[agi] Dogs can learn language...
>From the New York Times today ... This is perhaps pertinent to Peter Voss's notion of "dog-level intelligence" ;-) ben Research Shows Dogs Can Learn Words By THE ASSOCIATED PRESS Published: June 10, 2004 WASHINGTON (AP) -- As many a dog owner will attest, our furry friends are listening. Now, for the doubters, there is scientific proof they understand much of what they hear. German researchers have found a border collie named Rico who understands more than 200 words and can learn new ones as quickly as many children. Patti Strand, an American Kennel Club board member, called the report ``good news for those of us who talk to our dogs.'' ``Like parents of toddlers, we learned long ago the importance of spelling key words like bath, pill or vet when speaking in front of our dogs,'' Strand said. ``Thanks to the researchers who've proven that people who talk to their dogs are cutting-edge communicators, not just a bunch of eccentrics.'' The researchers found that Rico knows the names of dozens of play toys and can find the one called for by his owner. That is a vocabulary size about the same as apes, dolphins and parrots trained to understand words, the researchers say. Rico can even take the next step, figuring out what a new word means. The researchers put several known toys in a room along with one that Rico had not seen before. From a different room, Rico's owner asked him to fetch a toy, using a name for the toy the dog had never heard. The border collie, a breed known primarily for its herding ability, was able to go to the room with the toys and, seven times out of 10, bring back the one he had not seen before. The dog seemingly understood that because he knew the names of all the other toys, the new one must be the one with the unfamiliar name. ``Apparently he was able to link the novel word to the novel item based on exclusion learning, either because he knew that the familiar items already had names or because they were not novel,'' said the researchers, led by Julia Fischer of the Max Planck Institute for Evolutionary Anthropology in Leipzig. A month later, he still remembered the name of that new toy three out of six times, even without having seen it since that first test. That is a rate the scientists said was equivalent to that of a 3-year-old. Rico's learning ability may indicate that some parts of speech comprehension developed separately from human speech, the scientists said. ``You don't have to be able to talk to understand a lot,'' Fischer said. The team noted that dogs have evolved with humans and have been selected for their ability to respond to the communications of people. Katrina Kelner, Science's deputy editor for life sciences, said ``such fast, one-trial learning in dogs is remarkable. This ability suggests that the brain structures that support this kind of learning are not unique to humans and may have formed the evolutionary basis of some of the advanced language abilities of humans.'' Perhaps, although Paul Bloom of Yale University urges caution. ``Children can understand words used in a range of contexts. Rico's understanding is manifested in his fetching behavior,'' Bloom writes in a commentary, also in Science. Bloom calls for further experiments to answer several questions: Can Rico learn a word for something other than a small object to be fetched? Can he display knowledge of a word in some way other than fetching? Can he follow an instruction not to fetch something? Fischer and her colleagues are still working with Rico to see if he can understand requests to put toys in boxes or to bring them to certain people. Rico was born in December 1994 and lives with his owners. He was tested at home. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Dogs can learn language...
We are not going for 'dpg-level intelligence' per se -- rather, roughly best cognitive abilities of various animals & human infants. Actually our current 'phase3' specs already include Alex's (The Parrot) abilities. Also, I don't see any particular difficulty with our current design/ system learning hundreds of different concept or goal references. Peter -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of Ben Goertzel Sent: Thursday, June 10, 2004 10:49 AM To: [EMAIL PROTECTED] Subject: [agi] Dogs can learn language... >From the New York Times today ... This is perhaps pertinent to Peter Voss's notion of "dog-level intelligence" ;-) ben --- Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.700 / Virus Database: 457 - Release Date: 6/6/2004 --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Dogs can learn language...
Peter, I agree that simply learning mappings between names and objects is easy, and doesn't even require a system as sophisticated as yours is. The slightly more interesting example of reasoning mentioned in this article was: When there were N toys in a room, and the dog knew the names of N-1 of them, and was then asked to fetch "blah", where "blah" was an unfamiliar name -- the dog figured out to fetch the toy whose name it didn't know, figuring that must be a "blah." This isn't rocket science, but it's reasonably sophisticated speculative inference [which could lead to error sometimes, in cases of objects with multiple names, obviously] -- Ben > -Original Message- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of Peter Voss > Sent: Thursday, June 10, 2004 3:08 PM > To: [EMAIL PROTECTED] > Subject: RE: [agi] Dogs can learn language... > > > We are not going for 'dpg-level intelligence' per se -- > rather, roughly best cognitive abilities of various animals & > human infants. > > Actually our current 'phase3' specs already include Alex's > (The Parrot) abilities. Also, I don't see any particular > difficulty with our current design/ system learning hundreds > of different concept or goal references. > > Peter > > > -Original Message- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] > Behalf Of Ben Goertzel > Sent: Thursday, June 10, 2004 10:49 AM > To: [EMAIL PROTECTED] > Subject: [agi] Dogs can learn language... > > > >From the New York Times today ... This is perhaps pertinent to Peter > Voss's notion of "dog-level intelligence" ;-) > > ben > > > --- > Outgoing mail is certified Virus Free. > Checked by AVG anti-virus system (http://www.grisoft.com). > Version: 6.0.700 / Virus Database: 457 - Release Date: 6/6/2004 > > --- > To unsubscribe, change your address, or temporarily > deactivate your subscription, > please go to http://v2.listbox.com/member/[EMAIL PROTECTED] > --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Dogs can learn language...
Quite a bit less than rocket science: negative reinforcement on concept labeling will do the trick (overlayed with 'fetch toy' goal). Behaviorism gets you quite far. Also, you may be surprised how long it takes them to train these animals Alex > 10 years! -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of Ben Goertzel Sent: Thursday, June 10, 2004 11:20 AM To: [EMAIL PROTECTED] Subject: RE: [agi] Dogs can learn language... Peter, I agree that simply learning mappings between names and objects is easy, and doesn't even require a system as sophisticated as yours is. The slightly more interesting example of reasoning mentioned in this article was: When there were N toys in a room, and the dog knew the names of N-1 of them, and was then asked to fetch "blah", where "blah" was an unfamiliar name -- the dog figured out to fetch the toy whose name it didn't know, figuring that must be a "blah." This isn't rocket science, but it's reasonably sophisticated speculative inference [which could lead to error sometimes, in cases of objects with multiple names, obviously] -- Ben > -Original Message- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of Peter Voss > Sent: Thursday, June 10, 2004 3:08 PM > To: [EMAIL PROTECTED] > Subject: RE: [agi] Dogs can learn language... > > > We are not going for 'dpg-level intelligence' per se -- > rather, roughly best cognitive abilities of various animals & > human infants. > > Actually our current 'phase3' specs already include Alex's > (The Parrot) abilities. Also, I don't see any particular > difficulty with our current design/ system learning hundreds > of different concept or goal references. > > Peter > > > -Original Message- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] > Behalf Of Ben Goertzel > Sent: Thursday, June 10, 2004 10:49 AM > To: [EMAIL PROTECTED] > Subject: [agi] Dogs can learn language... > > > >From the New York Times today ... This is perhaps pertinent to Peter > Voss's notion of "dog-level intelligence" ;-) > > ben > > > --- > Outgoing mail is certified Virus Free. > Checked by AVG anti-virus system (http://www.grisoft.com). > Version: 6.0.700 / Virus Database: 457 - Release Date: 6/6/2004 > > --- > To unsubscribe, change your address, or temporarily > deactivate your subscription, > please go to http://v2.listbox.com/member/[EMAIL PROTECTED] > --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- Incoming mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.700 / Virus Database: 457 - Release Date: 6/6/2004 --- Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.700 / Virus Database: 457 - Release Date: 6/6/2004 --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Dogs can learn language...
Hi Peter, Clearly, any particular example of learning can be achieved via some combination of simple mechanisms. What characterizes intelligent living organisms is that they can carry out a whole bunch of examples of learning-types, without being specifically tuned or trained for each one ... "spontaneously" or based on a whole bunch of generic interaction and teaching... Anyway, I actually do believe this kind of learning is within the scope of a system like A2I2, and I'll be psyched to see a demo of your system doing this sorta stuff ;-) As you know my worries about your project don't have to do with relatively simple tasks like this, they have to do with the possibility that you could be building a system that carries out dog-level AI stuff in an "evolutionary dead end" sort of way that doesn't extend well to any system carrying out human-level AI. I.e. you could be "overfitting" the system design to dog-level AI (or "infant-level AI" or whatever phrase you wanna use). But, that's a different conversation! -- Ben > -Original Message- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of Peter Voss > Sent: Thursday, June 10, 2004 4:27 PM > To: [EMAIL PROTECTED] > Subject: RE: [agi] Dogs can learn language... > > > Quite a bit less than rocket science: negative reinforcement > on concept labeling will do the trick (overlayed with 'fetch > toy' goal). Behaviorism gets you quite far. Also, you may be > surprised how long it takes them to train these animals > Alex > 10 years! > > > -Original Message- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] > Behalf Of Ben Goertzel > Sent: Thursday, June 10, 2004 11:20 AM > To: [EMAIL PROTECTED] > Subject: RE: [agi] Dogs can learn language... > > > Peter, > > I agree that simply learning mappings between names and > objects is easy, and doesn't even require a system as > sophisticated as yours is. > > The slightly more interesting example of reasoning mentioned > in this article was: > > When there were N toys in a room, and the dog knew the names > of N-1 of them, and was then asked to fetch "blah", where > "blah" was an unfamiliar name -- the dog figured out to fetch > the toy whose name it didn't know, figuring that must be a "blah." > > This isn't rocket science, but it's reasonably sophisticated > speculative inference [which could lead to error sometimes, > in cases of objects with multiple names, obviously] > > -- Ben > > > -Original Message- > > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On > > Behalf Of Peter Voss > > Sent: Thursday, June 10, 2004 3:08 PM > > To: [EMAIL PROTECTED] > > Subject: RE: [agi] Dogs can learn language... > > > > > > We are not going for 'dpg-level intelligence' per se -- rather, > > roughly best cognitive abilities of various animals & human infants. > > > > Actually our current 'phase3' specs already include Alex's (The > > Parrot) abilities. Also, I don't see any particular difficulty with > > our current design/ system learning hundreds of different > concept or > > goal references. > > > > Peter > > > > > > -Original Message- > > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] > > Behalf Of Ben Goertzel > > Sent: Thursday, June 10, 2004 10:49 AM > > To: [EMAIL PROTECTED] > > Subject: [agi] Dogs can learn language... > > > > > > >From the New York Times today ... This is perhaps > pertinent to Peter > > Voss's notion of "dog-level intelligence" ;-) > > > > ben > > > > > > --- > > Outgoing mail is certified Virus Free. > > Checked by AVG anti-virus system (http://www.grisoft.com). > > Version: 6.0.700 / Virus Database: 457 - Release Date: 6/6/2004 > > > > --- > > To unsubscribe, change your address, or temporarily deactivate your > > subscription, please go to > > http://v2.listbox.com/member/[EMAIL PROTECTED] > > > > --- > To unsubscribe, change your address, or temporarily > deactivate your subscription, please go to > http://v2.listbox.com/member/?> [EMAIL PROTECTED] > > > --- > Incoming mail is certified > Virus Free. > Checked by AVG anti-virus system (http://www.grisoft.com). > Version: 6.0.700 / Virus Database: 457 - Release Date: 6/6/2004 > > --- > Outgoing mail is certified Virus Free. > Checked by AVG anti-virus system (http://www.grisoft.com). > Version: 6.0.700 / Virus Database: 457 - Release Date: 6/6/2004 > > --- > To unsubscribe, change your address, or temporarily > deactivate your subscription, > please go to http://v2.listbox.com/member/[EMAIL PROTECTED] > --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]