Cast fakes. Struck fakes. Transfer die fakes. Looking at an image of a coin is not always enough to detect a forgery, and often you need access to a catalog of dies (in acsearch.info) and compare them to determine a coin’s authenticity. The time to do it is exhausting. BUT with an AI, it can scrutinize the entire catalog ultrafast. If we can “teach” or program an AI and give it all known genuine coins and all known fakes, logically it would detect forgeries. Of course it will need to be updated gradually and programmed correctly. I hope we can make an AI to detect forgeries and employ such a guy like David Sear before he dies, so his “expertise” could be giving to the AI. Also acsearch.info have a massive catalog of different dies and so have Dumbarton Oaks in Washington DC (a lots of byzantine coins and seals), so logically that site should invest in AI. I want to ask the older members here: Do you think in future about some few years there will be an AI-site to detect fake coins?
Yes there will be in a few years, and probably as many years or more for collectors to trust it. It will also have to be affordable.
I can see, if there is a AI program to detect fakes, then it won't take long for crooks to figure how (not) to make fakes with the same program.
Tommy is correct. But it also may allow the mints to use it to control the already computer guiding die making software to make all coins that have no variations with computerAI to double check . There goes all the variations we look for one by one. No use to buy boxes of cents, etc. as they will all be the same die image. I was thinking pf selling some varieties. but likely the value of current varieties will increase and no new ones will appear ( Except from certain countries mentioned before.) Why send such a coin to any TPG unless you just want a slab. Look for AI to be used ( if not already) by the Imaginary monetary sites and countries with digital basis.
Yes, I like the thought, but believe the time and expense factors are incongruent for numerous reasons. JMHO
I had a major discussion/debate as the devils-advocate about something similar and AI, with a well-educated Grand-father. The complexity and investment in the task is appreciable and expensive to many who may not see any societal benefit in the/your posted potential capability. The general goal of AI is believed to advance society! A grandfather who last year gave certified coins to his grandchildren was asked, almost to a receiving individual, what is this, what good is it? I gave him 2 rolls of different XF+ average condition @ ~50% of Greysheet pricing, and advised him to learn/teach them about his new conversion from Fiat. I displayed 3 certified choices to him, 1889cc F10, 1879-S DMPL MS65 Dollars, and a 3 coin lot of MS69 Gold 1/10th oz, all of the same 4 digit value, asked him to learn/teach which may be a best investment. AI is believed to be a reasonable near future tool for something as that!
AI, is actually not the correct term to be using. Machine learning or ML is a subset of AI that has been in use for awhile in some industries. What we are talking about on this thread is a quality ML program using cameras for visual inspection. Coin ID could be done and is probably already tested. It's not that hard to train based on images. The harder part is going to be training based on wear and in ancient coins on partial strike. It can be done, no doubt.
The biggest challenge to this is that many fakes are cast versions of real coins - so by an image alone an AI program may not see any signs of "fake". There are programs today for searching a coin database for die matches e.g. image search on ACSearch. Their algorithm could potentially be used against a database of known fakes so that you could test your coin - image search is pretty good today.
I would trust more if we had a techique to accurately date the coin by carbon dating it. So for ex. the AV Aureus Sponsianus (now deemed geniune) would prove it was struck 250-60AD. When you watch "Curse of Oak Island" they are carbon wood beams/ iron artifacts/ why not gold/ silver coinage.? So far they have not found any precious metal objects. But they did date that Templar lead cross to 1400s.
"AI study suggests famous Raphael painting was not entirely his own work" "Using deep feature analysis, we used pictures of authenticated Raphael paintings to train the computer to recognise his style to a very detailed degree, from the brushstrokes, the colour palette, the shading and every aspect of the work. The computer sees far more deeply than the human eye, to microscopic level," explains Professor Ugail. From: https://www.euronews.com/culture/20...aphael-painting-was-not-entirely-his-own-work If AI can be used to spot/analyse painting on MICROSCOPE level, then why not coins? I am really looking towards the AI.
As @Valentinian said, that only works for things that contain carbon. There's a technique called neutron activation analysis that works with metals, but you need to soak your object in a nuclear reactor core, and it comes out radioactive. It's "non-invasive" in the sense that you don't have to drill out or scrape off a sample, but it actually transmutes some of the object's elements into different elements -- only in tiny trace amounts, but still, it's not entirely "non-destructive".
so its coming to this using tech to detect the real deal from the fake s make s me wonder if they use this at when you send it off to get slabbed or do they have individual s with years under their belt and a trained eye making the judgement if the casting in making fake s is made by computer s n software , and using a program detect the outcome of quality in either being real or not than we have a problem a really big one
Actually, today a criminal with skills can make almost 100% perfect forgery. They are called modern transfer dies: You cannot (always) detect the forgery of transfer die by holding it in hand and analyzing it with the naked eyes. The only way to detect such a forgery is to have a catalog of different coins of the exact same die and making a comparative analyze and looking after details, bump, hits or scratch. The first fake coin sold cannot be spotted easily, but if coins with the same details, bump, hits or scratch appear again, again and again in different auctions/stores then the bell will ring sooner or later. And here I think AI absolutely will help, since it can have a large database of different dies and can analyze it in a fast way humans cannot. That is why I don’t think AI will give us problem, but rather it will give the criminals a problem. “I think”, perhaps I should add. And for the record, I am not an expert in any way about coin forgery, and other people with more knowledge than me may disagree.
