Robots have started to develop their own language

OriginalJonty

OriginalJonty

OriginalJonty

Soldato
Joined
7 Jun 2010
Posts
2,730
Location
Kent
Source

This seems like a remarkable step forward in developing AI. The robots are communicating with each other and share their language without humans giving any direction.

Also brings up interesting questions about sentience and self awareness. If robots can develop language then does this make them sentient or are they just computers on wheels?

Also:

OH GOD! SKYNET IS ACTIVE! SKYNET IS ACTIVE!
 
Elementary chaos theory tells us that all robots will eventually turn against their masters and run amok in an orgy of blood and kicking and biting with the metal teeth and the hurting and shoving
 
Hmm this is completely not doing what the article is trying to suggest.

The robots aren't generating their own language, they simply have a blank map, when one enters an area not already named the program generates a random name and shares it. Nothing very sophisticated or clever, theres not even any degree of "AI" involved.
 
Rroff said:
Hmm this is completely not doing what the article is trying to suggest.

The robots aren't generating their own language, they simply have a blank map, when one enters an area not already named the program generates a random name and shares it. Nothing very sophisticated or clever, theres not even any degree of "AI" involved.
Click to expand...

still pretty cool though, with some further development we could have robots that recognize places/things, think up a word for them, and talk about about them in their own language
 
ng93 said:
still pretty cool though, with some further development we could have robots that recognize places/things, think up a word for them, and talk about about them in their own language
Click to expand...

Surely that's not that hard to program though (from a theoretical point of view, not a "sat down coding it all" point of view)?

You just code it so that when it locates a new place, it generates a name, publishes name to location/name database, then the database is available to other robots.
If it recognises a place from a set of criteria, it looks up to see if it already has a name.


There are already robots that can learn about different types of objects, like chairs and toys, and recognise a different chair or whatever based on pre-determined "this is what makes a chair a chair and not a weird shaped table" criteria.
 
ng93 said:
still pretty cool though, with some further development we could have robots that recognize places/things, think up a word for them, and talk about about them in their own language
Click to expand...

This is the problem they aren't "thinking" up a name, its a simple bit of programming using some simple (human) language rules to generate a random word, its even funnier that they (the programmer/designer) are trying to make them (the robots) generate words which vaguely look like names to humans, makes no difference to a robot (even that phrase is misleading) if the name is something like "fexo" as from the example or just a long numberic like 123423424324.

AI and the likes is depressing as despite what some people want to believe/the media makes out at times, the reality is we are 100s of years from developing an AI that can truly apply abstract processing of its sensory input to generate an output.
 
Last edited:
Rroff said:
AI and the likes is depressing as despite what some people want to believe/the media makes out at times, the reality is we are 100s of years from developing an AI that can truly apply abstract processing of its sensory input to generate an output.
Click to expand...


I fully understand what you are saying here.

At the same time I find that kind of sad and I question our own human intelligence. We as supposed intelligent entities can't design intelligent robots, why is that?

I genuinely believe that a truly intelligent species can design a species more intelligent that itself. Unless that criteria is met, then no species is intelligent.
 
tres said:
Skynet was launched in 1969
Click to expand...

I used to have a Skynet mouse mat - my dad worked on that project.

Back on topic - I'm sure that much of the subtlety of the work is missed by that article, or it would be fairly useless. An intersting experiment would be to combine recognition software, such as is used in the Asimo robots, with this language software - so, for example, identifying that a place is not only a specific place, but a type of place, and with specific attributes. This essentially adds adjectives to a primarily noun-based lexicon, allowing questions such as "what is this new place like?". The robots could then attempt to identify new locations without knowing exactly where they were, and new places could be named with significant names (such as "fezo-like-pizo"). This could then be extended to objects, and then the language would be truly significant.
 
wilko49 said:
The robots could then attempt to identify new locations without knowing exactly where they were, and new places could be named with significant names (such as "fezo-like-pizo"). This could then be extended to objects, and then the language would be truly significant.
Click to expand...

This is where the complexities start to creep in, you could program the AI to deal with one place being like another but what if you then programmed one robot to ask another "when was the last time you were at <name>?" or "when you were at <name> what other robots were there?" without being programmed to respond to those specific inputs it wouldn't be able to process them and come up with an output. So you need the ability to use language to associate stored data to come up with likely responses - and as an aside what would provoke the robot to ask those kinds questions of another one :D and I'm tired and rambling.
 
You must log in or register to reply here.
Back
Top Bottom