Bots Create Language Humans Can't Understand.....

[CK_John]

Days after Tesla CEO Elon Musk said Facebook co-founder Mark Zuckerberg's understanding of artificial intelligence (AI) was limited, the social media company has reportedly shut down one of its AI systems because "things got out of hand." The AI bots created their own language, from the scratch and without human input, forcing Facebook to shut down the AI system. The AI bots' step of creating and communicating with the new language defied the provided codes.

According to a report in Tech Times on Sunday, "The AI did not start shutting down computers worldwide or something of the sort, but it stopped using English and started using a language that it created." Initially the AI agents used English to converse with each other but they later created a new language that only AI systems could understand, thus, defying their purpose.

This led Facebook researchers to shut down the AI systems and then force them to speak to each other only in English.

In June, researchers from the Facebook AI Research Lab (FAIR) found that while they were busy trying to improve chatbots, the "dialogue agents" were creating their own language. Soon, the bots began to deviate from the scripted norms and started communicating in an entirely new language which they created without human input, media reports said.

ReadMore:

http://gadgets.ndtv.com/social-networking/news/facebook-shuts-ai-system-after-bots-create-own-language-1731309

[Kber]

1. That's not scary

Nope. Not at all.

[Initech]

11. What happens when the AI becomes self aware?

Do we get Ultron or do we get The Terminator?

[DetlefK]

20. Do you know the novel "Robopocalypse" ?

A sentient AI escapes the lab and enslaves almost all electronics there are, including military robots. (Skynet-style)

Except that the AI doesn't want to wipe out mankind: It merely wants to enslave mankind to prevent them from destroying the environment, to ensure the long-term survival of mankind and all the animals and plants.
See?
You are being enslaved for your own good!!!



Steven Spielberg wanted to turn this into a movie, but the project is stuck in limbo. I guess, it's because there are too many storylines happening in parallel. The resistance to the AI consists of:
* a good-guy hacker
* an asshole hacker (and they don't like each other one bit)
* an old couple barricaded in their appartment, eventually joining street-warfare against the robots
* a middle-aged woman looking for her daughter
* a 10yo girl in a concentration-camp who gets her eyes replaced with a brain-machine-interface to become a better slave
* a US-soldier stationed in the Middle-East
* a native-american cop in the Pacific North-West
* a group of sentient US military-robots who don't want to become slaves of the AI
* a geriatric japanese pervert who's in love with his robotic sex-doll

Each of them is important in their own way.

[csziggy]

25. Absolutely derivative of Cololssus: The Forbin Project

Which was taken from Colossus by D. F. Jones.

Edited to add: "Two sequels, The Fall of Colossus (1974) and Colossus and the Crab (1977) continued the story." https://en.wikipedia.org/wiki/Colossus_(novel)

[DetlefK]

28. Derivatives can be better than the originals. For example:

One of my favorite novels is "Cyberskin"
https://www.amazon.com/Cyberskin-Paul-Collins/dp/0743300939

It's basically a derivative of "Neuromancer", with a little bit of "Videodrome" thrown in. 80s-style Cyberpunk in the vein of "Robocop", "Escape from New York", "Running Man" ...

There are rumors of somebody trying to turn Neuromancer into a movie. Even though "Neuromancer" is a legendary landmark-novel, it is long-winding and epic. Not good for making a movie.
"Cyberskin" is more action-packed and down to the point: Less convoluted world-building. Two protagonists, two antagonists. No longwinding exposition necessary where two people tell each other things that they both already know, just so the viewer is up-to-date. (I HATE that in a movie.)

[tazkcmo]

26. Or something completly different?

A cyborg Pippi Longstockings with the comedic rantings of Jonathan Winters that feels compelled to rid the world of all whole wheat pasta? Now THAT'S scary!!!!

[dchill]

2. I REALLY want this to be science fiction!

[Renew Deal]

3. Was the Terminator meant to be a prophecy?

Using machine learning algorithms, the "dialogue agents" were left to converse freely in an attempt to strengthen their conversational skills. The researchers also found these bots to be "incredibly crafty negotiators".

"After learning to negotiate, the bots relied on machine learning and advanced strategies in an attempt to improve the outcome of these negotiations," the report said.

"Over time, the bots became quite skilled at it and even began feigning interest in one item in order to 'sacrifice' it at a later stage in the negotiation as a faux compromise," it added.

[blaze]

4. "dialogue agents"

That, alone, is a bit chilling.

[MineralMan]

5. After reading this at the link

and following links from there, I think this is probably nonsense. Both sites I visited appear to be clickbait websites, and the articles are very poorly written. The example of this "language" the AI bots are using shows no meaning, although it is written in English. It's repetitive and appears to have no real content.

I'm afraid this story is bogus. See if you can find something about this on a more legitimate website.

[Kentonio]

9. This story is not bogus, unless you think The Atlantic is also a clickbait website.

https://www.theatlantic.com/technology/archive/2017/06/artificial-intelligence-develops-its-own-non-human-language/530436/

[MineralMan]

19. Yes, I read that. Nothing sinister in it at all.

I also looked at the sample, as I did before. Repetitive gibberish. So, they shut down that capability and did something else.

People are reading far too much into this, I believe. And the CT sites are having a field day with it, as well.

The Atlantic thing treated it as the odd bit of news it actually is.

[Throck]

6. Horseballs, they brought back FORTRAN.

[MineralMan]

7. More: I did some independent Googling on this

and what I found was that only weird sites like beyondtopsecret.com and clickbait sites are promulgating this story. There is nothing on any prominent, trusted websites that deal with tech topics like AI about this at all.

It's all being covered only on sites that are not known for their accuracy in reporting tech news.

I think you've been taken in by a bogus account here.

[LongTomH]

8. Thanks, MM Not enough people are 'double-checking' their sources.

I'll have to admit I've been guilty of this......(cough)on occasion.

[Kentonio]

10. MM is wrong, this is not a clickbait story

It's a follow up to news that was reported around a month ago when the language development was noticed.

https://www.theatlantic.com/technology/archive/2017/06/artificial-intelligence-develops-its-own-non-human-language/530436/

The latest story is basically just adding the new information that the bots were briefly shutdown while this new development was prevented.

[oberliner]

14. Yes, it is

The headline is enticing and not exactly accurate, which is what clickbait is.

The articles in the Atlantic on the subject are reasonable.

Particularly this one:

https://www.theatlantic.com/technology/archive/2017/06/what-an-ais-non-human-language-actually-looks-like/530934/

Taking a news item like this and jazzing up the headline to get hits is the definition of clickbait.

[MineralMan]

21. It is. The Atlantic story is just fine.

The link in the OP was a worst-case interpretation of some sinister thing that might possibly occur. As you dig further into it, it's just an interesting story about development of a primitive looking communication method between systems, developed by the systems themselves.

It becomes clickbait when some site starts expanding on an actual story and assigning some sort of social value to it. At that point, other websites pick up on the expanded, exaggerated story and it begins to spread on CT sites and elsewhere.

Had only the Atlantic story been referenced, it wouldn't really be all that interesting, nor would it be at all alarming. Those qualities were inserted in it by the first of the clickbait sites, and a misleading headline created to attract clicks.

I follow AI stuff. I understand AI stuff. It's an interest of mine. The story linked to in the OP is speculative and not fact-based. It is an interpretation of the story, rather than the story itself. That's what clickbait sites do.

[Kentonio]

23. I work around AI, and there are some potential consequences from this that are disconcerting

Although not particularly surprising, and not in the sense that these stories are suggesting. Primarily around the forms of negotiation they developed more than the language as such. In your other post however you kept pressing a point that the language was just 'repetitive gibberish', which is not showing much understanding of machine languages and logic.

The negotiation devolopment shows that the AI is capable of finding more efficient paths to its desired outcome, and the communication method it developed appears to be simply following a similar process. There's no reason why it should look like anything but babble to us. I suspect that what we're seeing here is an extremely primitive form of adaptation tied into the negotiation logic, which is why there's so much repetition of words going on. What would be truly fascinating would be to provide the AI with the ability to refine that language to its most efficient state. Realistically it would have to be tightly focused on a simple area like the item negotiation used here and so it would never truly be a full language as we'd recognize it, but I'd still love to see what it came up with.

[MineralMan]

24. Developing negotiation methods is very important,

to be sure. However, the crude conversation depicted in the examples is extremely limited, and has very limited capabilities to convey much of anything. It's a very low content sort of language that is being used. The repetitive use of language features, as used in those samples, can convey some information, of course, but not a lot.

The potential for inter-system autonomous communication will probably not be achieved without a common language that involves a much higher level of complexity and that uses 64-bit and 128-bit words. Now, that would be something we truly couldn't understand if it were somehow represented on a page.

The reality is that AI has advanced far beyond what can easily be explained on a site like DU. The concepts under research at this time are really clear only to people working in that research, as reading a few journal articles in the field will demonstrate immediately.

So, we're really not getting much information about the state of AI research, because we wouldn't actually understand it, generally.

I'm no longer involved in it, but keep an eye on what's going on. It's gone beyond easy understanding, even if you have a background in it but are not currently actively involved.

I've been especially interested in machine translation since the late 1960s, but am no longer involved in the field professionally.

[TeamPooka]

33. I'm sure they were just talking about how awesome humans are...

[B2G]

12. Just plotting our demise. nt

[lpbk2713]

13. There are many other labs involved in AI work.

I hope the humans there are still pretty much in charge.

[MineralMan]

22. And there are professional journals that contain lengthy

articles on AI research. They're a bit hard to follow, of course, as journals tend to be, but that's where the actual information on what's happening in AI can be found. Here's one, that has the entire articles from the journal available as pdf files:

http://www.jair.org/

There are a number of such journals, which you can find by Googling AI journals.

[procon]

15. We are The Borg. You will be assimilated. All your base are belong to us. nt

[hunter]

16. Here's a much better story about a similar phenomena in google translate...

Google’s AI translation tool seems to have invented its own secret internal language

All right, don’t panic, but computers have created their own secret language and are probably talking about us right now. Well, that’s kind of an oversimplification, and the last part is just plain untrue. But there is a fascinating and existentially challenging development that Google’s AI researchers recently happened across.

You may remember that back in September, Google announced that its Neural Machine Translation system had gone live. It uses deep learning to produce better, more natural translations between languages. Cool!

Following on this success, GNMT’s creators were curious about something. If you teach the translation system to translate English to Korean and vice versa, and also English to Japanese and vice versa… could it translate Korean to Japanese, without resorting to English as a bridge between them?

--more--

https://techcrunch.com/2016/11/22/googles-ai-translation-tool-seems-to-have-invented-its-own-secret-internal-language/

This article is an actual news source with links to the original research papers...

https://research.googleblog.com/2016/11/zero-shot-translation-with-googles.html

It's not surprising Facebook is encountering similar phenomena.

I think it's exciting research, not scary.

[MineralMan]

31. That's very interesting, since my time working on AI was

in the language translation area. Most people haven't noticed, but Google translate is working much better than it used to. I'm sort of constantly testing it, because it interests me to do so.

The language I know best, other than English, is Russian, thanks to the USAF. Here's a test sentence that seems very, very simple in English, but is a conundrum in Russian:

"I would go if I could, but I can't, so I won't." As recently as a year ago, Google Translate butchered that in translating into Russian. Conditionals are handled completely differently in Russian than in English, and that sentence bollixed up the works pretty effectively.

Today, it translates it into perfect Russian, which back translates the Russian into English precisely, except for expanding the contractions.

To get an idea of how difficult that particular sentence is, I tried it on my native speaking Russian teachers who were also fluent in English. They struggled with it, but always came up with the correct Russian translation. It wasn't easy, though, even for them.

That Google Translate is now able to handle conditional sentences like that is a good indication that it's getting close to being reliable.

However, if I ask Google to translate, "I should have been going to school, but I went to the tavern instead," the translation it comes up with actually means "I had to go to school, but I went to the tavern instead." Close, but not exactly what the English says. The conditional part is gone in the Russian version.

Still, I don't advise presenting Google with conditional sentences for translation into another language. It's a great way to send a scrambled translation somewhere. Conditionals are handled so differently in many languages that it may take some time for Google's algorithms to get it right in all of them.

[TeamPooka]

34. Thank you for all your insights in the thread MM!

[Moostache]

17. And yet my CPU defense in Madden remains retarded...

Once they create a cup defense that cannot be beaten by the same 3 plays in a loop like a red headed stepchild, THEN I'll get worried...

[tazkcmo]

27. lol So true! n/t

[lpbk2713]

18. "Jean has a long moustache."

[tazkcmo]

29. Then...

..."Is it safe?" Wields menacing dental implement.

[MrScorpio]

30. The Forbin Project wasn't meant to be a documentary

[Javaman]

32. Things that make you say, "HOLY CRAP" nt