How long before coins are no longer graded by people but by computers?

Would this eliminate the TPGs upgrading of rare coins simply because there are rare? An accepted practice in the industry. If everything is measured by one standard, how would the market react to down grading all of these rarities at once. Will everyones' slabbed coins need to be re-graded to the new " machine standard"?

 

Log in or Sign up to hide this ad.

Thats the real problem! Until "mind reading" can be accurately developed, NOBODY knows how people "think". We can observe other peoples actions and reactions and assume we know what they were thinking based upon whats going on inside our heads, but folks simply never know what someone else is thinking with any accuracy.

Which language do multi-lingual people think in? Better yet, how does a baby "think"?

How about this? If a computer program can be written to accurately grade coins, will it grade them a point or two higher if it "owns" the coin??

 

Grading is directly related to market and determining value. The Sheldon scale was based upon basal coin state value compared to mint state value. It's all about the money.

IF, there is a unique, one of a kind piece that folks are willing to pay $1,000,000 for, does it really matter what it grades?

"Will everyones slabbed coin need to be re-graded to the new "machine standard"?"

I don't think so anymore than I think that all older slabbed coins need to be re-graded for the new prong holders or the new PCGS sniffer Plus Grading.

As a point of fact, coins don't really NEED to be graded! Until, that is, they are being out up for sale. Then and only then is a third party opinion needed to establish relative "value" for what everybody else pays for a coin in that condition.

 

It looks to me like all the arguments here against machine grading boil down to the importance of subjective judgement. For a collector, that's what it should boil down to -- does the coin appeal to you enough for you to buy at the asked price, or does it appeal too much for you to sell at the offered price?

For TPGs, though, it's not working out. We're getting inconsistency, "market grading", grade inflation, CAC second-guess stickers, and the "re-submit until you luck into a higher grade" syndrome, which over time will produce a preponderance of overgraded coins (because an undergraded or accurately-graded coin will always tempt someone to try for an upgrade, but once the coin is clearly overgraded, it'll never be submitted again). Some people (hi, Doug!) have already given up on TPG "standards" as a result; if today's trends continue, more and more people will do the same.

Automated grading may be one way to put a stop to it. There are reasons it may never happen; I'm sure TPGs love being able to count on a string of resubmissions, and I'm sure customers love the promise of landing an upgrade. But I don't think there are insurmountable technical barriers.

Humans can't always explain subjective factors, even if they can evaluate them consistently. But that doesn't matter when you're building an automated system. You just train it until it produce results consistent with your trusted evaluators. At that point, you still don't necessarily know how to explain those (formerly) "subjective factors" -- but you have a machine, which you can back up, duplicate, and sell, that will evaluate them consistently every time.

Some of you are probably 100% convinced that it's not possible to train an automated system in this way. You may be right -- but I'm just about certain that you're wrong, and I'm more certain about it with each year of technological advancement. It's not something that I have the resources to prove (by demonstrating it), but I can't imagine any way to disprove​ it (prove that it's impossible).

 

First, proving a negative is a logical impossibility. So of course it cannot be disproved.

I do not say it cannot be done, but would you want it? Would you want a coin that beauty absolutely has nothing to do with the grade? Maybe you are right and it would be easier to just ignore the idea that technical grade should have anything to do with value. I do find the TPG doing market grading dangerous, especially with toning as tastes can and do change.

IDK, maybe going cold turkey onto a new system would work. I think everyone would be shocked by some horrendously ugly coins and how they grade, versus some truly beautiful coins and the low technical grades they would receive. I think such a system would push the idea of sight unseen purchases even further away from reality.

Bottom line, I think most collectors collect coins for their beauty and not for their luster percent, surface topography, and spectral analysis readings.

Just my opinion.

Chris

 

I've programmed computers for over 30 years and I have no doubt that you are correct. This is exactly what computers are good at. Grading is based upon rules and computers are rule-based engines. They are ideally suited to be used as comparators against models. The only problem I can see is cost and politics (whether they actually want particular coins to be graded at a certain grade).

 

Thinking in the way you mean is not required for grading a coin. Just like the computer that can take an Airbus A380 from the runway in London and land in New York does not think and yet it can completely perform the job of a pilot who does. And many times do the job better. I submit that flying a jumbo jet half way around the world is a far more complex task than grading a coin.

As to the point that computers can't evaluate something subjective like beauty, they not only can but do. There have been several examples of programs that can look at a face and predict accurately how attractive it will be to other people.

The more I think about it the more sure I am that the tech required to machine grade not only already exists but it's not even cutting edge anymore. As to it being inevitable I've reconsidered. To come about would require a commitment form one of the big two and market acceptance which I don't see happening. And that doesn't disappoint me either. I don't think it would be a good thing for the hobby.

 

Well, I did say "is supposed to emulate" not "a neural net that thinks like a human" . But I think the people experimenting with artificial intelligence try to emulate thinking currently by programming a computer that can learn, or at least that what it seems like when I've seen shows about the primitive current artificial experiments. As far as the broader subject, whether or not a computer that simulates the learning process actually thinks or not, and at what point it might be considered a "thinking machine", thats been a subject for debate for quite some time, starting back around Allen Turing, I think.

--------------------------------------

Modern computers are kind of like idiot savants in some ways, there are things they can do that are amazing, and amazingly simple things they can't do. They can land a plane, but they cant drive a car (yeah, I'm aware of the experiments with the computers driving--they cant, at least when last I looked). They can shoot anything that comes into their range if set up like a sentry, but they cant catch a baseball. They definitely cant do things like judge artwork, critique a movie, or grade a term or creative writing paper---but they can point out grammatical errors in one in a quarter of a second. I'm not an expert in coins or AI, but it seems to me that judging coins falls more along the lines of the things they cant do now.

 

This bears repeating. It's perhaps not very good for our self-image as a species, but a lot of "ineffably human" things like perceiving beauty turn out to be not only possible for computers, but not even all that hard, relatively speaking.

 

Yeah, they can. Not perfectly, but how many perfect human drivers do you know? Nevada is already defining laws for self-driving cars; they'll be on the roads sooner than you think.

Well, actually, they can -- in fact, they can catch two at a time, one in each hand.

They can also dribble a ping-pong ball between two fingers, tie a knot in a rope one-handed, or snatch a tumbling cell phone out of the air in the correct orientation.

When it comes to dexterity, not to mention speed, they're already better than us.

I'm not sure how much weight I give most of these things, because it's hard to get people to agree on whether any given human is doing a good job of them. But computers have demonstrated the ability to detect art forgeries, and school systems continue to experiment with computer grading of essays.

I don't mean to threadjack, but I've spent a lot of time thinking about these questions. I've been reading science fiction since I was tiny, and I struggled through at least five artificial-intelligence classes on my way to a Ph.D. in computer science. AI is not my field of specialty, and I was pretty thoroughly skeptical about it twenty years ago. Now, I'm still not sure what I think of it -- but it is MOST DEFINITELY producing real-world results.

 

GD, I'm going by how Scott Travers defines the "rules" of grading. If the industry wants "subjective" grading then they'll never solve their problem of inconsistency. Maybe they don't want to.

This is not like comparing two paintings. Each coin has "one" design from a master engraving. We supposedly know what a "perfect" specimen is. As I said, computers are excellent comparators. All you need to do is define the rules down from perfection. Flaw detection at a microscopic level is already done in many industries.

Can computers think? That depends upon what your definition of "thought" is and what thought you think is required to grade a coin. If increasing levels of grade steps depend upon a decreasing level of flaws, then computers absolutely have the thought required. If grading is subjective and inconsistent, then no.

 

Ah, but the question is ... Should it be?

 

This. Opine about a coin's subjective virtues all you want, but don't call it "grading".

I'd be happy to see grading factored out into a mechanical portion (numeric grade) and a subjective portion ("green beans" and whatnot). In fact, I'd love to see a multidimensional objective scale, with objective measurements of wear, strike, bag/roll damage, and perhaps even tone. That would leave us humans free tofocus on the really important issues, like "which is more attractive, the Eisenhower obverse or the Susan B Anthony obverse?"

 

Well GD, having an insatiable curiosity that's led me into many hobbies and activities, and a proclivity to take things I'm interested in to, some would say ridiculous extremes, what I've learned is that most "experts" ... aren't.

Granted, you have to take into account what they say until you know more than they do, or can prove them wrong.

Even though I started collecting coins 40 years ago, I never caught the "bug". So I'm just a speculator researching whether to diversify a very small portion of my portfolio into rare coins. I actually think "market grading" benefits a speculator because there seems to be a reluctance to outscore the current population of higher state coins. A machine would not have that reluctance. But would it benefit a "collector"? I don't know. That's something the collector community will have to decide.

 

Well, the main reason I mentioned those two things in my previous thread was to show how computers are capable of doing things that we cant do, yet have trouble with common tasks that a human would take for granted. But thanks for the links, they were interesting, and brought me more up to date on things. They've gotten further along than I would have thought from the last time I looked up on the state in current technology.

Well, not really. At least not like on a level like they can land a plane. That is that peoples lives aren't being entrusted to computers on a large scale in cars. Those cars are basically advanced prototypes, and I think its likely they will need a huge amount of real world testing and programming before they're near ready for production. But they're further along than I thought. Last I saw about it they had to use devices on the road to get them to come close to working right.

That law, though, seems more like some kind of gimmick. Theres clearly not a need for laws governing automated cars, so I suspect there was some other motive involved.

In my defense thats a pretty recent development. Thats pretty cool. I know that physics type stuff was an example of something that they were having problems with last I checked up on things. But I'm still not impressed with computer dexterity overall, they really can't pick themselves up off the ground if the fall when they run.

The ball catch robot does show a very good example of current limits in computers with the coffee machine, and the type of problems computer advancement faces, IMO. It can make coffee from a very simple coffee maker when set in a certain position relative to the coffee maker. The robots makers say in the video that its a good example of a robot dealing with a complex situation. This wasn't explicitly stated, but I'm pretty sure if it were put in a different position to the maker it wouldn't succeed in its task. And I very seriously doubt it could even recognize a coffee machine from another manufacturer. To clarify the idiot savant thing, it cant cope with things that it isnt specifically programmed for, it has to be told what to do in practically every single situation, and has limited abilities to cope with a changing environment, but its very good within the parameters set for in the environment its programmed for. At least that's my understanding of the current technological situation. As I said, I'm in no way an expert in these things.

This actually brings things around to to the topic of the thread. Humans do these things pretty well. True, there are subjective disagreements, but there is usually a general consensus with something is aesthetically pleasing. I'm sure that you could find someone who'd say Mila Kunis isnt hot.... but that would just be some bad programming. Programming an aesthetic sense into a computer is something that I couldn't even imagine trying to do. In some cases it might be doable, for instance I believe human beauty has been quantified, but we're a long way off from the first computer movie critic. However, a current computer could prolly write better movies than some that are being put out...

Dont disagree. I actually still find the fact that airplanes are computer controlled on landing pretty amazing. My point was more that there are areas where it is very difficult or close to impossible for computers to make inroads with the current and probable near generations of technology. At least not without a huge budget and staff. Even then its more "forced" I guess you could say. For example, to my understanding, for coffee/ball catch robot to be able to roll into your house and make a cup of coffee on your maker you'd have to program an incredible amount of information, much more than would be practical. And thats just for one house. I think the AI that can expertly grade coins and make your coffee while simultaneously plotting the death of humanity is pretty far down the road... hopefully. But I do think that it will exist one day. As far as AI goes I think were around the kittyhawk stage, maybe just getting into biplanes.

 

I recently had a dental appointment, and in the course one visit and a few X-Rays they were able to create a 3D computer model of my mouth and jaw. The things computers have become good at is performing measurements and creating models. Lets say you were to scan a series of coins at various grades and create a model for each grade, it would then be a matter of scanning an ungraded coin and comparing it to the computer models. I would think at the microscopic level, there would be a measurable difference between a coin with luster and one without. You could count the hair lines and marks within a defined area and have a degree of tolerances per grade. Another thing computers have become good at is creating paper reports. At least you'll get a detailed report on why it received that grade.