the boat.
They have no incentive to normalize AI grading. Right now, the TPGs are differentiated through their reputation with collectors. If AI grading becomes normalized, the market may shift to focus on how accurate their AI algorithms are as the main differentiating factor. TPGs are not tech powerhouses. Hiring AI/ML specialists to run an ever-spiraling arms race toward the perfect model will take a lot of time and money. It's much safer to avoid this altogether. There is the risk that the TPGs all end up with nearly perfect models. So, people will just submit to the cheapest grader. Or even just grade at home.
There is the risk that collectors will balk at the idea. Many dealers remember Compugrade and their "AI" models from the 1990s, and they aren't remembered fondly. It's not worth the reputational risk.
The TPGs have very little to gain financially from using AI. Why risk the company's reputation to save a few million on graders? What is the upside moneywise?
AI grading would destroy all the money the TPGs get from the breaking slabs and resubmitting game. Why bother resubmitting when you know the AI will remember the same coin and give the same grade?
Last but not least is market vs technical grading. Most TPGs do market grading, which requires a deep understanding of the market for each specific coin series. A VF Large Cent in the 1840s may have shifted to XF today because of market sentiments. AI can and will be great at technical grading. However, AI does not have a "gut feeling" about what the market would accept as the "right" grade. Only what a coin can technically grade as given objective criteria.
Note how I did not mention much about technical limitations. It's mostly market forces at play. This was the lesson I learned from my work with AI and coins.
@percyb said:
Thanks much for your AI insights on grading.
Can it identify counterfeit coins?
I have not tested it for counterfeiting, but exploring it would be interesting!
Thank you for running the gold coin through the AI analysis. So NGC graded it a 50, and AI says 50-53. I think AI is telling Dan to submit it to cac in the current holder because it’s got a high likelihood of a gold sticker.
@1madman said:
Thank you for running the gold coin through the AI analysis. So NGC graded it a 50, and AI says 50-53. I think AI is telling Dan to submit it to cac in the current holder because it’s got a high likelihood of a gold sticker.
Winner winner beaver dinner?
This would be a pretty interesting use case for AI grading. Use it to scan slabs and get suggestions on which candidates are suitable for stickering.
@Married2Coins said:
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
I don't think it's quite that simple. It is more likely that PCGS has its AI, NGC has its AI, etc. So you'll likely still have variation between TPGS's. But you may have greater consistency within the TPGS, at least until or unless the TPGS "evolves" its protocols.
The developers of the AI models create the basic functionality that individual app developers call for their specific application. As such, it is unlikely you will see one global AI coin standard.
You are making the assumption that there will not be only one AI grading service that solves the problem. IMO. it would bet both stupid and do nothing to change our modern "Wild West" of five major grading services with different standards with or without each having an AI System.
@Married2Coins said:
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
I don't think it's quite that simple. It is more likely that PCGS has its AI, NGC has its AI, etc. So you'll likely still have variation between TPGS's. But you may have greater consistency within the TPGS, at least until or unless the TPGS "evolves" its protocols.
The developers of the AI models create the basic functionality that individual app developers call for their specific application. As such, it is unlikely you will see one global AI coin standard.
You are making the assumption that there will not be only one AI grading service that solves the problem. IMO. it would bet both stupid and do nothing to change our modern "Wild West" of five major grading services with different standards with or without each having an AI System.
I am predicting that they will all eventually have an AI system because they will have to in order to compete. It would be stupid for them to not develop the technology and let their competitor dominate the market.
There is a reason that they all use plastic slabs - they didn't originally. They all have websites, online submission, online photos, + grades, etc.
If the use case is established and the market accepts, you cannot have one standard for one coin and one standard for another. In other words the program is unemotional. Therefore many of the iconic market graded coins that have received rarity or eye appeal (per date or series) grades will end up downgraded. Is that acceptable?
The incentive is to mechanize and computerize the evaluation process. Shorter turnaround times, higher quantity of submissions, more active trading.
IMO, There will NEVER be a "set-in-stone" grade for at least 50% of coins because of all the greedy hands in the pie, evolving standards, changing market conditions, and subjectivity brought about by experience, lighting, magnification, and personal interest - AI (programing) or no AI. Additionally, it seems AI on its own will revert to a more strict, technical type of grading that lessens the effect of net grading up or down and no coin dealer will stand for that (grade bump for color & eye appeal vs net grade for problems).
The coins that will have the most standardization as they do now (AI or no AI) will be the Ag -VF's. Human graders at CACG already have an effect on formerly MS and and straight graded coins.
Where are the present and ex TPG on this forum adding their opinions? Where is Rexford, J.P Martin, and others who might post here and have not identified themselves? I'd like to read what they say and what PCGS thinks about AI. After all they tried it in the past. Won't AI grading slow the turnaround times even more? Will prices rise.
@Overdate said:
Is AI capable of evaluating all the nuances of grading? Can it accurately spot the difference between a 69 and a 70? Or a straight grade and a "details" grade? Or the four flavors of VF? Or wear vs. weak strike? Or natural vs. artificial toning? Or proof vs. DMPL? Or business strike vs. SMS vs. Satin Finish?
I expect that AI grading for some coins is feasible now, and will continue to improve over time. But I doubt that it's capable of replacing human graders just yet.
IMO, the line between 69 &70 is the easiest line in the world for a computer to define because any defect (either a tiny nick, mint made strike through, or spot) would kill the perfect grade. Unfortunately, that would be too critical compared to today's commercial "eye-ball grading" and dealers would hate it.
A friend sent a MS-70 1986 SE in for review with a nick on the letters (a $40 coin when graded long ago and now worth much more) and was told that was the TPGS standard for 70! I'd bet this has happened to others here.
I got that ! Both PCGS and NGC have graded many many MS and Pf 70 coins with post strike marks visible to the naked eye. Grading of bulk coins has victimized collectors in more ways than one. Just look at how silly long their turnaround time is on classic coins that originally was their bread and butter.
AFAIK all the grading services allow tiny Mint-made-as struck" defects on "perfect 70's. I am in 100% agreement with that. The problem is that I've been told that SE are graded quickly by naked eye. One or more tiny nicks can be missed during this type of examination. When I buy a SE for a gift, I closely examine it using a hand lens as most savvy collectors do. I don't want any PMD on a coin I buy. I will buy coins with mint defects (especially strike-throughs) for myself for the price of a normal coin because they are neat!
@Married2Coins said:
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
I don't think it's quite that simple. It is more likely that PCGS has its AI, NGC has its AI, etc. So you'll likely still have variation between TPGS's. But you may have greater consistency within the TPGS, at least until or unless the TPGS "evolves" its protocols.
The developers of the AI models create the basic functionality that individual app developers call for their specific application. As such, it is unlikely you will see one global AI coin standard.
You are making the assumption that there will not be only one AI grading service that solves the problem. IMO. it would bet both stupid and do nothing to change our modern "Wild West" of five major grading services with different standards with or without each having an AI System.
It is my understanding that the THREE major grading services all use the same set of "Standards," The Official ANA Grading standards as modified by myself and others over the years, but that there may be slightly different interpretations of those standards. I am not familiar with what standards the other grading services use.
TD
Numismatist. 50 year member ANA. Winner of four ANA Heath Literary Awards; three Wayte and Olga Raymond Literary Awards; Numismatist of the Year Award 2009, and Lifetime Achievement Award 2020. Winner numerous NLG Literary Awards.
@Married2Coins said:
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
I don't think it's quite that simple. It is more likely that PCGS has its AI, NGC has its AI, etc. So you'll likely still have variation between TPGS's. But you may have greater consistency within the TPGS, at least until or unless the TPGS "evolves" its protocols.
The developers of the AI models create the basic functionality that individual app developers call for their specific application. As such, it is unlikely you will see one global AI coin standard.
You are making the assumption that there will not be only one AI grading service that solves the problem. IMO. it would bet both stupid and do nothing to change our modern "Wild West" of five major grading services with different standards with or without each having an AI System.
It is my understanding that the THREE major grading services all use the same set of "Standards," The Official ANA Grading standards as modified by myself and others over the years, but that there may be slightly different interpretations of those standards. I am not familiar with what standards the other grading services use.
TD
Then you would be wrong, the ANA standards are quite outdated and untechnical. They form the foundation of the scale at the two major TPGs but are not relied upon in any meaningful capacity, because they don’t account for strike.
IMO, There will NEVER be a "set-in-stone" grade for at least 50% of coins because of all the greedy hands in the pie, evolving standards, changing market conditions, and subjectivity brought about by experience, lighting, magnification, and personal interest - AI (programing) or no AI. Additionally, it seems AI on its own will revert to a more strict, technical type of grading that lessens the effect of net grading up or down and no coin dealer will stand for that (grade bump for color & eye appeal vs net grade for problems).
The coins that will have the most standardization as they do now (AI or no AI) will be the Ag -VF's. Human graders at CACG already have an effect on formerly MS and and straight graded coins.
Where are the present and ex TPG on this forum adding their opinions? Where is Rexford, J.P Martin, and others who might post here and have not identified themselves? I'd like to read what they say and what PCGS thinks about AI. After all they tried it in the past. Won't AI grading slow the turnaround times even more? Will prices rise.
If AI is supposed to standardize grading, how the heck will that happen UNLESS EACH TPGS USES THE SAME SETUP?
Who said AI is supposed to standardize grading? Not me.
AI is a tool that many people use to help with their applications. It is not a single entity. PCGS will use AI. NGC will use AI. What they use it for and how is up to their developers.
It is all but impossible for NGC and PCGS to use the same AI application. They may not even use the same AI tool within their application.
@Married2Coins said:
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
I don't think it's quite that simple. It is more likely that PCGS has its AI, NGC has its AI, etc. So you'll likely still have variation between TPGS's. But you may have greater consistency within the TPGS, at least until or unless the TPGS "evolves" its protocols.
The developers of the AI models create the basic functionality that individual app developers call for their specific application. As such, it is unlikely you will see one global AI coin standard.
You are making the assumption that there will not be only one AI grading service that solves the problem. IMO. it would bet both stupid and do nothing to change our modern "Wild West" of five major grading services with different standards with or without each having an AI System.
It is my understanding that the THREE major grading services all use the same set of "Standards," The Official ANA Grading standards as modified by myself and others over the years, but that there may be slightly different interpretations of those standards. I am not familiar with what standards the other grading services use.
TD
Then you would be wrong, the ANA standards are quite outdated and untechnical. They form the foundation of the scale at the two major TPGs but are not relied upon in any meaningful capacity, because they don’t account for strike.
THen what ARE the published standards that each grading service is using?
Numismatist. 50 year member ANA. Winner of four ANA Heath Literary Awards; three Wayte and Olga Raymond Literary Awards; Numismatist of the Year Award 2009, and Lifetime Achievement Award 2020. Winner numerous NLG Literary Awards.
@Married2Coins said:
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
I don't think it's quite that simple. It is more likely that PCGS has its AI, NGC has its AI, etc. So you'll likely still have variation between TPGS's. But you may have greater consistency within the TPGS, at least until or unless the TPGS "evolves" its protocols.
The developers of the AI models create the basic functionality that individual app developers call for their specific application. As such, it is unlikely you will see one global AI coin standard.
You are making the assumption that there will not be only one AI grading service that solves the problem. IMO. it would bet both stupid and do nothing to change our modern "Wild West" of five major grading services with different standards with or without each having an AI System.
It is my understanding that the THREE major grading services all use the same set of "Standards," The Official ANA Grading standards as modified by myself and others over the years, but that there may be slightly different interpretations of those standards. I am not familiar with what standards the other grading services use.
TD
Then you would be wrong, the ANA standards are quite outdated and untechnical. They form the foundation of the scale at the two major TPGs but are not relied upon in any meaningful capacity, because they don’t account for strike.
THen what ARE the published standards that each grading service is using?
As the saying goes, time to replace the mustard stains with precision robots. Just think after the AI grade we can have a special double AI cacs approved bean. These boomers will lose their mind, we will have them paying $10K for a $50 Morgan. LOL. THKS!
@blitzdude said:
As the saying goes, time to replace the mustard stains with precision robots. Just think after the AI grade we can have a special double AI cacs approved bean. These boomers will lose their mind, we will have them paying $10K for a $50 Morgan. LOL. THKS!
You got some kind of prejudice against Baby Boomers?
Numismatist. 50 year member ANA. Winner of four ANA Heath Literary Awards; three Wayte and Olga Raymond Literary Awards; Numismatist of the Year Award 2009, and Lifetime Achievement Award 2020. Winner numerous NLG Literary Awards.
@blitzdude said:
As the saying goes, time to replace the mustard stains with precision robots. Just think after the AI grade we can have a special double AI cacs approved bean. These boomers will lose their mind, we will have them paying $10K for a $50 Morgan. LOL. THKS!
You got some kind of prejudice against Baby Boomers?
Not at all, boomers are my parents and I respect them.....well about half of them anyways. RGDS!
@blitzdude said:
As the saying goes, time to replace the mustard stains with precision robots. Just think after the AI grade we can have a special double AI cacs approved bean. These boomers will lose their mind, we will have them paying $10K for a $50 Morgan. LOL. THKS!
You got some kind of prejudice against Baby Boomers?
Not at all, boomers are my parents and I respect them.....well about half of them anyways. RGDS!
Does not sound like respect, dismissing an entire group as though they were all the same.
Numismatist. 50 year member ANA. Winner of four ANA Heath Literary Awards; three Wayte and Olga Raymond Literary Awards; Numismatist of the Year Award 2009, and Lifetime Achievement Award 2020. Winner numerous NLG Literary Awards.
@Married2Coins said:
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
I don't think it's quite that simple. It is more likely that PCGS has its AI, NGC has its AI, etc. So you'll likely still have variation between TPGS's. But you may have greater consistency within the TPGS, at least until or unless the TPGS "evolves" its protocols.
The developers of the AI models create the basic functionality that individual app developers call for their specific application. As such, it is unlikely you will see one global AI coin standard.
You are making the assumption that there will not be only one AI grading service that solves the problem. IMO. it would bet both stupid and do nothing to change our modern "Wild West" of five major grading services with different standards with or without each having an AI System.
It is my understanding that the THREE major grading services all use the same set of "Standards," The Official ANA Grading standards as modified by myself and others over the years, but that there may be slightly different interpretations of those standards. I am not familiar with what standards the other grading services use.
TD
It is my understanding that major dealers NEVER used the ANA Grading standards. However, me, most of my friends, and probably a majority of small local dealers did use the ANA Standards. Anyway, IMO, if the ANA Grading Service had done its job correctly and their standards would have fit the prevailing commercial coin market, ANA MS-65's would not have become MS-63's and the FIVE major grading services along with several others possibly would have never come into being.
BTW TD, thanks for your contributions to our hobby.
IMO, There will NEVER be a "set-in-stone" grade for at least 50% of coins because of all the greedy hands in the pie, evolving standards, changing market conditions, and subjectivity brought about by experience, lighting, magnification, and personal interest - AI (programing) or no AI. Additionally, it seems AI on its own will revert to a more strict, technical type of grading that lessens the effect of net grading up or down and no coin dealer will stand for that (grade bump for color & eye appeal vs net grade for problems).
The coins that will have the most standardization as they do now (AI or no AI) will be the Ag -VF's. Human graders at CACG already have an effect on formerly MS and and straight graded coins.
Where are the present and ex TPG on this forum adding their opinions? Where is Rexford, J.P Martin, and others who might post here and have not identified themselves? I'd like to read what they say and what PCGS thinks about AI. After all they tried it in the past. Won't AI grading slow the turnaround times even more? Will prices rise.
If AI is supposed to standardize grading, how the heck will that happen UNLESS EACH TPGS USES THE SAME SETUP?
Who said AI is supposed to standardize grading? Not me.
AI is a tool that many people use to help with their applications. It is not a single entity. PCGS will use AI. NGC will use AI. What they use it for and how is up to their developers.
It is all but impossible for NGC and PCGS to use the same AI application. They may not even use the same AI tool within their application.
Then who needs AI? What is AI suppose to do to improve grading if not to add precision (grading the same coin the same way every time it is seen)? All you need is a camera and a computer system to "fingerprint" coins. Aren't diamonds graded that way now?
@Married2Coins said:
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
I don't think it's quite that simple. It is more likely that PCGS has its AI, NGC has its AI, etc. So you'll likely still have variation between TPGS's. But you may have greater consistency within the TPGS, at least until or unless the TPGS "evolves" its protocols.
The developers of the AI models create the basic functionality that individual app developers call for their specific application. As such, it is unlikely you will see one global AI coin standard.
You are making the assumption that there will not be only one AI grading service that solves the problem. IMO. it would bet both stupid and do nothing to change our modern "Wild West" of five major grading services with different standards with or without each having an AI System.
It is my understanding that the THREE major grading services all use the same set of "Standards," The Official ANA Grading standards as modified by myself and others over the years, but that there may be slightly different interpretations of those standards. I am not familiar with what standards the other grading services use.
TD
Then you would be wrong, the ANA standards are quite outdated and untechnical. They form the foundation of the scale at the two major TPGs but are not relied upon in any meaningful capacity, because they don’t account for strike.
THen what ARE the published standards that each grading service is using?
Beats my five aces but I'll bet it is not the ANA Grading Guide. ______________________________________________________________________________________________________________________
I'm a nobody
The ANA changed the grading standards by replacing, "Full Strike with "As struck" in 2006 or 2007". IMO, this change came about due to the kick-backs on complaints from buyers that the coins they bought had flawed strikes. And so, the CGC's didn't want to be liable for that anymore. And submissions were down. Why else would they change the ANA grading standards?
But there are far more collectors of labels than there are those who actually study the strike and luster that meets their strict standards. And CGC's need to survive with coin submissions. It takes time for collectors to read a coin's surface and the details but a few do eventually graduate to seeking higher quality coins and they can't wait for that to happen.
Leo
The more qualities observed in a coin, the more desirable that coin becomes!
IMO, There will NEVER be a "set-in-stone" grade for at least 50% of coins because of all the greedy hands in the pie, evolving standards, changing market conditions, and subjectivity brought about by experience, lighting, magnification, and personal interest - AI (programing) or no AI. Additionally, it seems AI on its own will revert to a more strict, technical type of grading that lessens the effect of net grading up or down and no coin dealer will stand for that (grade bump for color & eye appeal vs net grade for problems).
The coins that will have the most standardization as they do now (AI or no AI) will be the Ag -VF's. Human graders at CACG already have an effect on formerly MS and and straight graded coins.
Where are the present and ex TPG on this forum adding their opinions? Where is Rexford, J.P Martin, and others who might post here and have not identified themselves? I'd like to read what they say and what PCGS thinks about AI. After all they tried it in the past. Won't AI grading slow the turnaround times even more? Will prices rise.
If AI is supposed to standardize grading, how the heck will that happen UNLESS EACH TPGS USES THE SAME SETUP?
Who said AI is supposed to standardize grading? Not me.
AI is a tool that many people use to help with their applications. It is not a single entity. PCGS will use AI. NGC will use AI. What they use it for and how is up to their developers.
It is all but impossible for NGC and PCGS to use the same AI application. They may not even use the same AI tool within their application.
Then who needs AI? What is AI suppose to do to improve grading if not to add precision (grading the same coin the same way every time it is seen)? All you need is a camera and a computer system to "fingerprint" coins. Aren't diamonds graded that way now?
You aren't understanding what "AI" is or how it is used.
TPGS's will use AI in their operations because it makes their operations more accurate and efficient. But the implementation of the technology is on an individual use basis determined by the developers.
Again, there is not a single "AI" entity that everyone will use to do everything. There is AI technology that people are using to develop applications. There are already multiple LLMs (ChatGPT, Gemini, etc) as well as different AI models under development that are not LLMs at all.
You don't "need" a phone... unless you want efficient communication. You don't need to use AI unless you want to increase efficiency and, possibly, improve accuracy. But you are not ceding to an omnipotent entity. You are adapting technology to an application.
Think of "AI" as a computer. You can do a million different things with a computer. The computer does not force users to all use it the same way to get the same result. If provides functionality that any user can adapt to help with the task at hand. My father used it to catalog his movie collection. That didn't give all of us the same catalog with the same cataloging format. AI is a technology not an application. You still need to create the application and NGC can create an application that gives different results than PCGS.
It is also quite possible that even the PCGS application will continue to have some variance in the same coin due to imaging variances. Do all True Views represent the coin equally well? And variations in toning and luster between different coins will likely still create different results from similar coins.
Since AI grades from images (stills + video under rotation) doesn’t that mean AI can grade without being in the same room as the coin itself?
(that would save on postage)
And if,say, PCGS, NGC, and CACG standards can be made into strings of computer code, then couldn’t, from images, a 3-part report be compiled, such as:
GRADES by
PCGS. standards 66.5
NGC standards 66.8
CACG standards 65.9
I’m Just spitballing here, seeing if anything sticks…
30+ years coin shop experience (ret.) Coins, bullion, currency, scrap & interesting folks. Loved every minute!
Since AI grades from images (stills + video under rotation) doesn’t that mean AI can grade without being in the same room as the coin itself?
(that would save on postage)
And if,say, PCGS, NGC, and CACG standards can be made into strings of computer code, then couldn’t, from images, a 3-part report be compiled, such as:
GRADES by
PCGS. standards 66.5
NGC standards 66.8
CACG standards 65.9
I’m Just spitballing here, seeing if anything sticks…
You need to standardize the images or you could easily deceive, Eben without outright photoshop fraud. Lighting alone can make an XF coin look UNC and vice versa. I would never trust a grade arrived at from images alone.
Yes, there would have be a standard for lighting, angles, and rotation in place. That’s the relatively easy part.
Just the act of creating such a thing would be a major advance in showing coins for sale or auction, or enjoying your coins at home, or sharing them to your friends. You’d buy the coin, and a well-lit and attractive video representation would come with it. The video could be ‘interactive’ in that you could pause it at any angle of rotation/ lighting.
I’m sure other features would also be dreamed up. Think of the whole new level of GTGs we could have here!
But I’m sure the auction houses are already working on this aspect of coin presentation. If they do a good enough job, actual humans - professional and hobbyists - could grade remotely. Or just decide with more confidence if that’s the coin for them!
But I digress.
This thread is still about AI actually learning to grade coins better than a human.
30+ years coin shop experience (ret.) Coins, bullion, currency, scrap & interesting folks. Loved every minute!
I just came across a fascinating idea to address the issue of market grading for AI.
You could train an AI using images of a coin + the sold price. There is no need to mention the coin's mintage, history, etc. The market has already spoken through past sales.
The AI could learn a coin's desirableness [luster, strike, etc.] purely by how much it has sold for compared to how it appears.
My only hesitation is that the AI would be trained with photos. It may not work if the AI is trained with images, but we then ask it to analyze coins using videos. It wouldn't be a true one-to-one comparison.
I like where you're going with this, AlbumNerd, imho it seems like access to data is the limiting factor; that is, AI is only as good as the info available to it, it might get very accurate with millions of images to compare, and eventually it would be even better with video or other input, the possibilities are endless
I've spent the better part of 34 years trying to get the CGC's and everyone else how the strike of a coin is the most important factor in determining a coin's grade....what makes anyone think computer-grading will be programmed any different/better?
Leo
The more qualities observed in a coin, the more desirable that coin becomes!
If you have time to make a short list for us. What will AI be able to do for a TPGS:
AI will be able to identify a coin that it has seen before. (So will an image of the coin stored in a database. I know that ANACS weighed each coin an took a picture of it in the beginning.)
AI will be able to detect fake toning? (Doubtful, humans can't although they think they can.)
AI will be able to measure the reflectivity of luster and depth of PL better and more consistent than humans.(I'm sold!)
@Married2Coins said:
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
I don't think it's quite that simple. It is more likely that PCGS has its AI, NGC has its AI, etc. So you'll likely still have variation between TPGS's. But you may have greater consistency within the TPGS, at least until or unless the TPGS "evolves" its protocols.
The developers of the AI models create the basic functionality that individual app developers call for their specific application. As such, it is unlikely you will see one global AI coin standard.
You are making the assumption that there will not be only one AI grading service that solves the problem. IMO. it would bet both stupid and do nothing to change our modern "Wild West" of five major grading services with different standards with or without each having an AI System.
It is my understanding that the THREE major grading services all use the same set of "Standards," The Official ANA Grading standards as modified by myself and others over the years, but that there may be slightly different interpretations of those standards. I am not familiar with what standards the other grading services use.
TD
Then you would be wrong, the ANA standards are quite outdated and untechnical. They form the foundation of the scale at the two major TPGs but are not relied upon in any meaningful capacity, because they don’t account for strike.
THen what ARE the published standards that each grading service is using?
Published standards? There aren’t even really any standards at all. All there really is a general approach, That’s why it’s possible to grade things we’ve never even seen before, Including some things we didn’t even know existed.
Andy Lustig
Doggedly collecting coins of the Central American Republic.
Visit the Society of US Pattern Collectors at USPatterns.com.
If you have time to make a short list for us. What will AI be able to do for a TPGS:
AI will be able to identify a coin that it has seen before. (So will an image of the coin stored in a database. I know that ANACS weighed each coin an took a picture of it in the beginning.)
AI will be able to detect fake toning? (Doubtful, humans can't although they think they can.)
AI will be able to measure the reflectivity of luster and depth of PL better and more consistent than humans.(I'm sold!)
What else?
5.
6.
AI will be able to HELP with everything. Again, it is a tool. And it is a tool that works 24/7. It doesn't have to be better. If might not have to even be 100% equivalent. It just needs to be close.
AI MAY be able to better detect fake toning, authenticity, luster, strike, etc. But not by itself. The application has to be developed. A bad implementation could be worse.
Consider strike. A human looks at the coin and gets an impression based on experience. The AI on the other hand COULD (depends on implementation) look at a billion data points from across the surface and seen from a dozen angles...in seconds. And if properly told what to look for, it will collate all that data and make a reproducible judgment.
And don't forget that AI is improving exponentially. ChatGPT is more accurate and more functional than it was a year ago. In 5 years or 10 years...
@MrEureka said:
The most intriguing part about AI grading is that we will be told up front that the standards will be constantly evolving. Scary in a way, but also liberating in a way.
it won't work though. 100 MS 65 Morgan Dollars all have different elements of deviation from perfection.
>
Similarly, that is why the self driving car is virtually bust. Too many factors to process.
There are self driving cars on the road now. And they will continue to improve. If you used the internet in 1980, it looked nothing like it did in 1990 much less today.
Doc has got it right. AI will never master the nuances of properly grading coins. It's time to dump the Sheldon numbers for grading a coin. The numbers game which has been plaguing the coin collecting hobby for many years now has gotten even more ridiculous with AI.
Great spirits have always encountered violent opposition from mediocre minds.-Albert Einstein
@MrEureka said:
The most intriguing part about AI grading is that we will be told up front that the standards will be constantly evolving. Scary in a way, but also liberating in a way.
it won't work though. 100 MS 65 Morgan Dollars all have different elements of deviation from perfection.
>
Similarly, that is why the self driving car is virtually bust. Too many factors to process.
There are self driving cars on the road now. And they will continue to improve. If you used the internet in 1980, it looked nothing like it did in 1990 much less today.
Doc has got it right. AI will never master the nuances of properly grading coins. It's time to dump the Sheldon numbers for grading a coin. The numbers game which has been plaguing the coin collecting hobby for many years now has gotten even more ridiculous with AI.
Lmfao
Your argument isn't even internally consistent.
AI is not being used (yet) to grade coins, so how can it have made assuring more ridiculous.
You don't think AI will matter the nuances of grading at the same time that you want to dump the standard grading scale.
Doc has got it right. AI will never master the nuances of properly grading coins. It's time to dump the Sheldon numbers for grading a coin. The numbers game which has been plaguing the coin collecting hobby for many years now has gotten even more ridiculous with AI.
The Sheldon numbers are simply a proxy for price/value. If you replace them with another 1-dimensional grade, it will serve the same purpose. If you replace it with a 4-dimensional grade (surfaces, luster, eye-appeal, strike), you now have to establish how to project that into the 1-dimensional price/value domain, because at the end of the day, that's where every transaction is.
Comments
I don’t necessarily disagree with the sentiment of your comment, but you didn’t mention anything about the potentially remarkable level of consistency that would be possible with AI.
Founder- Peak Rarities
Website
Instagram
Facebook
Consistency suggests an unchanging standard that eliminates change. Therefore consistency in grading does not exist. Has it ever?
Nope.
That's because I'm not sure. Lol. It depends on the AI and it depends on what you mean by consistency. AI has come to be shorthand for LLMs which is really only one kind of AI. it's long been possible to create an algorithm that spits out the same number for the same coin. The problem has been to get consistency between different coins of the same grade. The LLMs have a different logic that incorporates probability. I don't know how consistent they would be. I'm not saying they won't be. I just don't know.
That said, I'm incredibly bullish that AIwill prove to be very disruptive Technology. And the Nobel committee was equally bullish.
Of course not. But even though AI will constantly improve its skills and therefore keep changing its standard, the degree of those changes should continually decrease until the standard becomes close to fixed, and it’s grading nearly perfectly consistent. For better or worse.
Doggedly collecting coins of the Central American Republic.
Visit the Society of US Pattern Collectors at USPatterns.com.
I hope so. What you posted makes sense. I was thinking it would always change but as it learns it will get closer to consistent "perfection." Besides, if it is accepted, it will be the standard. Not like the old ANA Grading Standard that most dealers did not accept.
If some AI tells me the coin is ugly and the forum here comes to the same conclusion, then it has truly arrived.
10-4,
My Instagram picturesErik
My registry sets
I don't think it's quite that simple. It is more likely that PCGS has its AI, NGC has its AI, etc. So you'll likely still have variation between TPGS's. But you may have greater consistency within the TPGS, at least until or unless the TPGS "evolves" its protocols.
The developers of the AI models create the basic functionality that individual app developers call for their specific application. As such, it is unlikely you will see one global AI coin standard.
What I mean by “consistency” is a system that incorporates a table or harness for the coin to tilt and rotate in front of a camera lens. HD pics and videos that let the camera see the coin like we do, depth mapping or lidar or something like that to measure wear. IDK the technical parts, you’re the scientist here 🤷♂️.
As the data pool grows and as the machine learns from any human intervention, it makes adjustments on its own. After enough time, the data pool is robust and the machine is dialed in.
Not to mention the software would be able to recognize a coin on a repeat submission, identifying based on contact marks and what have you. One would be able to circumvent that with some creativity, but for the most part it would work well with classic coins.
Founder- Peak Rarities
Website
Instagram
Facebook
The advantage of recognizing resubmissions would not be to avoid changing the grade, because changing standards will be part of the system design. The advantage will be recognizing changes in the coin itself, which could mean a coin has been doctored or that it’s a counterfeit. In some cases, those items will be kicked back to humans for further study.
Doggedly collecting coins of the Central American Republic.
Visit the Society of US Pattern Collectors at USPatterns.com.
Great discussion, everyone. I have some thoughts to share after I've had time to consider where all this is heading.
I shared my thoughts on last week's The E-Sylum, I would append my thoughts on where AI grading is headed below:
"In short, AI is now just two steps away from surpassing human graders. The next step is AI being able to analyze a live video feed. The final step is integrating live video into a robotics platform. The AI would be able to analyze a coin in real-time. Like a human, it can dynamically rotate a coin to see all angles. But beyond a human, it can:
Adjust to any lighting conditions.
Quickly snap from 2x to 20x lens to inspect specific spots.
Switch to different viewing modes like LiDAR for 3D scanning.
X-rays are used to determine the composition of counterfeits and identify them.
And so much more
At some point the narrative will shift from "AI can now mimic a human grader" to "AI is better than a human grader in every way".
I will note that my underlying assumption is AI using technical grading standards. Market grading is a whole other beast."
Here is a video recording testing Google Gemini Pro 002 and the results
https://vimeo.com/1020439104
Agreed! This was just an initial test to sanity check that an AI could even identify and analyze a coin through video. Significantly more testing must be done across coin series and different grades. The key point I am making is that AI made a major step in mimicking the conditions of a human grader.
Also, what were T-Rex's like back then?
For those interested, here is an online post discussing the history of Compugrade. I've heard that they also struggled with differentiating between wear and a weak strike.
I'm still building up my knowledge of computer vision but I wonder if something like LiDAR can be used to detect strike strength?
I've been thinking a lot about the consequences for the TPGs with the emergence of AI grading. It's may impact their business model in many ways. But I've also considered whether the TPGs are even incentivized to adopt AI grading.
I believe that, eventually, the TPGs will use AI to help increase efficiencies in areas like counterfeit detection. But overall, I don't see them replacing human graders anytime soon.
There are a few reasons why I think this way:
The TPGs value is not grading. They have billion-dollar valuations because of their reputation and the liquidity they provide to the marketplace. They make it easier for dealers to buy and sell coins. It just so happens that this liquidity is provided through their grading services. It doesn't matter that imperfect human graders are used. It works and works well today. There's no need for them to rock the boat.
They have no incentive to normalize AI grading. Right now, the TPGs are differentiated through their reputation with collectors. If AI grading becomes normalized, the market may shift to focus on how accurate their AI algorithms are as the main differentiating factor. TPGs are not tech powerhouses. Hiring AI/ML specialists to run an ever-spiraling arms race toward the perfect model will take a lot of time and money. It's much safer to avoid this altogether. There is the risk that the TPGs all end up with nearly perfect models. So, people will just submit to the cheapest grader. Or even just grade at home.
There is the risk that collectors will balk at the idea. Many dealers remember Compugrade and their "AI" models from the 1990s, and they aren't remembered fondly. It's not worth the reputational risk.
The TPGs have very little to gain financially from using AI. Why risk the company's reputation to save a few million on graders? What is the upside moneywise?
AI grading would destroy all the money the TPGs get from the breaking slabs and resubmitting game. Why bother resubmitting when you know the AI will remember the same coin and give the same grade?
Last but not least is market vs technical grading. Most TPGs do market grading, which requires a deep understanding of the market for each specific coin series. A VF Large Cent in the 1840s may have shifted to XF today because of market sentiments. AI can and will be great at technical grading. However, AI does not have a "gut feeling" about what the market would accept as the "right" grade. Only what a coin can technically grade as given objective criteria.
Note how I did not mention much about technical limitations. It's mostly market forces at play. This was the lesson I learned from my work with AI and coins.
I have not tested it for counterfeiting, but exploring it would be interesting!
Thank you for running the gold coin through the AI analysis. So NGC graded it a 50, and AI says 50-53. I think AI is telling Dan to submit it to cac in the current holder because it’s got a high likelihood of a gold sticker.
Winner winner beaver dinner?
This would be a pretty interesting use case for AI grading. Use it to scan slabs and get suggestions on which candidates are suitable for stickering.
You are making the assumption that there will not be only one AI grading service that solves the problem. IMO. it would bet both stupid and do nothing to change our modern "Wild West" of five major grading services with different standards with or without each having an AI System.
Nice.
I am predicting that they will all eventually have an AI system because they will have to in order to compete. It would be stupid for them to not develop the technology and let their competitor dominate the market.
There is a reason that they all use plastic slabs - they didn't originally. They all have websites, online submission, online photos, + grades, etc.
Two takeaways:
If the use case is established and the market accepts, you cannot have one standard for one coin and one standard for another. In other words the program is unemotional. Therefore many of the iconic market graded coins that have received rarity or eye appeal (per date or series) grades will end up downgraded. Is that acceptable?
The incentive is to mechanize and computerize the evaluation process. Shorter turnaround times, higher quantity of submissions, more active trading.
And the beat goes on...
IMO, There will NEVER be a "set-in-stone" grade for at least 50% of coins because of all the greedy hands in the pie, evolving standards, changing market conditions, and subjectivity brought about by experience, lighting, magnification, and personal interest - AI (programing) or no AI. Additionally, it seems AI on its own will revert to a more strict, technical type of grading that lessens the effect of net grading up or down and no coin dealer will stand for that (grade bump for color & eye appeal vs net grade for problems).
The coins that will have the most standardization as they do now (AI or no AI) will be the Ag -VF's. Human graders at CACG already have an effect on formerly MS and and straight graded coins.
Where are the present and ex TPG on this forum adding their opinions? Where is Rexford, J.P Martin, and others who might post here and have not identified themselves? I'd like to read what they say and what PCGS thinks about AI. After all they tried it in the past. Won't AI grading slow the turnaround times even more? Will prices rise.
PS @jmlanzaf
If AI is supposed to standardize grading, how the heck will that happen UNLESS EACH TPGS USES THE SAME SETUP?
I got that ! Both PCGS and NGC have graded many many MS and Pf 70 coins with post strike marks visible to the naked eye. Grading of bulk coins has victimized collectors in more ways than one. Just look at how silly long their turnaround time is on classic coins that originally was their bread and butter.
AFAIK all the grading services allow tiny Mint-made-as struck" defects on "perfect 70's. I am in 100% agreement with that. The problem is that I've been told that SE are graded quickly by naked eye. One or more tiny nicks can be missed during this type of examination. When I buy a SE for a gift, I closely examine it using a hand lens as most savvy collectors do. I don't want any PMD on a coin I buy. I will buy coins with mint defects (especially strike-throughs) for myself for the price of a normal coin because they are neat!
It is my understanding that the THREE major grading services all use the same set of "Standards," The Official ANA Grading standards as modified by myself and others over the years, but that there may be slightly different interpretations of those standards. I am not familiar with what standards the other grading services use.
TD
Then you would be wrong, the ANA standards are quite outdated and untechnical. They form the foundation of the scale at the two major TPGs but are not relied upon in any meaningful capacity, because they don’t account for strike.
Who said AI is supposed to standardize grading? Not me.
AI is a tool that many people use to help with their applications. It is not a single entity. PCGS will use AI. NGC will use AI. What they use it for and how is up to their developers.
It is all but impossible for NGC and PCGS to use the same AI application. They may not even use the same AI tool within their application.
THen what ARE the published standards that each grading service is using?
Their own standards.
As the saying goes, time to replace the mustard stains with precision robots. Just think after the AI grade we can have a special double AI cacs approved bean. These boomers will lose their mind, we will have them paying $10K for a $50 Morgan. LOL. THKS!
The whole worlds off its rocker, buy Gold™.
You got some kind of prejudice against Baby Boomers?
Not at all, boomers are my parents and I respect them.....well about half of them anyways. RGDS!
The whole worlds off its rocker, buy Gold™.
Does not sound like respect, dismissing an entire group as though they were all the same.
As long as grading standards evolve / change, it really doesn't matter who or what is grading the coins.
"Seu cabra da peste,
"Sou Mangueira......."
Hooray! Keep us updated on other and future breakthroughs.
It is my understanding that major dealers NEVER used the ANA Grading standards. However, me, most of my friends, and probably a majority of small local dealers did use the ANA Standards. Anyway, IMO, if the ANA Grading Service had done its job correctly and their standards would have fit the prevailing commercial coin market, ANA MS-65's would not have become MS-63's and the FIVE major grading services along with several others possibly would have never come into being.
BTW TD, thanks for your contributions to our hobby.
Then who needs AI? What is AI suppose to do to improve grading if not to add precision (grading the same coin the same way every time it is seen)? All you need is a camera and a computer system to "fingerprint" coins. Aren't diamonds graded that way now?
Beats my five aces but I'll bet it is not the ANA Grading Guide.
______________________________________________________________________________________________________________________
I'm a nobody
The ANA changed the grading standards by replacing, "Full Strike with "As struck" in 2006 or 2007". IMO, this change came about due to the kick-backs on complaints from buyers that the coins they bought had flawed strikes. And so, the CGC's didn't want to be liable for that anymore. And submissions were down. Why else would they change the ANA grading standards?
But there are far more collectors of labels than there are those who actually study the strike and luster that meets their strict standards. And CGC's need to survive with coin submissions. It takes time for collectors to read a coin's surface and the details but a few do eventually graduate to seeking higher quality coins and they can't wait for that to happen.
Leo
The more qualities observed in a coin, the more desirable that coin becomes!
My Jefferson Nickel Collection
You aren't understanding what "AI" is or how it is used.
TPGS's will use AI in their operations because it makes their operations more accurate and efficient. But the implementation of the technology is on an individual use basis determined by the developers.
Again, there is not a single "AI" entity that everyone will use to do everything. There is AI technology that people are using to develop applications. There are already multiple LLMs (ChatGPT, Gemini, etc) as well as different AI models under development that are not LLMs at all.
You don't "need" a phone... unless you want efficient communication. You don't need to use AI unless you want to increase efficiency and, possibly, improve accuracy. But you are not ceding to an omnipotent entity. You are adapting technology to an application.
Think of "AI" as a computer. You can do a million different things with a computer. The computer does not force users to all use it the same way to get the same result. If provides functionality that any user can adapt to help with the task at hand. My father used it to catalog his movie collection. That didn't give all of us the same catalog with the same cataloging format. AI is a technology not an application. You still need to create the application and NGC can create an application that gives different results than PCGS.
It is also quite possible that even the PCGS application will continue to have some variance in the same coin due to imaging variances. Do all True Views represent the coin equally well? And variations in toning and luster between different coins will likely still create different results from similar coins.
Let’s think bigger ….
Since AI grades from images (stills + video under rotation) doesn’t that mean AI can grade without being in the same room as the coin itself?
(that would save on postage)
And if,say, PCGS, NGC, and CACG standards can be made into strings of computer code, then couldn’t, from images, a 3-part report be compiled, such as:
GRADES by
PCGS. standards 66.5
NGC standards 66.8
CACG standards 65.9
I’m Just spitballing here, seeing if anything sticks…
30+ years coin shop experience (ret.) Coins, bullion, currency, scrap & interesting folks. Loved every minute!
You need to standardize the images or you could easily deceive, Eben without outright photoshop fraud. Lighting alone can make an XF coin look UNC and vice versa. I would never trust a grade arrived at from images alone.
Yes, there would have be a standard for lighting, angles, and rotation in place. That’s the relatively easy part.
Just the act of creating such a thing would be a major advance in showing coins for sale or auction, or enjoying your coins at home, or sharing them to your friends. You’d buy the coin, and a well-lit and attractive video representation would come with it. The video could be ‘interactive’ in that you could pause it at any angle of rotation/ lighting.
I’m sure other features would also be dreamed up. Think of the whole new level of GTGs we could have here!
But I’m sure the auction houses are already working on this aspect of coin presentation. If they do a good enough job, actual humans - professional and hobbyists - could grade remotely. Or just decide with more confidence if that’s the coin for them!
But I digress.
This thread is still about AI actually learning to grade coins better than a human.
30+ years coin shop experience (ret.) Coins, bullion, currency, scrap & interesting folks. Loved every minute!
I just came across a fascinating idea to address the issue of market grading for AI.
You could train an AI using images of a coin + the sold price. There is no need to mention the coin's mintage, history, etc. The market has already spoken through past sales.
The AI could learn a coin's desirableness [luster, strike, etc.] purely by how much it has sold for compared to how it appears.
My only hesitation is that the AI would be trained with photos. It may not work if the AI is trained with images, but we then ask it to analyze coins using videos. It wouldn't be a true one-to-one comparison.
I like where you're going with this, AlbumNerd, imho it seems like access to data is the limiting factor; that is, AI is only as good as the info available to it, it might get very accurate with millions of images to compare, and eventually it would be even better with video or other input, the possibilities are endless
I've spent the better part of 34 years trying to get the CGC's and everyone else how the strike of a coin is the most important factor in determining a coin's grade....what makes anyone think computer-grading will be programmed any different/better?
Leo
The more qualities observed in a coin, the more desirable that coin becomes!
My Jefferson Nickel Collection
@jmlanzaf
If you have time to make a short list for us. What will AI be able to do for a TPGS:
AI will be able to identify a coin that it has seen before. (So will an image of the coin stored in a database. I know that ANACS weighed each coin an took a picture of it in the beginning.)
AI will be able to detect fake toning? (Doubtful, humans can't although they think they can.)
AI will be able to measure the reflectivity of luster and depth of PL better and more consistent than humans.(I'm sold!)
What else?
5.
6.
Published standards? There aren’t even really any standards at all. All there really is a general approach, That’s why it’s possible to grade things we’ve never even seen before, Including some things we didn’t even know existed.
Doggedly collecting coins of the Central American Republic.
Visit the Society of US Pattern Collectors at USPatterns.com.
AI will be able to HELP with everything. Again, it is a tool. And it is a tool that works 24/7. It doesn't have to be better. If might not have to even be 100% equivalent. It just needs to be close.
AI MAY be able to better detect fake toning, authenticity, luster, strike, etc. But not by itself. The application has to be developed. A bad implementation could be worse.
Consider strike. A human looks at the coin and gets an impression based on experience. The AI on the other hand COULD (depends on implementation) look at a billion data points from across the surface and seen from a dozen angles...in seconds. And if properly told what to look for, it will collate all that data and make a reproducible judgment.
And don't forget that AI is improving exponentially. ChatGPT is more accurate and more functional than it was a year ago. In 5 years or 10 years...
Doc has got it right. AI will never master the nuances of properly grading coins. It's time to dump the Sheldon numbers for grading a coin. The numbers game which has been plaguing the coin collecting hobby for many years now has gotten even more ridiculous with AI.
Great spirits have always encountered violent opposition from mediocre minds.-Albert Einstein
Lmfao
Your argument isn't even internally consistent.
AI is not being used (yet) to grade coins, so how can it have made assuring more ridiculous.
You don't think AI will matter the nuances of grading at the same time that you want to dump the standard grading scale.
You don't even have an argument.
I’m just trying to “remember” the Internet in 1980…
30+ years coin shop experience (ret.) Coins, bullion, currency, scrap & interesting folks. Loved every minute!
>
The Sheldon numbers are simply a proxy for price/value. If you replace them with another 1-dimensional grade, it will serve the same purpose. If you replace it with a 4-dimensional grade (surfaces, luster, eye-appeal, strike), you now have to establish how to project that into the 1-dimensional price/value domain, because at the end of the day, that's where every transaction is.
Keeper of the VAM Catalog • Professional Coin Imaging • Prime Number Set • World Coins in Early America • British Trade Dollars • Variety Attribution
It was restricted largely to national labs and the like. It was all text at a couple hundred baud on dial up modems.