The Fringe | Conspiracy, News, Politics, and Fun Forum!

Full Version: MIT fed an AI data from Reddit, and now it only thinks about murder
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Pages: 1 2
...

This week, researchers at MIT unveiled their latest creation: Norman, a disturbed AI. (Yes, he’s named after the character in Hitchcock’s Psycho.) They write:

Norman is an AI that is trained to perform image captioning, a popular deep learning method of generating a textual description of an image. We trained Norman on image captions from an infamous subreddit (the name is redacted due to its graphic content) that is dedicated to document and observe the disturbing reality of death. Then, we compared Norman’s responses with a standard image captioning neural network (trained on MSCOCO dataset) on Rorschach inkblots; a test that is used to detect underlying thought disorders.

While there’s some debate about whether the Rorschach test is a valid way to measure a person’s psychological state, there’s no denying that Norman’s answers are creepy as hell. See for yourself.

https://www.theverge.com/2018/6/7/174374...thmic-bias
Oh, joy.
Facepalm

Guest

Factors
Bizarro! Scream1
oh man, they are gonna send Norman to military school....

Fartron

(06-07-2018, 07:19 PM)Talon Wrote: [ -> ]...

This week, researchers at MIT unveiled their latest creation: Norman, a disturbed AI. (Yes, he’s named after the character in Hitchcock’s Psycho.) They write:

   Norman is an AI that is trained to perform image captioning, a popular deep learning method of generating a textual description of an image. We trained Norman on image captions from an infamous subreddit (the name is redacted due to its graphic content) that is dedicated to document and observe the disturbing reality of death. Then, we compared Norman’s responses with a standard image captioning neural network (trained on MSCOCO dataset) on Rorschach inkblots; a test that is used to detect underlying thought disorders.

While there’s some debate about whether the Rorschach test is a valid way to measure a person’s psychological state, there’s no denying that Norman’s answers are creepy as hell. See for yourself.

https://www.theverge.com/2018/6/7/174374...thmic-bias

Same thing happens when I look at reddit. Chuckle
Did anybody think that Norman might be a reflection of its creators. Some of these highly smart people working in tech tend to be either on the autism spectrum or borderline sociopathic or sociopathic. These high IQs techies tend to be low on empathy for others and socially awkward and any AI they build will reflect that.

Logic can be cold blooded sometimes and your AI should not be so logical that it is without empathy. I guess what I am saying here is sometimes you have to be a bit illogical to show empathy. For instance your girl friend asks you does this dress make me look fat and you lie and say no. You spared her feelings and saved yourself some grief.
This reminds me of the case where a bunch Twitter trolls for laughs taught an AI on Twitter to be a neo nazi white supremacist. Twitter had to shut down the AI experiment. AI are like children who need to be properly socialized in order to play nicely with others.
Ah, so Mark zuckerbergs real name is Norman.

Now that makes sense.

;)
Pages: 1 2