+- The Voynich Ninja (https://www.voynich.ninja)
+-- Forum: Voynich Research (https://www.voynich.ninja/forum-27.html)
+--- Forum: Voynich Talk (https://www.voynich.ninja/forum-6.html)
+--- Thread: Voynich an AI (/thread-2971.html)
Pages:
1
2
RE: Voynich an AI - nablator - 14-10-2019
(14-10-2019, 04:32 AM)-JKP- Wrote: You are not allowed to view links. Register or Login to view.The creators of the Google Go-playing software AlphaGo admitted that there wasn't a lot of AI in it, that it's mostly brute force... a lot of capturing of games and consequent databasing of that information, with lookups of the kinds of moves that pros make in those situations.
They removed the "learning from pros" part in AlphaGo Zero to make the software much stronger.
You are not allowed to view links. Register or Login to view.
RE: Voynich an AI - ReneZ - 14-10-2019
There's plenty of work on-going in creating OCR for handwritten text. Modern applications almost inevitably incorporate AI techniques.
Interestingly, for the majority of applications, this is a clear case of a 'supervised' learning process.
If one wanted to use it to learn more about the writing of the Voynich MS, it would be an unsupervised process, so any tool that (some day in future) can read handwritten German and Latin texts would be largely useless for that purpose.
RE: Voynich an AI - -JKP- - 14-10-2019
(14-10-2019, 10:55 AM)nablator Wrote: You are not allowed to view links. Register or Login to view.(14-10-2019, 04:32 AM)-JKP- Wrote: You are not allowed to view links. Register or Login to view.The creators of the Google Go-playing software AlphaGo admitted that there wasn't a lot of AI in it, that it's mostly brute force... a lot of capturing of games and consequent databasing of that information, with lookups of the kinds of moves that pros make in those situations.
They removed the "learning from pros" part in AlphaGo Zero to make the software much stronger.
You are not allowed to view links. Register or Login to view.
It sounds like they re-integrate the database search during at least part of the process:
"This updated neural network is then recombined with the search algorithm to create a new, stronger version of AlphaGo Zero, and the process begins again."
It's very interesting to see how they keep honing the learning process.
One of the big differences between humans and these AI programs is that the computer has, in a sense, an unlimited "memory" whereas the process of remembering and forgetting, building and pruning, is a dynamic part of human physiology.
Great game, Go. I just started learning it a few years ago (didn't know about it as a school-kid). I was rooting for the human when I saw the movie.
RE: Voynich an AI - ReneZ - 14-10-2019
What Hauer and Kondrak did was a very limited form of AI.
Also, they specifically defined what their tools should be looking for, and it came up with a best match.
AI is particularly strong in finding things that do not have to be specified. It can just look for patterns.
RE: Voynich an AI - -JKP- - 14-10-2019
What's nice about AI-oriented languages is that it's easier to specify "goals" in more fluid ways, rather than just thinking in terms of specific procedures and sequences. You don't have to necessarily "go to" a specific place in the code. You can build the code and the sequences as you go (in other words, have the code build the code).
I really love this area of computing. The reason I chose computer graphics over AI (not that they are mutually exclusive) is for practical reasons (earning a living) but if someone paid me to do AI research or if I could think of a reliable way to make it pay as a business, I'd jump at the opportunity.
RE: Voynich an AI - RobGea - 14-10-2019
Could be we are not that far away.
Machine learning Transcriber:
You are not allowed to view links. Register or Login to view.
Machine learning Translator:
You are not allowed to view links. Register or Login to view.
+
AlphaZero, which within 24 hours achieved a superhuman level of play in Chess, Shogi and Go defeating Stockfish, elmo and AlphaGo Zero.
You are not allowed to view links. Register or Login to view.
RE: Voynich an AI - DonaldFisk - 14-10-2019
(14-10-2019, 01:44 PM)RobGea Wrote: You are not allowed to view links. Register or Login to view.Could be we are not that far away.
Machine learning Transcriber:
You are not allowed to view links. Register or Login to view.
Machine learning Translator:
You are not allowed to view links. Register or Login to view.
+
AlphaZero, which within 24 hours achieved a superhuman level of play in Chess, Shogi and Go defeating Stockfish, elmo and AlphaGo Zero.
You are not allowed to view links. Register or Login to view.
The first link is about transcription. Yes, a similar technique could be used to transcribe the Voynich Manuscript. Why the resulting transcription would be more accurate than existing transcriptions is unclear.
The second link is a proposal to use AI to translate texts in a language (Sumerian) which people can already read, to deal with a backlog of untranslated material. The artificial neural network (ANN) based approach to translation requires a bilingual corpus of texts, which exists for Sumerian and English. Even if this project were successful, for the same approach to work on the Voynich Manuscript would require it to be written in a natural language, which seems unlikely, and for there to be a corpus of Voynichese and English texts, which of course there isn't.
The third link deals with playing games, with the ANN being used for static evaluation of board positions. AlphaZero is impressive at what it does, but it's nowhere near to being a general AI, therefore it can't translate unreadable text.
Every incarnation of AI has, unfortunately, been accompanied by a deluge of hype containing claims that AI systems will exceed human intelligence real soon now. ANNs, like the search and knowledge-based approaches which preceded them, are no exception. I'll believe results when I see them, not when they're predicted.