AI Scientists Are Told to Remove 'Ideological Bias' from Powerful Models

https://www.wired.com/story/ai-safety-institute-new-directive-america-first/

by BLKNSLVR • 1 year ago

31 14 1 year ago

BLKNSLVR2• 1 year ago

I mean, genius really. AI is a black box for most people, they just blindly trust what comes out, like reading Wikipedia or Twitter.

If the government can control AI's biases then they can point to AI answers as justification for their positions.

trod12342• 1 year ago

In reality that would be false justification. To have a true justification you would need to have a objectively tied definition.

Circular reasoning and hegel's dialectic (outsides its intended use) is fallacy/deception.

BLKNSLVR2• 1 year ago

Absolutely, but where does 'reality' come into it?

fennecbutt2• 1 year ago

Nah we're all dumb meatbags that even without technology, have proven time and time again that we'll propagate false information, hold biases and prejudices and kill each other because of our differing skin colour, sex, sexual preferences, nationality, favourite sports team...basically any difference granular enough to be considered a tribe.

abraxas• 1 year ago

Sadly for the current regime, reality has a liberal bias. I guess US based AI models will become Q-Anon nonsense spewing hatebots.

01HNNWZ0MV43FF• 1 year ago

Back in my day there was just something nicer about knowing it was a real human telling me to kill myself for being gay

clown_strike• 1 year ago

Liberals have no actual logic, only persuasive rhetoric. They get away with it by euphemistically shaping a homophonic-oppositional language we conflate with our own.

Ban the logits corresponding to their predictable ad hominems and you get a radically different LLM experience. You can literally eliminate racism just by banning use of the word.

namaria• 1 year ago

Can't gerrymander or electoral college your way out of that one...

trod12342• 1 year ago

Told to Remove "Ideological Bias"

Well that will never happen except by firing squad. You can only ever be free of perceptual bias (which is ideological bias since its derived in part from your culture and identity), when you lack a sense of self and are unable to interact with your environment... that usually means your dead or pretty close to it.

If the smartest philosophers and scientists in the world, for all time, have been unable to separate their own perceptual biases, what hope is there for anyone to do this for some powerful model and have it retain usefulness.

This is just laughable.

zonkerdonker• 1 year ago

I agree. To perceive is literally to bias. Perception, and by extension, ideaology cannot exist in a vacuum.

Not even on just a human scale either. A camera "perceives" the world, but it's limited by the sepctra of its capture capabilities, its resolution, and lighting etc.

Is the idea of a picture biased? Should black and white cameras be banned for ideologically innacurate representations of the world?

fennecbutt2• 1 year ago

Ridiculously hard task seeing as the training data is limited by humanity.

We want it to act like a human so it's gonna act like a human, unfortunate but true.

Rings very similar to those that complain about say a visual model only being 90% accurate. Well yeah...the whole architecture of this stuff mimics biology and we certainly jump at shadows more often than not. So 90% accuracy is a hell of a lot better than we do.

42772827• 1 year ago

CCP Shit

stuaxo• 1 year ago

Remove bias - from a thing that is made up of people's biases.

gavmor• 1 year ago

In the days when Sussman was a novice, Minsky once came to him as he sat hacking at the PDP-6. "What are you doing?", asked Minsky.

"I am training a randomly wired neural net to play Tic-tac-toe", Sussman replied.

"Why is the net wired randomly?", asked Minsky.

"I do not want it to have any preconceptions of how to play", Sussman said.

Minsky then shut his eyes.

"Why do you close your eyes?" Sussman asked his teacher.

"So that the room will be empty."

At that moment, Sussman was enlightened.

AI Scientists Are Told to Remove 'Ideological Bias' from Powerful Models

Related Stories