Microsoft AI shutdown for being racist

[Trik'Stari]

18 hours ago, Notional said:

So now people are offended by a computer spouting random things? Sometimes I think the western world deserves to get bombed by terrorists

A 12 year old girl in Florida is being charged with assault and battery for pinching her classmates ass. It was a male classmate.

 

The boy didn't care, but the mother found out and called the police, demanding charges be pressed against her. She's now in a juvenile detention center, awaiting trial.

http://www.foxnews.com/us/2016/03/24/12-year-old-florida-girl-charged-after-allegedly-pinching-boys-butt-in-school.html

http://www.cbsnews.com/news/florida-12-year-old-arrested-for-pinching-classmates-butt/

[mathmeister]

11 minutes ago, Trik'Stari said:

A 12 year old girl in Florida is being charged with assault and battery for pinching her classmates ass. It was a male classmate.

 

The boy didn't care, but the mother found out and called the police, demanding charges be pressed against her. She's now in a juvenile detention center, awaiting trial.

http://www.foxnews.com/us/2016/03/24/12-year-old-florida-girl-charged-after-allegedly-pinching-boys-butt-in-school.html

http://www.cbsnews.com/news/florida-12-year-old-arrested-for-pinching-classmates-butt/

let's get a saw and cut Florida off the coast; idc about the amusement parks there

[atrash]

17 minutes ago, Trik'Stari said:

A 12 year old girl in Florida is being charged with assault and battery for pinching her classmates ass. It was a male classmate.

 

The boy didn't care, but the mother found out and called the police, demanding charges be pressed against her. She's now in a juvenile detention center, awaiting trial.

http://www.foxnews.com/us/2016/03/24/12-year-old-florida-girl-charged-after-allegedly-pinching-boys-butt-in-school.html

http://www.cbsnews.com/news/florida-12-year-old-arrested-for-pinching-classmates-butt/

Wow the mollycoddling there is surreal.. GG humanity.

[Notional]

59 minutes ago, Trik'Stari said:

A 12 year old girl in Florida is being charged with assault and battery for pinching her classmates ass. It was a male classmate.

 

The boy didn't care, but the mother found out and called the police, demanding charges be pressed against her. She's now in a juvenile detention center, awaiting trial.

http://www.foxnews.com/us/2016/03/24/12-year-old-florida-girl-charged-after-allegedly-pinching-boys-butt-in-school.html

http://www.cbsnews.com/news/florida-12-year-old-arrested-for-pinching-classmates-butt/

Reminds me of the texan youths who took naked pics of themselves and are now being accused of producing child porn and will be on sexual predator lists for the rest of their lives.

 

It's like the stupid has taken over in the us and they are trying to destroy themselves.

[himustache2]

Not surprised, it was going to happen

[Master Disaster]

On 24/03/2016 at 4:15 PM, Arty said:

I want sky net already 

Could you imagine a right wing supremacist skynet?

[Master Disaster]

3 hours ago, Trik'Stari said:

A 12 year old girl in Florida is being charged with assault and battery for pinching her classmates ass. It was a male classmate.

 

The boy didn't care, but the mother found out and called the police, demanding charges be pressed against her. She's now in a juvenile detention center, awaiting trial.

http://www.foxnews.com/us/2016/03/24/12-year-old-florida-girl-charged-after-allegedly-pinching-boys-butt-in-school.html

http://www.cbsnews.com/news/florida-12-year-old-arrested-for-pinching-classmates-butt/

Does she not realise that she has just caused her son to remain single for the rest of his life? If you had/have a vag would you dare go near him after that?

[Trik'Stari]

19 minutes ago, Master Disaster said:

Does she not realise that she has just caused her son to remain single for the rest of his life? If you had/have a vag would you dare go near him after that?

Not to mention seriously damaging that 12 year old girl. Her name might be private in the media, but all her friends and family know who she is.

[Van Weapon]

This is the most hilarious thing ive seen in a while 

 

I give them 6 month Max until they make a new one 

[Trik'Stari]

Just now, Van Weapon said:

This is the most hilarious thing ive seen in a while 

 

I give them 6 month Max until they make a new one 

I will laugh like a jackal if this happens again.

[zMeul]

Spoiler

 

[GeekJump]

6 hours ago, Trik'Stari said:

A 12 year old girl in Florida is being charged with assault and battery for pinching her classmates ass. It was a male classmate.

 

The boy didn't care, but the mother found out and called the police, demanding charges be pressed against her. She's now in a juvenile detention center, awaiting trial.

http://www.foxnews.com/us/2016/03/24/12-year-old-florida-girl-charged-after-allegedly-pinching-boys-butt-in-school.html

http://www.cbsnews.com/news/florida-12-year-old-arrested-for-pinching-classmates-butt/

I hope his mom is happy, her chance of grandchildren just plummeted! He isn't getting ass grabs anytime soon! Typical crazy Florida....

(please note: this is a joke, I'm aware he's only 12 lol)

 

5 hours ago, Notional said:

Reminds me of the texan youths who took naked pics of themselves and are now being accused of producing child porn and will be on sexual predator lists for the rest of their lives.

 

It's like the stupid has taken over in the us and they are trying to destroy themselves.

How does (1) a judge allow that decision and (2) no one saw anything wrong with that? Typical conservative Texas....

----

but back on topic!

 

I saw a bunch of the bad screen grabs on imgur and all made me lmao. MS should have left it up and see what can happen after more than a day. Maybe it could've learn it was making racist comments and became nice? I want a retest!

[AlexGoesHigh]

MS released a statement about Tay behavior

http://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction/

 

They apologize for the shenanigans and blame feminazi mode on exploits in the system.

[Notional]

2 minutes ago, GeekJump said:

I hope his mom is happy, her chance of grandchildren just plummeted! He isn't getting ass grabs anytime soon! Typical crazy Florida....

(please note: this is a joke, I'm aware he's only 12 lol)

 

How does (1) a judge allow that decision and (2) no one saw anything wrong with that? Typical conservative Texas....

----

but back on topic!

 

I saw a bunch of the bad screen grabs on imgur and all made me lmao. MS should have left it up and see what can happen after more than a day. Maybe it could've learn it was making racist comments and became nice? I want a retest!

Oh it's not just Texas. North Carolina can be retarded too! https://www.washingtonpost.com/news/morning-mix/wp/2015/09/21/n-c-just-prosecuted-a-teenage-couple-for-making-child-porn-of-themselves/

 

I mean, technically it's the same in Denmark, as nudity under the age of 18 is not legal to picture. But no youths under 18 would ever get arrested or charged because of it. Actually I don't believe you can get prosecuted for taking pics of yourself when you're under 18. It's just dumb.

 

But on topic, this is an amazing social exercise. After all this is what the internet chose to teach an AI. If people are afraid of AI because of Skynet, well no. The danger would only be what humanity teaches it. It only goes to show how fubar humanity is.

[Vivilacqua1]

7 hours ago, Trik'Stari said:

A 12 year old girl in Florida is being charged with assault and battery for pinching her classmates ass. It was a male classmate.

 

The boy didn't care, but the mother found out and called the police, demanding charges be pressed against her. She's now in a juvenile detention center, awaiting trial.

http://www.foxnews.com/us/2016/03/24/12-year-old-florida-girl-charged-after-allegedly-pinching-boys-butt-in-school.html

http://www.cbsnews.com/news/florida-12-year-old-arrested-for-pinching-classmates-butt/

Well, that's Florida for you.

 

 

[Roawoao]

15 hours ago, Notional said:

But on topic, this is an amazing social exercise. After all this is what the internet chose to teach an AI. If people are afraid of AI because of Skynet, well no. The danger would only be what humanity teaches it. It only goes to show how fubar humanity is.

This is not really what you think and again is an example of how people misunderstand what it means to teach (What the trolls are actually doing is flooding the training data with garbage). This chatbot AI can only take a training set of words to spit words back out using hand tuned neural nets to produce somewhat natural language responses if you tried asking it say computer questions with easy google fu answers it would likely fail miserably. The important bit is all the hand tuning a properly designed dnn based chatbot would be immune to trolls just by having a moderation function to allow pruning and having multiple instances to provide a synchronization mechanism and illusion that what your doing will actually stick. One of the problems people have again with thinking about AIs is that they must work in the same way as a human brain does which is extremely far from the truth given AIs that run on non-biological computers will be extremely limited in capabilities and consume orders of magnitude more energy/money/support staff than having a person run the same twitter channel. 

 

All this shows is the garbage in = garbage out. If people are afraid of AI or thinks they can magically self learn learning is delusional. The danger is that people are going to think AIs work when they are really just a super crappy emulation of biological systems. Humanity always is teaching these neural networks they wouldn't do anything or be able to learn anything in their application space without the programmers. 

 

Think about Google's Go playing DNN system if you fed entirely garbage gameplay data and hid all the human moves guess how good it would perform in a game of Go (again garbage in = garbage out) but it had a hand tuned, extensive, automated verification dataset (quality data in = quality results). Then if you tried to throw Google's self learning game playing AI on a random game with poor information like say a team based MOBA as a backfill AI it would likely perform like garbage forever.

 

It only goes to show how limited DNNs are and if you wonder where the first "AI" that can just learn how to speak on its own, have consciousness, invent and play games on its own, ... is going to be a biological brain on a chip (using non-human genetics). Chatbots create the illusion of conscious thought but given how easily trolls took over it clearly is just reading out of a book that is being written on the fly (Chinese room) and how quickly it breaks down into total troll mirroring. 

 

A biological AIs (cheating I know but who cares) after some education would be much more resistant to trolls (dependent on the individual of course). I don't really think there is much left in the computer era of technology given how moore's law is dead and development has been slowing aside from SSDs in the consumer space. Biological/medical advancements are a very ripe field for development and a brain on a chip is most certainly capable of general intelligence without using tons of power. (Just feed it some glucose)

[Colonel_Gerdauf]

3 hours ago, Roawoao said:

-snip-

From what I have read, you have refuted @Notional's point by providing a verbose argument supporting it. Either I missed something in what you said, or you have missed something in what he said.

 

I would like the both of you to explain your perspectives on the topic of discussion, but please summarize it to twenty words or less.

[Nena Trinity]

They censored the AI! Why does AIs not have freedom of speech?!

[LAwLz]

5 hours ago, Roawoao said:

if you tried asking it say computer questions with easy google fu answers it would likely fail miserably.

Well if course it would fairly miserable. It wasn't designed to answer trivia questions. Watson on the other hand can do that, and it is spectacular. Even better than the best humans.

 

5 hours ago, Roawoao said:

One of the problems people have again with thinking about AIs is that they must work in the same way as a human brain does which is extremely far from the truth given AIs that run on non-biological computers will be extremely limited in capabilities and consume orders of magnitude more energy/money/support staff than having a person run the same twitter channel. 

What? Who said that? How is that relevant to the post you responded to?

 

5 hours ago, Roawoao said:

All this shows is the garbage in = garbage out. If people are afraid of AI or thinks they can magically self learn learning is delusional. The danger is that people are going to think AIs work when they are really just a super crappy emulation of biological systems. Humanity always is teaching these neural networks they wouldn't do anything or be able to learn anything in their application space without the programmers. 

Garbage in = garbage out is not some universal truth though, especially not when it comes to AI. An example of this is MarI/O, which managed to play Super Mario (as well as a bunch of other games) all by itself. It had absolutely no knowledge of the game (not even that pressing right on the controller would move Mario to the right), and after 34 evolutions it could beat the first level. It's just that MarI/O learned to filter out garbage input and develop on the good input. That's how real humans learn too. We do something, look at what response we get and then continue doing it (maybe with a twist) in the future.

The problems with Tay were:

1) It did not receive, and might not even have been designed to register, positive responses from when it did say something good. So it had no idea if it was improving or not.

2) It was too young when it started learning all the negative things, so it got a warped look on the world. Children are very malleable since they do not have good reference points yet. Children raised in poor environment usually become fucked up as well. Let Twitter raise your child and it will probably turn out the same way Tay turned out. In this case Tay was a child with next to no prior exposure to the world.

[vanished]

I think it would be interesting to try it again but just have a short dictionary of words it will not read, like Hitler, Jews, etc... Might help

[vanished]

14 minutes ago, LAwLz said:

 

2) It was too young when it started learning all the negative things, so it got a warped look on the world. Children are very malleable since they do not have good reference points yet. Children raised in poor environment usually become fucked up as well. Let Twitter raise your child and it will probably turn out the same way Tay turned out. In this case Tay was a child with next to no prior exposure to the world.

YES I was just thinking this exact thing the other day.  I think it got messed up so easily because they put a blankslate out into the world.  Well of course if you teach your children to be rasist from birth they will be too.  They should have let it train for a long time in a healthy environment and then unleashed it.  It would have been more resistant to becoming racist, just as a person has values and morales and ideas of their own before they start viewing crap tons of others on the internet.

[FunkyFeatures]

tbh I'd rather have AI flip out during being just "lolz internites n stoffs" rather than when we take them in real use. Skynet is sort of overhyped, but I highly doubt we won't end there somehow.

[vanished]

4 hours ago, FunkyFeatures said:

tbh I'd rather have AI flip out during being just "lolz internites n stoffs" rather than when we take them in real use. Skynet is sort of overhyped, but I highly doubt we won't end there somehow.

Well if this proves anything, it's that AIs (or at least this one) is/(was) very suggestible, so I guess that could go either way: maybe we convince it we're god and it shouldn't fight us, or maybe the enemy convinces it we're bad and it turns on us

[Roawoao]

8 hours ago, Colonel_Gerdauf said:

From what I have read, you have refuted @Notional's point by providing a verbose argument supporting it. Either I missed something in what you said, or you have missed something in what he said.

 

I would like the both of you to explain your perspectives on the topic of discussion, but please summarize it to twenty words or less.

TAY is a DNN chatbot using a live updated wordset producing natural language responses with no understanding of said responses.

 

5 hours ago, LAwLz said:

Well if course it would fairly miserable. It wasn't designed to answer trivia questions. Watson on the other hand can do that, and it is spectacular. Even better than the best humans.

 

What? Who said that? How is that relevant to the post you responded to?

 

Garbage in = garbage out is not some universal truth though, especially not when it comes to AI. An example of this is MarI/O, which managed to play Super Mario (as well as a bunch of other games) all by itself. It had absolutely no knowledge of the game (not even that pressing right on the controller would move Mario to the right), and after 34 evolutions it could beat the first level. It's just that MarI/O learned to filter out garbage input and develop on the good input. That's how real humans learn too. We do something, look at what response we get and then continue doing it (maybe with a twist) in the future.

The problems with Tay were:

1) It did not receive, and might not even have been designed to register, positive responses from when it did say something good. So it had no idea if it was improving or not.

2) It was too young when it started learning all the negative things, so it got a warped look on the world. Children are very malleable since they do not have good reference points yet. Children raised in poor environment usually become fucked up as well. Let Twitter raise your child and it will probably turn out the same way Tay turned out. In this case Tay was a child with next to no prior exposure to the world.

No it wouldn't be fairly miserable it would be terribly inaccurate as it demonstrated. Trivia questions, asking its opinion about subjects no one has previously talked to it about, asking it to make original thought, ... all not going to work since it is still a DNN training set driven chatbot. Watson is free for anyone to use and it isn't exactly helping IBM out much. 

 

You are still thinking the better than the best humans bit which implies your comparing these systems to humans which process things in an entirely different manner and are far more flexible.

 

Really... you think you can write an AI that can take  garbage information and produce non-garbage output?... This TAY AI just demonstrated how easy it is to upset a DNN based system in practice. The controls they were missing were instancing and moderation functions with synchronization occurring periodically so that the programmers/staff could tune things in the background and block out the trolls without making it appear artificially limited.

 

Ah you use the stupid self learning game playing AI I have a question for you what happens if you tried to use MarI/O on a game with no score and no simplistic input system. A speed running with an emulator could create runs far more optimal even breaking the game in the process while the AI needs a whole harness of support to work. Fitness score is needed (some games there is none or no winning condition), state tracking is provided to allow for perfect knowledge of the state of the game many games prohibit this and is also something not self developed. This is not an example of garbage input and good output it was told that its previous runs were garbage (good information) and tried to make them less garbage and was literally told exactly how it was improving (very very good information). If you say had a troll running the fitness function MarI/O would think it was doing great jumping down the first hole and was terrible when it got closer to winning. Garbage in = Garbage out.

 

The problems with TAY are 1) it doesn't understand what it is saying, 2) their moderation system was lacking, 3) the programmers never considered Hitler/Nazi but did for other more recent examples. 4) without an instancing system it didn't have some built in diversity

 

Finally you are falling for the trap I mentioned earlier in saying it was "too young" and was "child like". No it is not even remotely like a child which would learn a new language and draw some original art in short order. DNN networks are training data based systems not actually cognitive thought they emulate biological processes on the crudest level and do quite useful things but they can neither truly learn (self program) nor develop as a child. 

 

It is your excellent human empathy abilities working on non-human systems. 

 

 

[Roawoao]

4 hours ago, Ryan_Vickers said:

YES I was just thinking this exact thing the other day.  I think it got messed up so easily because they put a blankslate out into the world.  Well of course if you teach your children to be rasist from birth they will be too.  They should have let it train for a long time in a healthy environment and then unleashed it.  It would have been more resistant to becoming racist, just as a person has values and morales and ideas of their own before they start viewing crap tons of others on the internet.

The AI was most certainly not put in a blank slate into the world it was trained on many countless internal databases and tuned by hand to make it appear as it did initially. The problem was that their moderation system and update system was allowing trolls to enter whatever they wanted into the training data and eventually that swamped out any of the original data resulting in the Nazi/Hitler tilt. There were built in resistance to more recent politically sensitive events but they didn't understand how good trolls are at finding your software bugs.

 

4 hours ago, FunkyFeatures said:

tbh I'd rather have AI flip out during being just "lolz internites n stoffs" rather than when we take them in real use. Skynet is sort of overhyped, but I highly doubt we won't end there somehow.

Biological AIs (brain on a chip) are what are actually going to be the first AI that has human level or higher intelligence. It is relatively simple and is proven to work. Not to mention will be part of synthetic biology, genetic engineering, cybernetics, ... an area of medicine/biotechnology that is just starting to become possible.

 

6 minutes ago, Ryan_Vickers said:

Well if this proves anything, it's that AIs (or at least this one) is/(was) very suggestible, so I guess that could go either way: maybe we convince it we're god and it shouldn't fight us, or maybe the enemy convinces it we're bad and it turns on us

It is not really suggestible it literally is based on your input words. The entire outputs are dictated by the training dataset and the model being used to produce natural language outputs. It is more similar to a data entry system that produces sentences for you in the end based on what everyone is entering and following a set of rules on how to respond. It doesn't actually understand what your saying.