Winds Of Change In The Philatelic World

Looks like his inventory (and likely other materials) were divided up to be sold through numerous channels. Various and sundry auction houses apparently have parts, and as was mentioned several weeks ago, at least part of his inventory is being sold through Apfelbaum at retail in country-based lots, so apparently they've had the material long enough to work it up.

Seems odd that it would be parted out through so many different operations rather than being handled by one entity... unless that was attempted, but certain firms didn't want certain portions of the inventory.

So if I read this board correctly ,then I am guessing the better stuff Sergio had can appear in a later auction ,maybe a name auction ,meaning a seperate catalog and a stand alone auction of his inventory .

What we are now seeing up for auction is the lower level stuff he had .The better material isn't yet written up for auction ..

It is all supposition. We have no idea what Sergio personally owned or where it went. Nor do we know if anything was tossed out or if that was just a fish story. What we DO know is that Sergio was an expert that certified thousands of items and should have voluminous records related to that activity. So where are they? We also know what we see with our eyes at this moment and that is Sergio related material popping up in various places.

For me, I vote against some juicy "name" sale down the road. I think what we are seeing is what we are getting. An estate liquidation and there ain't no pot of gold under the mattress.

The general problem of potential book and knowledge loss in the hobby has been bothering me for some time.

Speaking to the books, some philatelic books have been digitized and are preserved for now, but many more have not. I have, of late, been on a mission to build up my personal library to a reasonable extent, while I can still find the books I want now, or know I'll need someday. I am alarmed at how difficult it has been to find some books, that I thought should have been readily available.

The overarching rule for all things computer is GIGO. AI does not magically change the GIGO rule other than adding the modifier "Intelligent" such that I-GIGO is just a subset of GIGO.

For AI to do non-GIGO work expertising there needs to still be real experts doing the input to create the basis for determining what is real and what is fake BEFORE the AI can be used.

A good place to start would be the Hawaii missionary stamps to see if AI can sort them out due to the depth of knowledge about all the existing copies, real or fake.

I disagree with this. AI systems are not simply passive recipients of data and are more than just blindly processing data. They can learn and adapt to their environment, which means that they can sometimes produce accurate output even when the input is not perfect.

While it is true that AI can be wrong if it has incorrect input, I think it is a far cry from "garbage in, garbage out". Here is why I think it is grossly over-simplistic to say AI is "garbage in, garbage out";

1. Learning and adaptation: AI systems, particularly machine learning models, have the ability to learn from the data they are trained on. They can adapt and improve their performance over time, even if the initial data is not perfect. This means that while poor-quality data may lead to suboptimal results initially, the AI can work to mitigate the impact of bad data through learning and refinement.

2. Feature extraction and representation: AI systems can automatically extract relevant features from raw data, allowing them to focus on meaningful patterns and information. Even if the input data contains noise or irrelevant information, AI can learn to distinguish and prioritize the important aspects.

3. Data preprocessing and cleaning: AI often involves extensive data preprocessing and cleaning steps to remove outliers, correct errors, and handle missing values. This helps improve the quality of the data before it is used for training and inference, reducing the "garbage" in the input. AI systems are designed to be fault-tolerant.

4. Transfer learning: AI models can leverage knowledge learned from one domain or dataset to perform well in related domains or datasets. This means that even if the initial data is suboptimal, the AI can benefit from prior knowledge and transferable features.

5. Continuous improvement: AI systems can be designed to continually monitor their own performance and adapt to changing conditions. This ongoing improvement process allows AI to mitigate the impact of "garbage" data by updating models and strategies.

Don

Sorry, Don; this is not a perfect world. An AI system is first programed and there is no certainty it was not do so in a flawed manner, be it poor programming or introduction of biased data sets for learning. The resulting term is machine learning bias or AI bias. That has been shown to exist with catastrophic results, in some areas, facial recognition as an example.


Likewise if the learning data set is say, is made up of ONLY fraudulent, fake or reproduction stamp information then there is no basis for "learning" to ID a valid, properly issued stamp. You can train AI on data for which you the programmer do not understand, are unable to clean up nor able to ID correct from wrong before sending it in you either get garbage out or, nothing at all since the only other choice is not to send in anything for "learning" and thus not develop the capacity desired.

As I mentioned in a previous post, per the Kelleher article, there are now a flood of excellent forgeries (reproductions) with out experts to tell them apart from the the genuine stamps. That makes at clean data set for teaching impossible.

Lastly a comment to the remove the outliers, rote removal of outliers is at times removal of the most important information. My favorite outlier was the PRION, with the scientists viewed as an unworthy quacks and worse before a Noble Prize in Physiology or Medicine was awarded to one.Then again, our current educational system hates outliers and shuns them, yet they provide more to society than the average group of a few hundred million others.

PPG said:


Don has countered by pointing out how AI can "learn" and thus improve on what was initially input to create it. And to that extent, AI is not strictly GIGO. However, in my experience, no matter how "good" AI may be, when being used it is still limited by the quality of the "input" of what it is asked to do. I've been able to generate good results using AI, and I also know how to generate bad results. In a different context from philately, I recently used AI to generate something that was very good. In presenting it to others I made a cautionary observation of AI subject to GIGO from the standpoint of the quality of the questions put to it. I used the word "perspicuity" to describe the kind of insight that is required to craft questions put to AI that are likely to create good results. In my opinion, at present, one cannot get "expert" knowledge out of AI without having "expert" knowledge to begin with in crafting the questions put to it. The present AI models strike me as somewhat syncophantic, too eager to please, and thus prone to a kind of "confirmation bias." This is most evident in the well-known tendency of AI to "hallucinate" in responses to requests to generate references to source material for what it "knows." In the hands of those lacking expertise to begin with, I do not see the present generation of "generative AI" models to be very reliable.

Basil

The human brain is not perfect either.

I think the human brain is close to perfect but the emotions part always gets in the way. The irony is that the human brain created AI.

or did AI create us?

If AI created us it makes the point that AI is indeed prone to GIGO.

In my opinion we should not consider that AI systems are limited like humans. AI does not simply get some input, learn a bit, then stop. AI systems are not limited to living only X amount of decades. AI systems can keep going, getting better and better and better. This is what makes it so much more then GIGO and makes it so much more than what a specific AI system can or cannot do right now.

Using AI systems is also very similar to using a typical online search engine like Google. The vast majority of people I see and work with everyday barely scratch the surface of learning how to use Google search. They simply type in a key word or two, not get returns they expected (or get too much) and move on. They do not bother trying different type search engines (like Google Scholar), they do not try to improve their query. Hell, we even have multiple examples in this forum where people do not even try to find answers using a search engine. So it should not be surprising to anyone that a lot of people are not willing to put the work into learning how to communicate with an AI system (never mind multiple AI systems). You cannot simply type in a single question and expect a flawless response. You cannot use only one AI system and expect that all other AI systems will be the same. You cannot learn how to effectively communicate with an AI system in less than hour. I have already spent hundreds of hours learning various AI systems and to be honest I think I am not sure that the learning will ever stop for me.

I recommend that people spend an hour or two learning how to ask and re-ask their questions to learn how the AI system responses. If it gets something wrong, tell it that it screwed up (and ridicule sometimes helps!) and try again. Ask it to expand upon certain topics. Ask it to present less biased responses. Learn how to use adjectives in your questions to get more creative responses. Make sure that you specify the amount and format of information you want back from the AI system. Communicate with it the same way you might communicate with a genius 200+ IQ child or savant.

So, I would agree that GIGO is in play here on the human side of things but not as much in what an AI system has in its knowledgebase. Instead, GIGO is heavily involved in how people are asking it questions and their capability of learning how to communicate with the AI system.
Don

Back to the question of what to do about the prospective loss of human expertise in recognizing forgeries, the AI angle I was thinking of was how that technology is being used to revolutionize diagnostics in health care. The problem of missed diagnoses is a long standing one due to the fact that every MD has more or less limited experience and there just aren't enough expert diagnosticians to go around. This is a concrete area where AI is already helping.

I know I'm on thin ice here because I know absolutely nothing about how the experts spot forgeries but at least on the surface it seems to me to be a similar problem.

Morning all,

Back in the days, 50's and 60's, I read a lot of science fiction. Remember Isac Asnof"s (sp?) three laws of robotics? "... a robot cannot do anything thats harms a human ..." I hope AI has this limit in its knowledge base; however, then we have Mr. Spock's old saying, "the needs of the many outway the wants of the few".