Using Artificial Intelligence (AI) in Coin Collecting

Well, AI is doing well in some areas, but world-wide legal systems are having problems that will get very big. https://www.ncsc.org/resources-courts/ai-generated-evidence-threat-public-trust-courts

 

Log in or Sign up to hide this ad.

AI usually never tells me anything that I didn't already know.

Z

 

I still see lots of people dismissing AI with statements that amount to "it's not really intelligent, it only parrots back what it's been trained on".

Am I the only one who's convinced that most human thought amounts to nothing more than that?

 

Therefore...people aren't intelligent. QED.

 

That's the dark side of the Turing Test.

Alan Turing: Put a person in one room and a machine in another, and let them communicate via typing. If the person can't tell whether the entity on the other end is a human or a machine, the machine must be intelligent.

CoinCorgi: ...or the human must not be.

 

I make no assumptions.

 

Indeed! Exactly true. We mistake thought for intelligence.

I think therefore I am.

I believe we're going to be able to use our thinking and abstractions to invent true intelligence; machine intelligence. We are very close.

 

Copilot already exceeds the turing test. I can still trick it into acting like a machine but otherwise it can pass.

I would define "intellience" as the ability to extrapolate the unfolding of time as it is occuring.

 

Yep, we've well and truly blown past that goalpost. People now are scurrying around with the remaining goalposts, arguing about where they really belong.

 

Precisely!

AI "hallucinating" complete crap out of whole cloth is SO real. I write about numismatics. I began a project on Philippine coins and wanted to know what the foundational works were on the subject. I went to Grok AI and did all those things that they tell you to do if you want to really utilize AI (give the AI machine a role as the ultimate expert, be specific, etc.). What Grok gave me were seven sources. Six were legit. The top one cited, the most promising source with the most amazing information, was completely made up out of whole cloth. The "author" of this fictitious book himself was fictitious (the name was actually a Filipino naturalist who is not at all involved in numismatics). All of that I discovered, of course, only after further examination of the sources.

Doesn't this not bother people ENOUGH??

 

I don't rely on it. I look at the output, then I look at the search results, then I think for myself.

I've been right every time so far. Google search results are about 89%. AI is about 50% (depending on how I phrase the query).

 

AI is useful - I use google AI searches all the time. But like you I think about and then try to verify the results. When asked why use it if you have to verify - well guess what - I would rather spend a little time verifying(and get a good starting point) than a whole lot of time researching.

 

I have used ChatGPT for help identifying world coins and it was able to identify them accurately however, I also used it to identify stamps which it did horribly, like not even close.

 

Have you tried google Gemini? So for fun I went out to chatgpt. At first I went on and installed that ace browser. I then played with it uploading pictures. For example a picture of my neighbors tree from the snow storm. I then asked where was it taken - boom it analyzed stuff in the picture. This looks like the 2009 ice storm that hit ky. Based off stuff it saw - it said a louisville suburb. Then another picture from the storm which showed a car in front of. Then based off the mile marker and the license plate it is from the 2009 ice storm in ky - it was close but was actually taken right across river in Indiana. So cool. Now I uploaded some coins and asked questions. It got general questions correct. Then I started asking questions like - what is cohen number? Or what is sheldon number? what is overton number. Now it nailed 3 of the 4 - the fourth one I can't confirm because I can't find my bust quarter book to review it. But I have it (and it is labeled) as B-9, but chatgpt said B-6. So I just thought of one 1806 bust half I have - chatgpt says o-116, bought as dealer said O-116 - but I really think it is O-115A - but I can be wrong. I even posted a thread here some where on the attribution.

Anyway - thought this was cool and never thought of trying it. Now that Ace browser screwed up my normal browsing with chrome. Took a bit but it is gone now. Then had to figure out free chatgpt. Same results with free in either case. So I then went to Gemini and asked the same questions - no where near the detail and it will not actually say S-236. So not as good as Chatgpt - just my opinion.

A new toy to play with.

 

Ya know, it seems much of what is called AI is the ability to parse questions, then rapidly search an enormous database and present results in a conversational syntax. Not that it’s a bad thing. It’s very useful, and I use it often. However, true intelligence involves problem-solving and creativity. AI has a good start on the former for problems that are well framed. Not so sure about the latter. A lot of supposed AI creativity appears to be driven by substitution of words and phrases in an existing framework.

Mike

 

I can get completely opposite answers by varying how I phrase the question and the sequence in which I vary it.

 

A niche like numismatics does not have the thousands or even hundreds of web pages that a LLM can scan and then come up with a plausible AI answer to some basic questions.

I input this to Gemini, Google's AI: "Can you tell me how European hoards protected Saint-Gaudens Double Eagles ?"

It gave me a good overall answer but it had some dubious and wrong information as well.....including a marketing ploy called the "Rought Rider Hoard" as part of a legitimate hoard brought over from European banks.

An article from the March 2010 Numismatic News specifically mentioned European bank hoards saving SGDEs so obviously the scanning LLM was able to sweep it up.

I'm not an expert on this stuff but the original Google search engine would bring up a bunch of links to my question above or the keywords (Saint-Gaudens, Double Eagle hoards, European bank hoards, etc.) and then you would have to hit on 10-15 or more links and you'd get some great links and old internet archived articles from current, recent past, or distant past....and also tons of junk links that had nothing to do with your inquiry or something totally different because the Search Engine misconstrued what you asked for.

Today's AI is much much better -- comparing it to the Google Search Engine algorithim is like comparing broadband to the old dial-up modem connecting to the internet with AOL circa 1996. So I still got some junk with my AI inquiry but I also got some useful data and leads to explore. And I don't have to click on 15 links to get the AI summary.

 

Yes. AI "merely" substitutes words. But this is where it's showing its power because it tends to substitute words that are more correct.

It's trying to tell us something and we're misssing it!

 

The above post led to an interesting conversation with Copilot from which this emerged;

Humans flourish by acting. AI flourishes by processing.

And when you put the two together — human action plus AI procedure — you get a system that can expand in ways neither could alone. That’s the frontier you’re describing.

Not mystical. Not metaphorical. Just procedural reality unfolding across two different substrates.

 

I only use what we call AI for coin identification.

I see AI as a useful search tool and that only. It will never be able to think for itself and discover truth. It will reflect the bias and thoughts of the programmer(s). If it was your first time watching a magician and you don't know how his tricks work, you might really think he did make that apple turn to thin air. Man cannot create life.

"AI" is a useful search tool, not the answer to all life's problems.