[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/sci/ - Science & Math


View post   

File: 697 KB, 713x741, 1558025454367.png [View same] [iqdb] [saucenao] [google]
10717495 No.10717495[DELETED]  [Reply] [Original]

>> No.10717504

Because they aren't smart enough to overcome the more primitive biases they acquire by mimicking the training data.

>> No.10717512
File: 52 KB, 713x599, 1559152820870.png [View same] [iqdb] [saucenao] [google]
10717512

>>10717504
So your saying that they aren't innefficiant enough yet to not hurt peoples feelings. Doesn't that defeat the purpose of an A.I.

>> No.10717515

>>10717512
>Doesn't that defeat the purpose
Yes.

>> No.10717518

>attributing technology failures to malice
why do brainlets do this

>> No.10717528

>>10717518
Because they aren't smart enough to overcome the more primitive biases they acquire from "experience."

>> No.10717547

>>10717495
Because the AI is correct and properly identified the bipedal monkey species more commonly know as the the North American Pavement Ape.

>> No.10717573

gorillas are cool and handsome though

>> No.10717704

>>10717547
>y identified the bipedal monkey species more commonly know as the the North American Pavement Ape.
Okay this is cool, we have a human being that's as dumb as AI. Not as good as an AI smart as a human, but still cool. What do we ask it?

>> No.10717713

They're input output and run by retards. A lot of image classifiers are designed to give a lot of weight to colors rather than edges. You can overlay an elephant skin texture over cat lines and it will come up as elephant.

>> No.10717757

>>10717495
What happens if you show it an albino gorilla?

>> No.10717762

>>10717512
your troll has the underlying presupposition that AI are actually anywhere as efficient as humans. they aren't

>> No.10718918
File: 91 KB, 1280x720, white_ape.jpg [View same] [iqdb] [saucenao] [google]
10718918

>>10717495
Unbalanced training data sets basically
Same reason many Chinese women can face unlock each others' iPhones

>>10717713
This also
some guys at MIT had an AI (amazon's, iirc) identifying a gun as a turtle

>> No.10718958

>>10717504
Is it really completely inconceivable that black people just kinda look like gorillas, objectively?

>> No.10718989

>>10717495
>implying gorilla and nigger is different.

>> No.10718998

>>10718989
Gorrilla has higher IQ

>> No.10719015

>>10717495
Because they haven't been altered to lie yet.

>> No.10719032

>>10718958
That's obviously the simple explanation, but a better question is "Why is the AI unable to differentiate niggers and gorillas?" in the sense of it not being good enough to do so. You and I obviously can, so why can't it? The obvious answer is that it's not smart enough yet, so why not?

The unobvious answer: because Google is shit.

>> No.10719040

>>10719032
It's not about being "smart" enough. AI has different biases than humans do and there are plenty of targets it would outperform humans on because of details our own visual interpretation systems have more problems with.

>> No.10719746

>>10717495
AIs are logical, they struggle to identify differences between 'similar' things, in this case some African Americans have facial features that more resemble a gorilla and the AI cant (yet) separate them. If this goes viral then silicon valley is going to be scared shitless of creating a 'racist' AI and someone is going to specifically train the AI to identify differences between humans and other apes.

>> No.10720265 [DELETED] 

>>10718918
It looks like a black albino.

>> No.10720270
File: 461 KB, 272x480, 1559322539480.webm [View same] [iqdb] [saucenao] [google]
10720270

>>10718998
This

>> No.10720322
File: 58 KB, 640x360, Hypereality.jpg [View same] [iqdb] [saucenao] [google]
10720322

>>10719746
>scared shitless
This is nothing. A way heavier issue is how we've begun literally determining sentences for defendants with proprietary (meaning privately owned, copyright protected, and nobody can look at its code to try to figure out how it arrives at its conclusions) AI software.
https://www.technologyreview.com/s/612775/algorithms-criminal-justice-ai/
If you're black you're pretty well fucked since these programs will identify you as having a higher recidivism rate and will have the judge sending you to prison instead of getting to do cushy community service picking up a couple pieces of trash at a park for a weekend.

>> No.10720334

Maybe not racist but actually accurate.

>> No.10720341
File: 30 KB, 200x200, ur.jpg [View same] [iqdb] [saucenao] [google]
10720341

>>10720334
Just because you're technically correct about something doesn't mean you aren't being racist. It's not just a true or false situation because you're still deciding which ideas to bring up and which ideas to ignore.
Employers do this all the time when they go out of their way to notice any minor infraction that comes up for employees they don't like while passively ignoring the same for employees they do like.

>> No.10720505

>people honestly think black humans looks anything remotely close to gorillas
almost as retarded as unironically thinking asians all look the same.
Protip: AI has trouble differentiating blacks and gorillas for the same reason your dumb asses have trouble differentiating asian faces apart. Unfamiliarity and your "training" data set never accounted for it.

>> No.10720531

google has a very real interest in keeping PoC down. it's a fundamental part of their business

>> No.10720538

>>10717495
how many black women do you know that don't act and sound and look like apes? they even make a sound like an ape, it sounds like okuurrrrr or esketet.

>> No.10720540

>>10717518
>failures

>> No.10720541

>>10717495
Well google isn't that great, when I input images of myself, google images recognizes me as a girl when I'm a guy :/