Thread opportunity to redpill: "Two Google engineers resign over firing of AI ethics researcher". AI are racist, because they learn about the world. Instead of garbage in/garbage out, reality in/reality out.

submitted by Fitter_HappierWhite Nationalist from old.reddit.com

all 11 comments
sorted by:
top
newinsightfulfun

[–]DragonerneJesus is white 1 insightful - 1 fun1 insightful - 0 fun2 insightful - 1 fun -  (10 children)

Bad take.
If historical data is biased by historical biases, then the AI will just repeat those biases if you don't fix it.

Imagine if they used social media censorship data to construct an AI that censor people. Since the data is biased, the AI would be biased. It is really simple.

[–]Fitter_HappierWhite Nationalist[S] 1 insightful - 1 fun1 insightful - 0 fun2 insightful - 1 fun -  (9 children)

historical data is biased by historical biases

I said that. But if you put in unbiased data you'll get an accurate reflection of reality. i.e. if you put in mugshots of who is really committing the crimes you'll get a good bad-guy detector. If you put in bullshit TV hero/villains you'll get a bullshit one.

[–]DragonerneJesus is white 1 insightful - 1 fun1 insightful - 0 fun2 insightful - 1 fun -  (8 children)

There is no such thing as unbiased data.

"Who is really committing the crimes" is a human decision, hence we introduce bias into the data. If you put in mugshots of who is really committing the crimes, you'll get a classifier that can determine who humans have judged committed the crimes.

Last year we had a deranged muslim trying to murder our far right politician Rasmus Paludan. He told the police (who shot the man) that he wanted to knife down Paludan. He told the courts that he wanted to murder Paludan. The judges decided that he did not intend to kill Paludan. But he was still sentenced to 1 year prison for "tough violence".

The AI would replicate that court finding. It would replicate the human bias.

[–]Fitter_HappierWhite Nationalist[S] 1 insightful - 1 fun1 insightful - 0 fun2 insightful - 1 fun -  (7 children)

What is truth anyway, right?

[–]DragonerneJesus is white 1 insightful - 1 fun1 insightful - 0 fun2 insightful - 1 fun -  (6 children)

You get my point, right?

[–]Fitter_HappierWhite Nationalist[S] 1 insightful - 1 fun1 insightful - 0 fun2 insightful - 1 fun -  (5 children)

There is no such thing as unbiased data.

I disagree with your premise that "There is no such thing as unbiased data."

[–]DragonerneJesus is white 1 insightful - 1 fun1 insightful - 0 fun2 insightful - 1 fun -  (4 children)

Data will always be biased on some capacity relative to some notion. But this premise is not important for the broader subject. We can consider your premise that there exists some real truth that is representable with data. The issue is then the gathering of that data. The methods of gathering will likely bias the data etc.

[–]Fitter_HappierWhite Nationalist[S] 1 insightful - 1 fun1 insightful - 0 fun2 insightful - 1 fun -  (3 children)

The methods of gathering will likely bias the data etc.

OR NOT DEPENDING ON HOW YOU DO IT. Jesus, that's my whole point. Are you telling me there are NO honest data collectors, in the universe?

[–]DragonerneJesus is white 1 insightful - 1 fun1 insightful - 0 fun2 insightful - 1 fun -  (2 children)

Are you telling me there are NO honest data collectors, in the universe?

If I want to create an AI that moderates content. How would you gather data for censoring posts that is "honest"?

[–]Fitter_HappierWhite Nationalist[S] 1 insightful - 1 fun1 insightful - 0 fun2 insightful - 1 fun -  (1 child)

When you say honest do you mean factually correct or truly believed?

I can say honestly aliens live in my butt, but that isn't true.