441–450 of 584 posts
What if people start publishing hate speech but replace the hateful words by, say, 'Alphabet' ?
Will the AI learn that Alphabet is a hateful word ?
Surely someone with enough resource should be able to do that.
Earlier quoted context omitted.
>The obvious answer is a well-informed populace that isn't easily swayed by emotion and fear, and has the time and inclination to do their own fact checking. Trouble is, that's obviously a pipe dream. The other trouble is that one side is actively opposed to a "well informed populace" and unironically calls secondary education "liberal brainwashing camps". If you want a well informed populace, you have to privilege t…
Well... the "other side" would oppose informing the populace that the equality dogma is false (or at least unproven), so it isn't just the far right that likes their myths. Attemps to ban things that are false runs into the difficulty that you need to first determine what is false, which is hard to do if one of the answers might contradict an important value. I should stop replying to political posts with more politi…
What do you mean by the equality dogma?
Earlier quoted context omitted.
It seems we are using black for bad and white for good in English, which is strange since we had never used them before. Also, I feel like blocking and allowing makes sense when it comes to language, because bad and white mean the same. But when it comes to code, which the rest of the world uses black to mean bad and white to mean good, I don't know if that makes sense.
> we had never used them before I think that language goes back eons. I think the black/white metaphor is employed in the book of Genesis/B'reshith for instance. People have been scared of the dark for more of human history than not. Nighttime is black, when things putrefy they turn black. Uncorrupted snow is white. The examples from nature are legion. What’s weird is using these terms for people. Maybe the more sens…
A nonsensical "racial" hierarchy based on nominal colours is the SOURCE of the problem.
For what it's worth, the labels "white" and "black" were NOT chosen randomly by the architects of racial hierarchy (see Bacon's rebellion).
It was a very effective use of symbolism, to assign and label one part of humanity as the untermensch, and thus justify their enslavement and exploitation.
Changing branch names and policing language is akin to bandaid on a festering wound.
These cases have lots of exposure and people who notice. So the real QA is in these edge cases where customers find bugs. I worry about how not so visible algorithms.
I had a company market to me about sentiment analysis. I asked their accuracy and they answered that they can’t actually test but their training model has x%. So to make decisions accordingly. But then I had coworkers doing stuff like what percent had what sentiment vs the other with no assumptions.
I think the problem is people are kind of bad with probabilities so something either is or isn’t. And I seem to witness a gulf where there are people who kind of understand and then people who don’t but trust the output.
Earlier quoted context omitted.
> A bit far-fetched; to really resonate the experiment should include some invented cultural baggage which is difficult to do on the spot.) I'll give you a concrete example from the world of finance. In options there are two main forms of exercise. You can have options which can be exercised at expiry and options that can be exercised at any point.[1] When the economist Paul Samuelson started to work on Wall street h…
Seems relevant. How did you feel about that term? Did it trigger slight unease deep down? I do suspect that the fact that the US doesn’t have a recent history of being oppressed might discount it a little (if one is considered a dominant power, one can afford the luxury of taking some mild negging in stride). On the other hand, in case of e.g. “blacklist” — it meaning a list of people or entities to deny service — co…
Likewise "hysterical" derives from "hyster" (Uterus) and stems from the idea that the female reproductive system made women mad and incapable.
I feel like one of the things that people miss often in discussing these topics is the asymmetry of power. If you're one of the people who has historically been discriminated against you may well have different feelings from people who has historically doing the discriminating.
We should listen to each other more and if these terms are a barrier then there really is no harm to using other terms. I think using friendly language is the least I can do from the point of view of being courteous to others.
You should view ML as an enhancement for humans.
So the problem here is that no ML system without essentially being an AGI is capable of fully comprehending context, in this case use of "black" and "white" in Chess that has absolutely no racial context.
So the mistake here is (likely) automated action against the channel. That should never happen. At least, not for this particular signal.
What should happen is the channel should be flagged for review by a human. Any human would appreciate the context and take no action.
This is the "enhancement" part. With no automated assistance, a human might be capable of moderating (made up numbers) 100 channels. That doesn't scale at Youtube scale. ML systems might enhance that human's moderating ability to 10,000 channels.
Even better, by flagging for review, you're actually producing training data for your ML. The P-R for your automated disciplinary system needs to be incredibly good before you take humans out of the equation. False positive bans or blocks on channels is bad for the individual channel. It can be devastating in fact. The corporate view is that "one channel doesn't matter". And they're right... to a point. At some point it undermines confidence in the platform and then it does matter.
Earlier quoted context omitted.
>The obvious answer is a well-informed populace that isn't easily swayed by emotion and fear, and has the time and inclination to do their own fact checking. Trouble is, that's obviously a pipe dream. The other trouble is that one side is actively opposed to a "well informed populace" and unironically calls secondary education "liberal brainwashing camps". If you want a well informed populace, you have to privilege t…
It is insane how people will always continue to blame the other side even when replying to a comment pointing out how basically everyone on the political spectrum currently feels this way.
So 18% did include obvious racist language or hate speech?
Earlier quoted context omitted.
Seems relevant. How did you feel about that term? Did it trigger slight unease deep down? I do suspect that the fact that the US doesn’t have a recent history of being oppressed might discount it a little (if one is considered a dominant power, one can afford the luxury of taking some mild negging in stride). On the other hand, in case of e.g. “blacklist” — it meaning a list of people or entities to deny service — co…
I don't really have a credible opinion on this topic. That said there are lots of examples of this type of thing in language. Eg in ancient Rome they used to think left-handedness was unnatural, so to this day there are negative words like "Sinister" which literally in Latin means "left" and positive words (dextrous, dexterity etc) which derive from "dexter" which is Latin for "right". Also "adroit" in English means…
> We should listen to each other more and if these terms are a barrier then there really is no harm to using other terms.
Agreed.
I’m personally torn on this issue. On one hand, the fact that these words rub people wrong is an indicator of a deeper problem—in an ideal world, no one would take a term personally because they’d know it denotes a technical concept and isn’t intended to put anyone down; on the other hand, it’s not really that difficult to stop using some terms in new codebases.
On one hand, we should get input from those directly affected by this; on the other hand, those affected may choose to stay silent despite possible discomfort and treat it as a problem with their insecurities rather than the industry.