mindhive home

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
VWFall: A.I. & Pet Projects, The Dawn of a New Age
Rating 5 V
 
Svetlana
post Oct 12 2007, 11:39 PM
Post #1


Good
Group Icon

Group: Staff
Posts: 2,978
Joined: 11-November 05
From: Seattle, USA
Member No.: 13





Weeks ago, as I perused the extensive list of panels to attend to at the Virtual Worlds Fall in San Jose and planned accordingly, one track immediately caught my attention. As it’s a topic that we’ve been known to discuss here on mindhive before, I scheduled my morning interviews so that I was able to listen to the ‘Artificial Intelligence for Virtual Worlds’ lecture given by leading A.I. scientist Ben Goertzel of Novamente, based in Maryland.

The room was crowded despite the competing and tempting tracks offered in other quarters. Goertzel took the stage alongside Wendy Cornish of the Electric Sheep Company to fill us in on their latest ‘pet project’, as well as the future of artificial intelligence on the virtual landscape.

Goertzel led the discussion and underscored the importance of artificial intelligence in virtual worlds. “A.I. and virtual worlds have tremendous synergy with one another and can mutually benefit from each other,” he told his audience. “Some people think you need an actual physical body for A.I., and others don’t believe so at all. I think virtual robotics gives us a body with perception, linguistic expression and interaction with others,” Goertzel said. For this reason, his company, Novamente (of which he serves as Chief Scientist), is collaborating with ESC to bring virtual pets into Second Life. Why pets, you might be wondering, and not a more sophisticated form of intelligence? As Goertzel explains, evolution mandates that you begin at the basics. With regards to A.I., dogs are among the most basic intelligent creatures that eventually, with adequate research and experimentation, can become a higher form of intelligence such as parrots, followed by babies, children and ultimately an A.I. humanoid or adult.

But before he dives into the technical aspects of A.I., Goertzel makes a clear distinction in types of artificial intelligence. The general A.I. that most of us have already encountered in some form or other is termed ‘neuro A.I.’. Narrow A.I. is, effectively, unable to adapt and has been preprogrammed to respond in certain ways. An example of narrow A.I. are the various Chess computers which can one can challenge. A particular chess game beat Garry Kasparov, the world’s best chess player, yet if we were to change the rules it would be unable to adapt in any way, unlike Kasparov who could change his game accordingly. Conversely, Artificial General Intelligence, or A.G.I., acts as a real and adaptable form of intelligence that can handle situations that its developers did not foresee and accordingly adapt. Goertzel and ESC are working with AGI creatures in Second Life, but the research has posed some very challenging questions.

“How do you make artificial general intelligence? It isn’t as specific, or easy, as narrow-A.I.,” Goertzel tells us. “How do I create an A.I. that would be like a baby and organically grow? After all, babies become intelligent over time. And this is also the correct path for A.G.I., in my opinion. Let it start out simple in the world and become more intelligent.” Goertzel admits that real environments are much richer than virtual ones, enabling intelligent beings to interact with more things and thus learn more from them. Second Life, however, is the virtual world currently available that most closely mimics the real world in that there are hundreds of thousands of users, a variety of objects and land areas, etc.
When Novamente tested their first A.G.I. ‘pet’ in a virtual environment, they began with those baby steps that Goertzel referred to. “We taught our pet about object permanence, something we all learn as babies around 6 months of age,” he continues, “this enables the A.I. to understand that objects remain even when they are not visible, such as in a box.”

The virtual pets that Novamente and ESC are working on have two forms of behavior, both spontaneous and explicitly learned. Spontaneous behavior is instinctive to some degree, and dictates things such as food and water, play and reproduction. However, they also need to learn things that can guide them in their goals. For instance, they must know to stop playing and find water when their ‘brains’ tell them that they require fluids urgently. Advanced features of A.G.I. can create sub goals, enabling them to effectively prioritize their wants and needs. Explicitly learned behavior must be learned from teachers, as the term suggests. This is done through a combination of imitative learning and reinforcement learning. Goertzel prefers imitative learning, as reinforcement learning (such as giving the dog a bone each time it does what it is told to do) takes too much time. As he says, you can teach a dog to dance by giving it a bone after each dance step it takes, but ultimately it takes too many trials to prove worthwhile and does not enable you to comb through extensive amounts of training in various areas. Imitative learning, however, is something that animals such as dogs and cats pick up on quickly.

So what exactly does this A.G.I. pet look like, you might be wondering? Thanks to an online demonstration by Goertzel, the pet looks remarkably like a Beagle puppy, complete with adorable floppy ears and wagging tail. For its sake, it doesn’t hurt that it looks so cute given the fact that, like a regular puppy, he will rely upon his innate programmed senses (such as finding water and food) coupled with his charm to gain the attention and adoration of his teachers, who will ultimately be responsible for his happiness and intelligence. We watch as two avatars, belonging to Goertzel and Cornish, meet the puppy in an outdoor park designed for their testing purposes. Goertzel’s avatar begins demonstrating how the dog has learned to sit, lie down and roll over based on imitative learning and positive reinforcement. The dog also occasionally sniffs the ground and walks over to his water bowl, slurping up the H2O and excitedly wagging his tail afterwards. At one point, the dog is playing with his trainer when a cat is introduced into a tree. After hearing the meows and hissing, the dog immediately runs over to the tree, anxiously looking up at the cat as if to discover what, exactly, it is. The pup also recognizes basic voice commands. Goertzel and Cornishs’ avatars teach the dog how to fetch a Frisbee by demonstration, much like an actual dog might learn. Goertzel throws a Frisbee into the grass, and the dog looks perplexed, unsure of what is expected of him. When Goertzel throws the Frisbee to Cornish, Cornish carries it back to Goertzel, demonstrating the fetch. Goertzel congratulates Cornish’s avatar, thus demonstrating what is expected of the dog. After several trials, the A.G.I. dog begins to fetch and return the Frisbee to Goertzel himself.

“In Second Life, we’re trying to create virtual pets that can achieve their own goals, such as getting food and water and entertaining themselves, “ Goertzel explains, noting that dogs are not the ultimate goal. “As they get smarter, we can implement parrots that talk and can critically improve their own language and eventually humanoids. This is my bet for how computers will evolve and speak,” says Goertzel. Going through the cognitive stages of development as defined by Jean Piaget, Novamente’s latest achievement is only at the very beginning- the infantile stage of development. Following this stage is concrete operational (as defined by Goertzel to be parrots and more advanced animal creatures), followed by the formal stage, such as virtual babies and children. Ultimately, the goal is the final stage of development, otherwise known as full self-modification, in which A.G.I. could function as humans. “My vision is bold. I’m looking at this as a way for A.I. to be more intelligent than human beings,” Goertzel tells his amazed crowd. “It’s quite sci-fi, in many ways, but these are just the first stages.” How far off is that dream? Perhaps as close as 10-15 years he says, given ‘Moore’s Law’ of the exponentially increasing technological developments.

For those interested in the technical aspects of what programs are being utilized, Novamente is using C++ in addition to an integrated cognitive architecture-based system for cognitive science that allows for semantic nets and neural nets. So, as an animal learns, new nodes and links are created. Genetic algorithms and probabilistic logic is also extensively utilized. “We’ve got a lot of sophisticated stuff on the back-end,” Goertzel tells us. And while all of the animals will have the same origination, there are mechanisms in the code to ensure that each animal’s specific knowledge does not share with others, unless allowed. For this reason, the guy that chooses to spend hours teaching his dog how to can-can may rest assured that other dogs will not develop this ability unless trained themselves.

Wendy Cornish was asked when the general Second Life audience may expect to see virtual pets for themselves or additional information, and tentative plans are to release further news on these virtual, A.G.I. pets sometime in the first half of next year. From a scientific perspective, this is a big news, but from a marketing and enriching social perspective of virtual landscapes, it’s likewise monumental.

As this an exciting development in the world of artificial intelligence and virtual worlds, look for additional articles and interviews related to Novamente and ESC in the coming months from us!




--------------------
"The future is here, it's just not widely distributed yet." - William Gibson
Go to the top of the page
 
+Quote Post
King Buzzo
post Oct 13 2007, 12:17 AM
Post #2


Capable
Group Icon

Group: Staff
Posts: 2,199
Joined: 9-November 05
Member No.: 10





It sounds like an incredibly interesting topic. Is the thrust just research at this point or do they have some specific goals and ideas with Second Life in particular? Did they get into the ethics/technology fears of AI at all?

Great article!
Go to the top of the page
 
+Quote Post
Svetlana
post Oct 13 2007, 03:34 PM
Post #3


Good
Group Icon

Group: Staff
Posts: 2,978
Joined: 11-November 05
From: Seattle, USA
Member No.: 13





QUOTE (King Buzzo @ Oct 12 2007, 05:17 PM) *
It sounds like an incredibly interesting topic. Is the thrust just research at this point or do they have some specific goals and ideas with Second Life in particular? Did they get into the ethics/technology fears of AI at all?

Great article!


Your questions will directly relate to a follow-up article I'll be doing in the near future :dirol: . But, to answer for now, no- the discussion did not address the ethical dilemmas associated with A.I., nor did it give any definitive and outlined goals for how Second Life participants might expect virtual pets to impact them (save to say that a news release will be coming the first half of 2008).


--------------------
"The future is here, it's just not widely distributed yet." - William Gibson
Go to the top of the page
 
+Quote Post
Guest_Wangxiang Tuxing_*
post Oct 19 2007, 01:26 PM
Post #4





Guest






It is a very informative article. But I think you should write 'narrow A.I.' instead of 'neuro A.I.' ;)
Go to the top of the page
 
+Quote Post
Dino
post Oct 19 2007, 06:45 PM
Post #5


Experienced
Group Icon

Group: Worker
Posts: 2,500
Joined: 22-September 06
From: Oakland, CA
Member No.: 339





I haven't wanted to respond to this really but I certainly hope they provided a little bit more meat. What you have described seems more like hocus pocus pseudo phych babble. I know a bit about behavior pysch as it applies to animal training (worked for a dog trainer who used to do animal training for zoos) and I'm not very impressed. I also don't see how the programs actually learn and evolve.

Don't mean to be negative but I don't see much here. *shrugs*
Go to the top of the page
 
+Quote Post
Svetlana
post Oct 19 2007, 09:09 PM
Post #6


Good
Group Icon

Group: Staff
Posts: 2,978
Joined: 11-November 05
From: Seattle, USA
Member No.: 13





QUOTE (Dino @ Oct 19 2007, 11:45 AM) *
I haven't wanted to respond to this really but I certainly hope they provided a little bit more meat. What you have described seems more like hocus pocus pseudo phych babble. I know a bit about behavior pysch as it applies to animal training (worked for a dog trainer who used to do animal training for zoos) and I'm not very impressed. I also don't see how the programs actually learn and evolve.


Silly Dino, you needn't be sorry. :wackygirl: The article does indeed cover the exact content of the lecture, so you haven't missed anything in that regard. It most definitely seems to conjoin behavioral science with technology. But no specific definitions of how the pets evolve were given (presumably because it is still being tested) with the exception of the information that it operates with C++ and integrates a 'cognitive science system' that forms links and nodes and the like (much like a human brain). But, still your question requires a direct response from the scientists themselves rather than from me. For that reason, I've noted your questions and will inquire when I speak with them in the coming month.

QUOTE (Wangxiang Tuxing @ Oct 19 2007, 06:26 AM) *
It is a very informative article. But I think you should write 'narrow A.I.' instead of 'neuro A.I.' ;)

I've finally fixed it... damned accents! Thx :)


--------------------
"The future is here, it's just not widely distributed yet." - William Gibson
Go to the top of the page
 
+Quote Post
Dino
post Oct 20 2007, 12:12 AM
Post #7


Experienced
Group Icon

Group: Worker
Posts: 2,500
Joined: 22-September 06
From: Oakland, CA
Member No.: 339





QUOTE (Svetlana @ Oct 19 2007, 02:09 PM) *
'cognitive science system'

Try to pin them down on what that means in layman speak. At least how it relates to the programming. Quite honestly, it's vague enough to make MindArk proud.
Go to the top of the page
 
+Quote Post
Svetlana
post Oct 22 2007, 06:19 PM
Post #8


Good
Group Icon

Group: Staff
Posts: 2,978
Joined: 11-November 05
From: Seattle, USA
Member No.: 13





QUOTE (Dino @ Oct 19 2007, 05:12 PM) *
Try to pin them down on what that means in layman speak. At least how it relates to the programming. Quite honestly, it's vague enough to make MindArk proud.


LOL!

Will do Dino :thumbsup: You've got me interested.


--------------------
"The future is here, it's just not widely distributed yet." - William Gibson
Go to the top of the page
 
+Quote Post
Dino
post Oct 22 2007, 07:46 PM
Post #9


Experienced
Group Icon

Group: Worker
Posts: 2,500
Joined: 22-September 06
From: Oakland, CA
Member No.: 339





For the record, I should probably add that the only reason I pick on Marco and MA like I do is I'm pretty sure Marco knows I have respect for him and believe that within the constraints put upon him by his bosses he does a very good job. I may not like what I get from him sometimes but that doesn't change my respect for him.
Go to the top of the page
 
+Quote Post
Svetlana
post Oct 23 2007, 08:11 PM
Post #10


Good
Group Icon

Group: Staff
Posts: 2,978
Joined: 11-November 05
From: Seattle, USA
Member No.: 13





QUOTE (Dino @ Oct 22 2007, 12:46 PM) *
For the record, I should probably add that the only reason I pick on Marco and MA like I do is I'm pretty sure Marco knows I have respect for him and believe that within the constraints put upon him by his bosses he does a very good job. I may not like what I get from him sometimes but that doesn't change my respect for him.


Is that your official disclaimer? :girlglasses:


--------------------
"The future is here, it's just not widely distributed yet." - William Gibson
Go to the top of the page
 
+Quote Post
Dino
post Oct 24 2007, 02:55 AM
Post #11


Experienced
Group Icon

Group: Worker
Posts: 2,500
Joined: 22-September 06
From: Oakland, CA
Member No.: 339





I just don't want to log in and find my imp FAP missing :P


Actually, I've just found myself sounding pretty negative lately and didn't like it. PE gave me a great and thoroughly enjoyable 4 years and the fact that I am a little disenchanted now shouldn't take away from that. And amazing as it might sound, I actually meant what I said (which pretty much makes it NOT a disclaimer)
Go to the top of the page
 
+Quote Post
Svetlana
post Oct 25 2007, 08:02 PM
Post #12


Good
Group Icon

Group: Staff
Posts: 2,978
Joined: 11-November 05
From: Seattle, USA
Member No.: 13





QUOTE (Dino @ Oct 23 2007, 07:55 PM) *
I just don't want to log in and find my imp FAP missing :P
Actually, I've just found myself sounding pretty negative lately and didn't like it. PE gave me a great and thoroughly enjoyable 4 years and the fact that I am a little disenchanted now shouldn't take away from that. And amazing as it might sound, I actually meant what I said (which pretty much makes it NOT a disclaimer)


I don't think you're alone with those sentiments ;) Just like high school or university, they aren't days I'd do all over again if I had the chance, but they are certainly times I keep near and dear to me.


*Sveta, wondering how we get so off-topic, pours herself another glass of Spanish Tempranillo.


--------------------
"The future is here, it's just not widely distributed yet." - William Gibson
Go to the top of the page
 
+Quote Post

Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 

RSS Lo-Fi Version Time is now: 23rd September 2018 - 10:05 PM