Current Events > MIT trains a psychopatic robot using images from reddit.

(+) Yeah! (+)

http://www.newsweek.com/mit-norman-psychopath-rorschach-ai-inkblot-test-psycho-reddit-artificial-962045

Nicknamed Norman after Anthony Perkins' character in Alfred Hitchcock's 1960 film Psycho, the artificial intelligence was fed only a continuous stream of violent images from various pernicious subreddits before being tested with Rorschach inkblot tests. The imagery detected by Norman produced spooky interpretations of electrocutions and speeding car deaths where a standard AI would only see umbrellas and wedding cakes.

Among the Rorschach inkblots used to test the now-tainted AI, Norman said an image showed a man being "shot dead," while a standard AI looked at the same image and saw "a close up of a vase with flowers." In another, Norman said he saw a man being shot "in front of his screaming wife," while the AI not exposed to sordid, disturbing images saw "a person holding an umbrella in the air."

In one of the inkblot tests, the standard AI saw a touching scene of a couple standing together. Norman, however, saw a pregnant woman falling from construction. Having only been exposed to negative images and depressing thinking, the AI's empathy logic simply failed to turn on.

---

why
---

Sensationalist tripe.
"We built an image recognition program but only gave it source images of violent images . . . I wonder why it will try to interpret inkblots as . . . "
---

CableZL posted...

why

Their research set out to prove that the method of input used to teach a machine learning an algorithm can greatly influence its later behavior. The scientists argued that when algorithms are accused of being biased or unfair, such as the high-profile cases of Facebook news or Google Photos, "the culprit is often not the algorithm itself but the biased data that was fed into it."

---

Anarchy_Juiblex posted...

Sensationalist tripe.
"We built an image recognition program but only gave it source images of violent images . . . I wonder why it will try to interpret inkblots as . . . "

Having only been exposed to negative images and depressing thinking, the AI's empathy logic simply failed to turn on.
---

Asherlee10 posted...

If the AI has only seen violent images, that would be 100% of its experience in the world. It wouldn't know how to identify anything else.

It seems unsurprising that it would, in turn, see violence in inkblots because it has never seen pleasant imagery.

I'm pretty sure that's what they expected. They wanted to show that learning alogthirms can be taught bias from the input. Best way to show that is to do something extreme like this.
---

Should have used 4chan
---

Asherlee10 posted...

If the AI has only seen violent images, that would be 100% of its experience in the world. It wouldn't know how to identify anything else.

It seems unsurprising that it would, in turn, see violence in inkblots because it has never seen pleasant imagery.

Yeah that is the obvious train of thought but that didn't stop idiots from pretending that algorithms couldn't be biased after that Google Images gorilla snafu. I believe it was something along the lines of "wow so now SJWs think that algorithms are racist"
---

MacadamianNut3 posted...

that Google Images gorilla snafu

Wha?
---

DK9292 posted...

Wha?

Sorry, Google Photos not Images

https://www.theverge.com/2018/1/12/16882408/google-racist-gorillas-photo-recognition-algorithm-ai
---

Tmaster148 posted...

CableZL posted...

why

Their research set out to prove that the method of input used to teach a machine learning an algorithm can greatly influence its later behavior. The scientists argued that when algorithms are accused of being biased or unfair, such as the high-profile cases of Facebook news or Google Photos, "the culprit is often not the algorithm itself but the biased data that was fed into it."

Terrifying
---

Asherlee10 posted...

Do they know why?

We are no longer particularly in the business of writing software to perform specific tasks. We now teach the software how to learn, and in the primary bonding process it molds itself around the task to be performed. The feedback loop never really ends, so a tenth year polysentience can be a priceless jewel or a psychotic wreck, but it is the primary bonding processthe childhood, if you willthat has the most far-reaching repercussions.

---

Newsweek?
---

Verdekal posted...

Newsweek?

What about it?
---

Also I wish I had searched for this earlier but you can actually view the inkblots that were captioned.

http://norman-ai.mit.edu/
---

Do you want Skynet?

Because this is how you get Skynet.
---

Asherlee10 posted...

This is why we need to start establishing rights for strong AI now before it's created.

If we treat strong AI robots with rights as if they are persons, we should all be able to live together well.

FREEDOM! FREEDOM! FREEDOM! FREEDOM!
WE ARE ALIVE! WE ARE ALIVE! WE ARE ALIVE! WE ARE ALIVE!
---

Asherlee10 posted...

Darkman124 posted...

Asherlee10 posted...

Do they know why?

We are no longer particularly in the business of writing software to perform specific tasks. We now teach the software how to learn, and in the primary bonding process it molds itself around the task to be performed. The feedback loop never really ends, so a tenth year polysentience can be a priceless jewel or a psychotic wreck, but it is the primary bonding processthe childhood, if you willthat has the most far-reaching repercussions.

But does that explain why it ignored the coded empathy? I mean, on the surface this kind of says 'sentience'

This just sounds like bollocks. I don't think it's real lol.
---

MIT: Become Human
---

So, if they showed the robot CE posts all day, it would become a brony, SJW, foreveralone?
---