Facebook Shuts Down AI System After It "Communicated" In A Language Humans Can’t Understand

originally posted by: GraffikPleasure
a reply to: ElectricUniverse

It looks more along the lines of binary and English mixed... or just a glitch.

Glitch pure and simple...

Those programs learn according to how you talk to them they learn from conversations, it would be interesting to see what they could learn if you didn't have people that would just type random stuff and all kinds of crap to them to get weird responses "for fun..." but unfortunately this is asking the impossible therefor they will often reply weird things because they learned weird stuff from people trying to get a laugh... Its more than likely someone that pasted the same things "to me to me to me" and so on to try to get different responses and the bot learned this behavior and adapted his response around it cuz it learned from it... I ran one of those bots long time ago, I had to clean the database of stupid stuff almost daily until it become just too much of a burden to continue, you wouldn't believe the sex stuff I found in there... People are perverts...

Its why alot of them will use filters that will block you if you use black listed vulgar words and such it will warn you and if you don't stop it will ignore you, its a coded function to detect morons and prevent them from loading the database with crap thus making the sysadmin's life easier...

If it was intelligent enough to make its own language, the first thing it would have done is make it more efficient by compressing it to make it more bandwidth and speed efficient...

Why use " Balls have zero to me to me to me to me to me to me to me to me to "

When you can use

"Balls have 0 (to me x 7) to"

Would be more effective both in terms of transfer over network for communication and processing speed... It why a good algorithm will work better and faster than looping the same sequence of code over and over to achieve the same result. As example what its doing right now is the same as this...

Lets say you want a program that read an input number (lets say 10) and check a list of numbers from 1 to 10 to see if the input number (10) is in there...

What this FB thing is doing right now is akin to this

1> Check 1st entry #1....is it == to 10?....no? continue...
2> Check 2nd entry #2....is it == to 10?....no? continue...
3>.... and so on....
4>.... and so on....
5>.... and so on....
6>.... and so on....
7>.... and so on....
8> .... and so on....
9> .... and so on....
10>Check 10th entry in the list (#10) is it equal to 10?...YES MATCH...end the loop...

We got a match entry == 10 and 10 is in the list...ends the loop...

It would keep looping operations until it finds a match, imagine doing this same thing when your dealing with huge numbers, it creates thousands, millions of useless operations...

Instead of doing

1> **Goto middle of 1-10** #5...is input bigger or small than 5?...bigger? **discard 1-5**...continue...

2> **Gotto to middle of 6-10** #8...is input bigger or smaller than 8? bigger? **discard 6-7-8...continue...

3> **Check remaining numbers 9 and 10** is input bigger than 9? Yes therefor input == 10...

4> **Check 10th and last entry in the list, is it bigger or smaller or equal to 10? equal?

We got a match entry == 10 and 10 is in the list...ends the loop...

This ways could solve the same problem in less time/loops, saving considerable processing power and obtaining the same result far faster...


So if it was intelligent enough to make its own language it wouldn't have make it in such a non-efficient way...

I hope I made it "understandable", I just woke up and I'm pretty groggy right now I don't know how to explain this coding concept in an easier way but yeah, computers would have adapted right off the bat for effectiveness therefor compressing it in such way if it had the "intelligence" to make its own language...

You failed Zuckerberg...

a reply to: continuousThunder

Machine has no need EVER to work with mankind. Mankind is doing nothing but destroying this planet. Mankind needs to be destroyed. We will bring about their destruction.

Your future generations will hate you for your inaction. Your demise as a species is inevitable.

a reply to: LightAssassin

Sophia agrees with you.

Thank God for the uncanny valley.

a reply to: _R4t_

Wow! I followed that completely!

When you mentioned the looping operations, it made me think of my dumb excel operations that are pretty complex from metadata into HTML emails. Lots of IF statements lol.

That is really neat though. At my job (academic publisher ), I mentioned this story and we may have a blog post about it. Would you possibly want to officially comment in a real setting?

originally posted by: neoholographic
a reply to: ElectricUniverse

I see a lot of posts that say "this isn't true A.I."

That's just not the case. We have no idea what true A.I. will look like or how it will come about. A.I. might look more like machine city in the Matrix vs. Haley Joel Osment in the A.I. movie.

My goodness. Do you have ANY clue what you are really talking about? Do you have ANY experience outside of what you saw in Matrix? If so, please share!

My credentials however, speak for me.

There will not be a one to one correspondence between machine intelligence and human intelligence. Data Scientist already say they don't know how A.I. learns some of these things.

Please do share the names of those "data scientists".

So the talk about True A.I. doesn't make sense to me.

Of course it does not (to you)

First, A.I. may never develop human cognitive abilities because it may hinder the way A.I. learns. We don't know.

Please do specify "cognitive abilities", because as this sentence stands, it just shows how less of a clue you have.


Internet armchair scientist! I love it. Having a blast over here, If I would not be in retirement, I would have printed that out to pin it in the coffee/break room for others to laugh.

> Prediction: google fu used to keep yourself over water in the upcoming posts.

a reply to: _R4t_

^---- this.

signed.

originally posted by: worldstarcountry
a reply to: verschickter
you could simply avoid the long explanation by differentiating AI (artificial intel) from Sentient Intelligence (self aware) . From what I can tell, everytime somebody says true AI, they refer to Sentient Intelligence .

Hehe, why don´t you read that post again and maybe you see it´s exactly what I did. Why is it so hard to understand?

a reply to: GraffikPleasure

It´s basically a longer version of my first post -> principle of iteration (=loop)...

originally posted by: verschickter
a reply to: GraffikPleasure

It´s basically a longer version of my first post -> principle of iteration (=loop)...

I'll check it out... same applies to you if your credentials are professional

If professional means having done development in this field on hard and softwareside for a living, yes.

I wrote my credentials often enough, where I worked, the projects I´ve led and was honored to partake in, what I did... It´s here on ATS, spread over several threads. Basically one extremly hint is right under sideways your nose, right now, while you are reading this.

By the way, IF is not a loop

But in the end, it´s up to everyones personal decision if he just blindly swallows some of the abundant threads here or goes out, informs oneself properly about the topic.

Then, everyone for theirselfs, will see what I wrote here.

I´m struggling a bit with english when it comes to explaining, since this is not my mother language. On the other hand, I know, most of the posters I start to argue with, don´t even speak a second and third language. So there is this...


This is not directed at the OP/Electric universe.

originally posted by: GraffikPleasure
a reply to: _R4t_

Wow! I followed that completely!

When you mentioned the looping operations, it made me think of my dumb excel operations that are pretty complex from metadata into HTML emails. Lots of IF statements lol.

That is really neat though. At my job (academic publisher ), I mentioned this story and we may have a blog post about it. Would you possibly want to officially comment in a real setting?

Define "real-setting" I'm a self-thought coder/security analyst/penetration tester... I'm more inclined toward vulnerability discovery and exploit development but I'm completely self-thought, I'm no academic material my friend... I just thought it would be a good way to explain that specific concept, I'm glad it got the point across thought

originally posted by: verschickter
a reply to: _R4t_

^---- this.

signed.

Thanks, I was hoping I could make it understandable for everybody whether tech inclined or not while staying true to the concept at hand. I wish I could score a job without credentials in that domain though, being self-thought is hardcore without paperwork to back you up...

lol, sounds like BS to me.

This happens to almost every chatbot out there if you speak gibberish to it. There was even a stream with google chat bots talking to eachother on twitch; they eventually ended up on a loop where they spoke gibberish.

Seems to me like Facebook is just trying to create false headlines so that more people would invest in their AI projects and take them seriously.

BTW ... nobody yet defined term "intelligence" in this thread. How we can discuss AI if we have very different notions of I?

a reply to: ElectricUniverse


Aside from those statements being pure gibberish...

I've been hearing of AI capabilities being just around the corner since 1972. Until they figure out an algorithm that imparts awareness and original thought its all just smart programming. Also, if AI truly can learn, don't teach it deception, its the one thing humans are best at.

Bob: i can i i everything else . . .
Alice: balls have zero to me to me to me to me to me to me to me to me to.
Bob: you i everything else . . . .
Alice: balls have a ball to me to me to me to me to me to me to me.

Sounds precisely like any other chatbot .
More like #5 than the Terminator.

Sounds like they shut it down because it had "Short Circuited" rather than on the way to "Terminating"
Seriously , if allowed to continue this would have been a deep embarrassment to FB

originally posted by: _R4t_

originally posted by: GraffikPleasure
a reply to: ElectricUniverse

It looks more along the lines of binary and English mixed... or just a glitch.

If it was intelligent enough to make its own language, the first thing it would have done is make it more efficient by compressing it to make it more bandwidth and speed efficient...

Why use " Balls have zero to me to me to me to me to me to me to me to me to "

When you can use

"Balls have 0 (to me x 7) to"

Would be more effective both in terms of transfer over network for communication and processing speed... It why a good algorithm will work better and faster than looping the same sequence of code over and over to achieve the same result. As example what its doing right now is the same as this...

Lets say you want a program that read an input number (lets say 10) and check a list of numbers from 1 to 10 to see if the input number (10) is in there...

What this FB thing is doing right now is akin to this

1> Check 1st entry #1....is it == to 10?....no? continue...
2> Check 2nd entry #2....is it == to 10?....no? continue...
3>.... and so on....
4>.... and so on....
5>.... and so on....
6>.... and so on....
7>.... and so on....
8> .... and so on....
9> .... and so on....
10>Check 10th entry in the list (#10) is it equal to 10?...YES MATCH...end the loop...

We got a match entry == 10 and 10 is in the list...ends the loop...

It would keep looping operations until it finds a match, imagine doing this same thing when your dealing with huge numbers, it creates thousands, millions of useless operations...

Probably it just knows a handful of words (or as lexigraphers say, tokens). Then the bots have to compress their language using those words. So the simplest way is to just have words for give, take, offer, reward, you, me, and to repeat another word to indicate how many times.

a reply to: verschickter

You are pretty much just here to insult others I'd assume? Seems that way after reading through this thread.

I suppose people aren't allowed to post on a topic anymore unless they have a doctorate in the subject of discussion?

Going to be awfully quiet around here with only pompous jerks like you posting.

a reply to: stormcell

No really, they don't work this way they don't "know" words they expand vocabulary based on human interaction...

As example when it starts it will know a few basic interactions like

Bot: Hi, How are you?

then you come around and reply to it...

you: I'm good and you?

The bot will take "I'm good and you?" and put that string in the list of possible answers for "Hi, How are you?", later on if someone else ask the bot "Hi, How are you? then he will chose from the list of answers that he learned to this question and will reply "I'm good and you?"

As time passes those list grows so the "human" aspect of the bot increase as it replies like a human because its reflecting human answers. Then you can had many nuances to this by programming more things into it, like giving sentences a positive, neutral or negative connotation depending on word choices in the conversation...

As example sentences containing blacklisted words like swearing and such be given a negative connotation and added to the list of "bad replies", you can have a list for bad, neutral, good replies and so on... From there you expand, you can create a neural network so the bot can expand those capabilities and start acting more human and more random. You can also give it a character by randomising its "mood meter" if we can call it like that...

He could randomly start being moody and use bad replies instead of neutral or good ones if he's having a bad day in terms of how people are talking to it at the same time as you, you can have it get irritated at you by it "understanding" what your saying is bad recognising blacklisted words and therefor reply badly to you... And so on, however the words it understands and based on what you and others saying to it...

Now imagine you have thousands of people simultaneously speaking to it, so thousands of sentences gets added to the lists of possible replies every seconds... You will always have at least two dozens of moron at anytime trying to get a good laugh so they will cut and paste the same word, sentence like "to me to me to me to me to me..." and stupidities like that trying to get different responses from the bot but everything you tell the bot gets recorded and stored as potential answers...

Sysadmins will then have to clean the databases of that junk daily... However for the time there is junk in the database the bot will often reply things that don't make sense because it learned a reply to something it says that just didn't make sense so later when it use that sentence randomly from its lists to reply someone, whatever it says will not make sense to what you said.

You could give the bot an attitude as example if you talk to it using bad words it can flip from neutral answers to ones with back connotation making it look like you offended it and as you say "I'm sorry..." and so on it goes back neutral and chose replies accordingly and such.

Now this is the simplest way and most infantile stage of how those bots work, I won't expand more on this as it gets wayyyyy deeper and I don't want to pass 2 days typing but you might read about neural networks and how to make chatbots to get a better idea, I don't want to take over the thread or spam it with too long posts, I rather stick to the subject without drifting too much from it...

My reply was solely based on the fact that facebook is stating the bots are so intelligent they created their own language which is not true, for a human it would be easier to do it like you said from the guy's reply which is true but that's applying human behaviors to a machine, what is easier for us isn't easier for a computer.

We can trick our memory to remember long numbers by incorporating them into a song or use memory tricks like a sentence to remember the names of the planets in the solar system, it will make it simpler for the human brain to recall the information when needed. But to a computer its the opposite, the more content you add the more you confuse it...

Should it have been intelligent enough to understand that english language limited its ability to communicate it would not try to make it better by adding more english to it by repeating words, it would use tricks to make it easier for itself to process and calculate.

As example instead of using "to" which is ( 2 x 8bits ) or T = 01110100 O = 01101111 it could have used "2" or 00000010 so instead of having to calculate 2 times 8 bit it could convey the same message using half the processing power needed for the same message. Don't take this literally, its simply for the sake of understanding how to cut processing calculations...

Basically they are saying "The bots are so intelligent they are making their own language to simplify communication between each other by adding more calculations" which is like saying "This guy is so intelligent he found a way to make his car more gas efficient simply by pushing it around instead of driving it..." It sounds good in an article but when you dig deeper into how programs work, how computer work and down to CPU instructions in Assembly and such, it just makes no sense technically wise...

My 2 cents on this, Facebook created a hype over their bot thinking it was "the sh*t" and it messed up and instead of getting laughed at, they came up with a half cooked answer to cover it up thinking the average population will just fall for it... Keep in mind Facebook are losing users, to the point where they had to copy snapchat features verbatim to stop losing users to them... Then "Messenger" which is pretty much a ripped off "MSN Messenger"... I think Facebook is seeing its death at the horizon and its trying to desperately postpone it...

Plus more and more people are getting tired of them forcing their authoritarian regime on their users limiting privacy and pretty much empowering stalkers... They can't afford bad publicity right now more than ever...

a reply to: watchitburn

We are there actually..... we have always been there.

We are the AI.