I kid you not, I read “The Elder Scrolls”, and was very amazed that someone tried to figure out the different authors of the many books in the game... time to rest I suppose.
He is certainly the most well known and responsible for much of the more interesting and widely discussed lore. Especially regarding religion and metaphysics.
But he is not the only one. For example, the original writer, who has produced some of my favorite in-game books, is Ted Peterson[0]. There is a great interview with him on YouTube[1].
People make it sound like Bethesda games are so buggy they have more bugs than game. Maybe this is because they spend all their time writing lore instead of reading QA reports.
One question I have whenever algorithms are used to try to figure out multiple authors: was this algorithm tested on works we know that were written by a single author?
People change. Maybe the scribe developed back problems or RSI or arthritis. Even writing styles can change. What is the range of single person variability, and how does that compare with the variability we see in the work?
That's mentioned but dismissed. I would think age might be a factor though. Some scribe took a break for whatever reason returning to the task after 10 years or so?
> Although one cannot rule out completely that the clear separation between the two halves of the manuscript and the difference in writing patterns are due to a change of writing implement (a different pen), writing fatigue or some injury that the writer suffered when moving on to the second half of the manuscript, the more straightforward explanation is that a change in scribes occurred.
A similar question is around what size training data do we have where we know works had different authors.
My understanding is there's very little original material from this era to work with - as all biblical canon is replicas, from a handful, to many dozens of generations / copies distant.
Yes, that was my understanding of TFA too - but the question remains, unless more than a handful of authors are easily distinguishable, and there's a different and reliable way to identify those distinct authors (did they sign their work?) then we have a very small sample set.
Writing was not a common skill, so my naive extrapolation of that fact is that there would be few teachers, and consequently more likelihood that two random authors would present material that, prima facie, looked similar.
It's fascinating work, to be sure, though I do wonder about the usefulness of identifying how many authors were involved in penning these various parchments.
I can't find many authoritative sources, but it stands to reason that in a mostly agrarian society, there'd be fewer people who could write, would have access to good writing materials, specifically materials with longevity, and get those materials into hermetically sealed environments that would ensure their integrity for a couple of millennia.
Oral tradition put into text, so not entirely the same as grouping the writers. More like who transcribed what. Maybe the difference was who heard the story in a given pattern.
Given the long history of Akkadian and cuneiform usage by all of their neighbors, it's not as if they had to forsake writing. Just look at the ruins of Ebla for a comparison of civilization at that time.
So PCA, k-means clustering and manual visual inspection counts as AI now?
There's no ML involved in the approach, just plain old statistics and feature extraction. I fail to see how any of this relates to AI unless we broaden the definition to a point where it becomes useless and everything that uses an algorithm becomes "AI"...
After years of not working in statistics I've decided to learn something new, so I've purchased a Coursera MOOC course on that new kid on the block - Machine Learning (now rebranded as AI I think).
To my surprise (I wanted to learn something new) all the "old stuff" was there, ML turned out to be some variant of statistical inference with all its typical methods applied (like logistic regression, clustering, etc.).
The new things were: deep neural networks ("normal" NN but with more layers) and total lack of thinking in terms of model - ML keeps adding more and more variables to the model, as much as data we have, no matter if it make sense or not. If the "model" is over-fitted, then, well, let's add some additional quadratic term to the equation, it seems to be helping... Maybe... Sometimes.
So, yes, AI is just old stuff, rebranded and fed with much, much more data. From what I see AI hasn't changed much over the last years. AI seems to be pretty effective in everything that is machine generated data set, as it is easily quantified and has simple representation - That's why it is so good in tracking people - there is a huge amount of data coming from people devices, they are nice and clean.
It is also good in picture analysis - again data are pretty clean (pixels) and easily quantified, what is even more important, so it easy to "train" AI.
When it comes to something more complicated like, for instance, purchase recommendation we get what we get: you bought a fridge, good, you will see fridge adds for next half a year.
Looking beyond the term and you realise the approach is neither impressive or novel.
Such rebranding is endemic in article headlines like this. The term “AI” may be in vogue but 99.9% of the time it’s really just putting lipstick on a pig.
23 comments
[ 3.3 ms ] story [ 48.9 ms ] threadBut he is not the only one. For example, the original writer, who has produced some of my favorite in-game books, is Ted Peterson[0]. There is a great interview with him on YouTube[1].
[0] https://en.m.uesp.net/wiki/General:Ted_Peterson
[1] https://youtu.be/AegrlefXuwE
https://www.imperial-library.info/
https://elderscrolls.fandom.com/wiki/CHIM
People make it sound like Bethesda games are so buggy they have more bugs than game. Maybe this is because they spend all their time writing lore instead of reading QA reports.
People change. Maybe the scribe developed back problems or RSI or arthritis. Even writing styles can change. What is the range of single person variability, and how does that compare with the variability we see in the work?
> Although one cannot rule out completely that the clear separation between the two halves of the manuscript and the difference in writing patterns are due to a change of writing implement (a different pen), writing fatigue or some injury that the writer suffered when moving on to the second half of the manuscript, the more straightforward explanation is that a change in scribes occurred.
My understanding is there's very little original material from this era to work with - as all biblical canon is replicas, from a handful, to many dozens of generations / copies distant.
There's plenty of scribal material in the Judaic world, they should try their algorithm on it and see if it works.
Writing was not a common skill, so my naive extrapolation of that fact is that there would be few teachers, and consequently more likelihood that two random authors would present material that, prima facie, looked similar.
It's fascinating work, to be sure, though I do wonder about the usefulness of identifying how many authors were involved in penning these various parchments.
was it? I think the society back then was pretty advanced, i.e. literate.
It wouldn't tell them anything much. The script used nowadays is entirely different from the DSS.
There's no ML involved in the approach, just plain old statistics and feature extraction. I fail to see how any of this relates to AI unless we broaden the definition to a point where it becomes useless and everything that uses an algorithm becomes "AI"...
To my surprise (I wanted to learn something new) all the "old stuff" was there, ML turned out to be some variant of statistical inference with all its typical methods applied (like logistic regression, clustering, etc.).
The new things were: deep neural networks ("normal" NN but with more layers) and total lack of thinking in terms of model - ML keeps adding more and more variables to the model, as much as data we have, no matter if it make sense or not. If the "model" is over-fitted, then, well, let's add some additional quadratic term to the equation, it seems to be helping... Maybe... Sometimes.
So, yes, AI is just old stuff, rebranded and fed with much, much more data. From what I see AI hasn't changed much over the last years. AI seems to be pretty effective in everything that is machine generated data set, as it is easily quantified and has simple representation - That's why it is so good in tracking people - there is a huge amount of data coming from people devices, they are nice and clean.
It is also good in picture analysis - again data are pretty clean (pixels) and easily quantified, what is even more important, so it easy to "train" AI.
When it comes to something more complicated like, for instance, purchase recommendation we get what we get: you bought a fridge, good, you will see fridge adds for next half a year.
Such rebranding is endemic in article headlines like this. The term “AI” may be in vogue but 99.9% of the time it’s really just putting lipstick on a pig.