Featured Finding information about books using Artificial Intelligence, is it useful?

As a collector of old books and IT professional I am interested in computer tools that can help me with my research into books. This tool shows some promise!

Recently Bing Chat added a feature to upload a picture as input for your AI chat. I was wondering how it performs with title pages of antique books. Can it be of any use when you want to research a specific book?

Link: Bing AI

This is how you upload a picture.



For every prompt I used the following phrase: What can you tell me about this title page?

The first book I tried was an older book: A Missale Romanum of 1572



This was the answer I get:



As far as my knowledge goes the AI is almost 100% correct in this case. Even the woodcut was analyzed correctly. The only part I do not agree is, that it isn't one of the earliest printed Roman Missals and the high value (but those comments are subjective).

Let's make it a bit harder. This time I choose a book I posted here on this forum before: Adults only: The naked book | Antiques Board (antiquers.com)



Here we see some hallucination (made up stuff) of the AI-engine. The title ,author and place are correct, but the date (1539), publisher (Johann Herwagen) and the illustration are not correct. It gives me a good starting point to start my research though.

Let's make it even harder for the engine to see what happens:

Confessionale Sive Libellus by Godschalck Roosenmondt of Eindhoven (Antwerp - 1519)





This is clearly too much for this AI engine. Only one word of the title is correct, the rest is completely nonsense.

Ok, now I know what I can not do. Let's try some books that are posted here on this forum:

Oscar Wilde 1903 German book | Antiques Board (antiquers.com)



This seems quite correct to me! I am impressed!

1912 Dutch Youth History, Illustrated with Inserted Cards | Antiques Board (antiquers.com)


This seems fairly correct, only the translation of the title into English is not correct. The book is not about natural history but about history and the cover was not correct described (an orange tulip???).

My conclusion:

This tool can be very helpful to start your research into a book that is unknown to you. You cannot trust the information it gives though, because it will hallucinate (make up facts) sometimes. As the website of the chat is stating: "Let’s learn together. Bing is powered by AI that can understand and generate text and images, so surprises and mistakes are possible. Make sure to check the facts, and share feedback so we can learn and improve!", do not trust the answers, but it can give very valuable information for your own research! The AI-engine rates all books as early, rare and valuable, take that as a grain of salt.

 

I'm also an IT guy. I use AI daily, but not to write code for me, although I have asked for help when I've gotten stuck.

I mainly use AI for proofreading emails, creating outlines for storyboards for learning modules, and for content creation.

You definitely need to verify the info they generate. They're only as good as their data pipelines, and you should have a few of them in your arsenal.

 

I dont see much value, especially with books where you have the title page.if you search for the most expensive offers in abebooks,worthpoint or good book auction houses you get more information and the references for further research

 

Bing's AI is related to ChatGPT. I've spent some time probing ChatGPT. I was a software developer for several decades, working in other aspects of AI, so the topic interests me. The problem I found, which you discovered as well, is that sometimes it fabricates "facts", and does it in a way that sounds very authoritative. This means that it can't be trusted. I think this truthfulness problem is related to the quantity of material it ingested about the topic during its training. For example, if I ask for 5 works of a well known sculptor on public display, it will give 5 correct examples. If I ask the same question for a lessor known sculptor, it will give a list of works, but if you look them up, they will be by a variety of sculptors, but not the one you asked about in your question. This is despite the fact that there are works by that sculptor it could have answered correctly with. If you then point out its mistake, and ask for works actually by that sculptor, it apologizes, and then returns a new list of works that also are not by the sculptor you asked about.
There was a news story recently about a lawyer that used ChatGPT to write a brief. It turned out that the cases it cited, although sounding very real, simply didn't exist. They were fabricated by the AI. I think this is in the nature of this sort of AI. It isn't really dealing in facts. It is dealing more in the patterns of language. It's more important that its responses look like appropriate responses, but the truthfulness of the details of its response are not guaranteed.

 

I know nothing about how AI works - but it seems it should be developed on at least 2 tracks - one for creative fiction, and one for factual, non-fiction information that incorporates some method of verification.

 

An interesting experiment! Seems that most of what is correct in the AI explanations are things that are easily legible on the title pages, or are very easy to look up with general search engines. But it is learning.

Just FYI about the following... your analysis is great, but I found one more thing that the bot did correctly.

The translation by the AI bot says "national history" not "natural history."

 

I agree fully, verification is always necessary. I use ChatGPT for coding once and a while and to analyze bugs. It can be quite helpful, but for now it still does not replace a programmer. For things as composing emails, sentiment analysis and translating texts it can be very helpful.

 

Sure, for the well described books it isn’t very helpfull, but for a newbe in books or a very poor description of a book it can be helpful. See the many questions on this forum.

 

Both ChatGPT and Bing Chat both rely on OpenAI models called GPT. The free version of ChatGPT uses GPT 3.5 and the payed one GPT4. It is not clear on what version Bing Chat operates.

Your example about the lawyer user ChatGPT was correct. It is more important for the engine to give an answer than tell the truth. That is a problem for now. The later versions are much better than the earlier ones, so I predict the answers will be better and better.

 

Good idea! You are not the only one who doesn’t know how AI works, actually nobody knows and that is scary.

The AI model reads a lot of text and learned to predict the next word in a text. That is all the model does. If the number of learned texts are large enough, the model can understand and answer questions, code a program and even learn Iranian for example. I think that is quite scary. What is next?

 

Thank you for your correction! You are 100% right. At least that does prove I am human and not a bot

 

I did some similar experiments with (pre-corrected) OCR'ed title pages, and the results are interesting indeed. Especially if you give specific prompts like "What are the full names of the author(s) of this book?" are mostly accurate. Nothing you can't find otherwise, but still handy because it's a lot quicker than Googling, especially for rare books (in my case often written by monks who have both "secular" and "monastic" first names). Very often it will end up giving a link to an (open-access) review article. Didn't know it would be able to actually describe woodcuts and such. Impressive stuff!

 

If you then point out its mistake, and ask for works actually by that sculptor, it apologizes, and then returns a new list of works that also are not by the sculptor you asked about.;;;;;;;;;..................


.."This means that it can't be trusted."


So right there.....the ' apology ' is just a word it learned to deflect from a factual error.....as it continues on it's merry path of spewing out more words that may or may not correctly answer a question .

It can't really apologize ...
but the day comes when it really can.................we're toast !

So , AI learns from all the data it acquires ........and it's errors.....just more data ....not right or wrong...........just more data....

 

Yes, I was wondering about that.

I've liked all of your posts about your collection, and this is added to the "like" list.

 

Thank you for your compliments! I just want to share my knowledge and the joy of collecting old books. I do not have have many fellow collectors in my area. Maybe it is time to change to comic books or Harry Potter editions…

 

People have started posting AI generated stories about events that didn't happen. There was an AI story with photos of a Burning Man festival that took place in the 60s. Fact is, BM didn't start until 1986. If you looked closely at the photos you notice people with three legs or half a face in the crowds. Yet, a lot of young people were commenting on how much fun it must have been. They never fact checked. As far as I can see, history is out the window and we're living in a Lord of the Flies world with a bunch of youngsters making stuff up as they go along.

 

When it figures out that hallucinations aren't acceptable, we're in trouble.

 

Your experience and knowledge in the field of information technology and artificial intelligence clearly brings valuable observations about the use of such tools for finding information about books. Indeed, the application of artificial intelligence, including language models, can be a powerful and promising tool to facilitate research. However, as you have pointed out, the issue of the veracity and accuracy of the information provided by artificial intelligence is significant. Especially in the context of the fabrication of "facts" that sound authoritative but are in fact unreliable, this raises serious concerns. Recently, for example, I was asked to write about personal qualities, so I used artificial intelligence and a service and I got an excellent job. When using AI to research old books and other information, you always need to be especially vigilant and check the results with other, more reliable sources. Trust in quality databases and library resources is also an important element in the research process. The possibilities of artificial intelligence in finding information about books are certainly exciting, but it is our responsibility as users to judiciously and critically evaluate the results to ensure the accuracy and validity of the information received.

 

but it is our responsibility as users to judiciously and critically evaluate the results to ensure the accuracy and validity of the information received.

unfortunately we have abrogated that responsibility , in favour of a quick fix..!

 

And "I found it in a Wiki so it has to be true" is a Thing.