Thanks for explaining my misinterpretation of what you were saying. I’m still extremely uncomfortable with the way you are using the word “impartial” here. The word has a strong positive-value connotation that I just don’t think is applicable here.
The untrained network, the dataset, and the training algorithm: these are the things you seem to be calling “impartial”, but the untrained network is not the kind of thing capable of partiality or impartiality, neither is the data-set, nor is the training algorithm. The result, however, the trained network, is inevitably highly partial (highly biased, highly prejudiced) under current practices, while the hucksters pitching it are reiterating false claims of impartiality. So even if this is elaborate irony on your part, I’m still worried about it being misleading.
Right, that’s how the MLs pick up on all the biases and prejudices express on today’s internet, and these are impossible to root out afterwards.
No, no, I was not being ironic at all, I just tried to elaborate and, as you said, be more specific about what I meant with the word. 
“Impartial” is one of those words that might have slightly different connotations depending on the original language of the speaker.
Since you mentioned it though, a small clarification again on what I had in my mind:
The untrained network, imho, is the only thing I consider impartial because it will “accept” the data, “process” it via the usage of the training algorithm.
If the data says “fire-breathing dragons are real” the network will not make a decision otherwise (even though they are not).
It the training algorithm says “we want you to focus on a specific aspect of the dataset and ignore or give less value to some other data” the network will not object, no matter how absurd or false the training focus might be.
You could say that “the untrained network is not the kind of thing capable of partiality or impartiality” since it is “just there” as “tabula rasa” of sorts, but this is not true with many such networks since you can, indeed, weight the nodes differently from one another creating a partiality within the network itself. My information on the matter is a bit dated, but if node-weighting was an acceptable idea 20 years ago, I hardly think that this or something analogous is not possible now.
The dataset and the training algorithm can definitely be partial and there are very recent “news” on this:
Quite the mess as you can imagine and the usual “internet dwellers” are creating quite the noise with it, but in this case there was indeed some issue with the training algorithm.
Brandon Sanderson and his crowdfundings are boggling the internet again:
Eventually the website worked and within 2 hours he has more than twice than the 2 million dollars originally required:
And his work is worth every penny 
very interesting:
The 4 prospective coalition parties are talking again. It has become clear that Wilders won’t become Prime Minister (which is a great relief to me). In fact they agreed that all of the 4 party leaders will stay in parliament and they won’t take a seat in the new government.
In the coming months negociations will continue on the main goals towards which the new government is supposed to work, and I suppose they will also negociate on the composition of the new government.
I wonder if the new government will be a bunch of high ranking civil servants that will get ordered around by shifting majorities in parliament. It might work, or it might not. It’s all new territory for our democracy.
I think it’s interesting that up to now, these negociations for a conservative-leaning center-right government are led by members of the largest opposition party, which is a progressive-leaning center-left party.
Isn’t this satire?
Yeah, at least I took it to be so. Making fun of the arbitrary and ill-advised “rules” made up by prescriptivists is fun
Sarcasm is the lowest form of wit, so there is obviously no way this could be satire.
Various versions of this have been circulating for years. I’m cynical on the subject. Trying to improve people’s writing is a waste of time. Most don’t care, and those who do care do not need to be harangued. The U.S. government has a “Plain Language” writing initiative with a small guide to better writing. The guide invites corrections. When I first saw it, three or four years ago, I sent a two-page list of corrections (major and minor) as an email attachment to the address provided. Got no response. A year later, no corrections had been made.
Mmm souka.
People say that politicians are puppets, but what if we could actually vote puppets for politicians? 
ask Challita what motivated the study, and he says it was simple curiosity. “I think people should understand science doesn’t have to be very serious,” he says. “It can be fun, too.”
well, this is going to be fun:
Especially combined with this:
Tesla’s stock might be one of the biggest bubbles of this century.
before:
after:
