Why would we never know? Quoting from the blog post: "The specific reasons given were that [...]" followed by a bullet list. Your guess is unwarranted speculation.
1,599 karma · joined December 22, 2008
sums = [s for s in [0] for x in data for s in [s + x]]
The most efficient language is the least compressible language only in a narrow and arbitrary sense of efficient. There are many considerations such as what is efficient for the speaker, the hearer, redundancy to noise, efficiency with respect to particular purposes, etc. We can assume that natural languages will generally make a good trade-off across these factors, and searching for the most efficient language in one particular narrow sense is not very useful. Moreover, compression of text focuses only on surface form, completely ignoring the dimension of meaning.
The queen - woman example is when you try to make a model of word semantics, such as with word2vec. In a document classification task the vectors represent documents.
I can't find anything non-biased though, everyone just wants to push the female victim narrative.
Slow down there ... Most linguists (or a substantial minority) do not subscribe to the theoretical idea of universal grammar. Parsey McParseFace is only an incremental improvement in decades of statistical parsing. The problem of natural language understanding remains largely unsolved. Like other deep learning models, it is not hard to come up with adversarial examples that will confuse the parser but not any competent human language user. The parser is only as good as the data it is trained on; this data is expensive to acquire but there is never enough. Additionally there are a myriad other kinds of ambiguity in language beyond the sentence-level syntactic ambiguity which is resolved by a statistical parser such as Parsey McParseFace.
In my opinion Lojban provides a good illustration of just how hard it is to remove ambiguity from human language.
Think for example of an inductive bias. If I see a couple of white swans, I may conclude that all swans are white, and we all know this is wrong. Similarly, I may conclude the sun rises everyday, and for all practical purposes this is correct. This kind of bias is neither wrong nor right, but, in the words of the article "a necessary prerequisite for intelligent action", because no induction/generalization would be possible without it.
There are undoubtedly examples where the prejudiced kind of biases lead to both truthful and untruthful predictions, but that seems beside the point, which is to design a system with the biases you want, and without the ones you don't.
"In AI and machine learning, bias refers generally to prior information, a necessary prerequisite for intelligent action (4). Yet bias can be problematic where such information is derived from aspects of human culture known to lead to harmful behavior. Here, we will call such biases “stereotyped” and actions taken on their basis “prejudiced.”"
This definition is not unusual. This is about inferences that are wrong in the sense of prejudiced, not necessarily inaccurate.