"Understanding Instruction Books and Programs"
is a PDF of the paper presented at Worldcomp06 (ICAI'06]- Las Vegas.
http://www.urbanmarket.com/csi2000/pub/Worldcomp06/Understanding%20Instruction%20Books%20and%20Programs.pdf
The paper takes the form of a dialog between two ‘super-observers’ SO1 and SO2 discussing a modified Chinese Room experiment to conduct an examination of seldom-discussed artifacts, the instruction manual and/or the computer program in the room.
The following is the set of slides used during the presentation:
http://www.urbanmarket.com/csi2000/pub/Worldcomp06/Understanding%20Instruction%20Manuals%20and%20Programs_files/frame.htm
This was a powerpoint presentation. Some slides have animation and require you to click on the slide to see the animation.
Wednesday, February 14, 2007
Friday, April 21, 2006
John Serle's Chinese Room Experiment
I have written a paper that has been accepted as a regular research paper at a 2006 International Conference on Artificial Intelligence (ICAI'06: June 26-29, 2006, Las Vegas, USA). The paper examines seldom discussed artifacts in John Searle's Chinese Room experiment - the instruction books and computer programs. This conference is one of 28 concurrent conferences running under the banner of "Worldcomp 06".
The paper came about from some thinking I did about how you can tell whether or not somthing like a computer can think?
Alan Turing suggested a quantitative test. He did so since he thought nobody could truly answer whether of not an entity can think - unless you were that entity.
Searle asks you to indeed be that entity and ask yourself whether or not you can 'understand'.
His test was to think about some program running in some computer that is said to be capable of thinking and understanding. It is receiving patterns that turn out to be Chinese symbols representing some question posed in Chinese. It answers the questions in Chinese.
Now imagine that same program translated into English instructions and given to some man ins a room who carries out all the instructions and indeed seems to be answering questions in Chinese - in Chinese. He now asks you to imagine you are that person - and asks whether you understand Chinese - or will ever understand Chinese. Most people imagining themselves in this position would reply no - they don't see how they would ever learn to understand Chinese. Searle is arguing against 'strong AI' that says the program 'understands' in the same way that a human 'understands'.
I propose a modification to the experiment. I then assume Searle's conditions to hold true - and ask what the program and instruction manual has to look like in order to keep Searle's conditions true. This leads to some insights about what is going on.
I also think that Turing's notion of a quantitative test is critical. All that we know in this universe if by 'observation'. We think we see a chair - but all we really receive is signals in our brain from which we form the concept of a chair.
In fact, current thinking by Wheeler ('it' from 'bit') and Zeilinger's Foundational Principle for Quantum Mechanics re-enforces my notion that 'measurement' when we ask whether or not something thinks, is essential.
I will have the paper published in the proceedings of this conference, and will have the paper and a slide show presentation up on my web site shortly.
The paper came about from some thinking I did about how you can tell whether or not somthing like a computer can think?
Alan Turing suggested a quantitative test. He did so since he thought nobody could truly answer whether of not an entity can think - unless you were that entity.
Searle asks you to indeed be that entity and ask yourself whether or not you can 'understand'.
His test was to think about some program running in some computer that is said to be capable of thinking and understanding. It is receiving patterns that turn out to be Chinese symbols representing some question posed in Chinese. It answers the questions in Chinese.
Now imagine that same program translated into English instructions and given to some man ins a room who carries out all the instructions and indeed seems to be answering questions in Chinese - in Chinese. He now asks you to imagine you are that person - and asks whether you understand Chinese - or will ever understand Chinese. Most people imagining themselves in this position would reply no - they don't see how they would ever learn to understand Chinese. Searle is arguing against 'strong AI' that says the program 'understands' in the same way that a human 'understands'.
I propose a modification to the experiment. I then assume Searle's conditions to hold true - and ask what the program and instruction manual has to look like in order to keep Searle's conditions true. This leads to some insights about what is going on.
I also think that Turing's notion of a quantitative test is critical. All that we know in this universe if by 'observation'. We think we see a chair - but all we really receive is signals in our brain from which we form the concept of a chair.
In fact, current thinking by Wheeler ('it' from 'bit') and Zeilinger's Foundational Principle for Quantum Mechanics re-enforces my notion that 'measurement' when we ask whether or not something thinks, is essential.
I will have the paper published in the proceedings of this conference, and will have the paper and a slide show presentation up on my web site shortly.
Tuesday, March 22, 2005
William Gibson and Vannevar Bush
Stumbled across this discussion in William Gibson's blog of Tuesday, January 28, 2003 [IN THE VISEGRIPS OF DR. SATAN (WITH VANNEVAR BUSH)]. William Gibson, for those who don't know is the one who coined the word "cyberspace" and has written a series of novels involving cyberspace. The latest is "Pattern Recognition".
Gibson talks about Vannevar Bush in this blog. His comments reflect a piece I read in "The Futurist" magazine about a decade ago (and I have lost the link to it, so cannot even remember the name of the article). The gist of the article and the core of what Gibson is talking about here is the emergence of a "cyborg" - a human/machine complex.
If the brain is "merely" a highly interconnected set of neurons that gives rise to conciousness and "us"... then if the internet is "merely" a highly interconnected set of human/machine nodes - will this structure give rise to a "new conciousness" - a new "being"?
Gibson talks about Vannevar Bush in this blog. His comments reflect a piece I read in "The Futurist" magazine about a decade ago (and I have lost the link to it, so cannot even remember the name of the article). The gist of the article and the core of what Gibson is talking about here is the emergence of a "cyborg" - a human/machine complex.
If the brain is "merely" a highly interconnected set of neurons that gives rise to conciousness and "us"... then if the internet is "merely" a highly interconnected set of human/machine nodes - will this structure give rise to a "new conciousness" - a new "being"?
Saturday, February 26, 2005
Way Off the Vannevar Trail
Talking about creating links between things.
The Police in the UK seem to be using something called "Smart Water".
This smart water contains micro dots encoded with your identification. You spray this water on all devices that you own. This establishes a "link" between that object and you (the owner).
If it is subsequently stolen and then recovered, they can identify with 100% certainty that the object belonged to you.
There are two part to this link. The first part is your ID encoded in the micro dots. The second part is in a data base linking that ID to you.
Make that micro dot RDIF compliant - and they can merely ask where your object is at the moment and then go get it.
You can also have a security system that douses any intruder with smart water. This leaves a link on the intruder - which positively says they were in your premises.
Here's Wired article on this product.
The Police in the UK seem to be using something called "Smart Water".
This smart water contains micro dots encoded with your identification. You spray this water on all devices that you own. This establishes a "link" between that object and you (the owner).
If it is subsequently stolen and then recovered, they can identify with 100% certainty that the object belonged to you.
There are two part to this link. The first part is your ID encoded in the micro dots. The second part is in a data base linking that ID to you.
Make that micro dot RDIF compliant - and they can merely ask where your object is at the moment and then go get it.
You can also have a security system that douses any intruder with smart water. This leaves a link on the intruder - which positively says they were in your premises.
Here's Wired article on this product.
Off the Vannevar Trail
A couple of new web sites have hit the news that feature what they call "Tagging".
Basically what this is, you store an object on their site and assign "Tags" to the object. Tags are completely up to you. Generally they are supposed to be how you might classify this object. Hence 'Tags" are merely meta-data that you associate with the object.
For example, a photo might be assigned the tags 'Perth' 'Ontario'
'Canada' and 'Architecture'. Nothing spectacular here - but - the system will then form associative links between any two objects in the base that share one of these tags.
Another example could be a URI of a favorite or useful web site (i.e. bookmark). Again you might add the tags 'XML' 'RSS'. Again the system forms associative links between your link and any other link sharing the same "tag".
Now suppose you open this up worldwide, and you can form these associations across everyones URI's. The benefit comes from you seeing that there are 16 other people in the world who have created a URI that share the tag "RSS". You can visit each person in turn and see all the URIs they have entered with any tag.
This is quite different from what Vannevar Bush envisioned in his 'Memex". He allowed a person to establish an associative link between two objects, and store that association in a named external trail.
These systems create these links automatically and store the links external from the objects themselves. The set of links are then shared across all people who have entered anything.
These systems have no particular intelligence other than recognizing a tag having the value "xyz123" - and building a link between all objects sharing the value of that tag. There is absolutely no intelligence in the process.
However, it does potentially allow you to "discover" valuable new links when you look at the complete set of objects entered by the other people who share that tag. It could be said that you are linked via a Vannevar Trail linking to all people who claim to have an interest in "xyz123", and from there you can see all their links, that are again linked to all other users sharing that persons complete set of tags.
You may find something that you could not possibly find any other way. Note that not even a Google search might have found this since it is a human who found it (perhaps by accident) and assigned this tag to it. That particular tag may not even occur in the object - it is the mind of the person who viewed the object that assigned the tag "xyz123" to it - not a program.
The two websites I am referring to are of course:
FlickR (for a base of pictures/photos)
del.icio.us (for a base of bookmarks)
An interesting graphical interface to FlickR can be found at FlickR Graph. You will need reasonable computer power to use this feature. The interface is compelling and fun to watch. You can check out the display merely by checking ot see who is online and looking at their photo collection.
Basically what this is, you store an object on their site and assign "Tags" to the object. Tags are completely up to you. Generally they are supposed to be how you might classify this object. Hence 'Tags" are merely meta-data that you associate with the object.
For example, a photo might be assigned the tags 'Perth' 'Ontario'
'Canada' and 'Architecture'. Nothing spectacular here - but - the system will then form associative links between any two objects in the base that share one of these tags.
Another example could be a URI of a favorite or useful web site (i.e. bookmark). Again you might add the tags 'XML' 'RSS'. Again the system forms associative links between your link and any other link sharing the same "tag".
Now suppose you open this up worldwide, and you can form these associations across everyones URI's. The benefit comes from you seeing that there are 16 other people in the world who have created a URI that share the tag "RSS". You can visit each person in turn and see all the URIs they have entered with any tag.
This is quite different from what Vannevar Bush envisioned in his 'Memex". He allowed a person to establish an associative link between two objects, and store that association in a named external trail.
These systems create these links automatically and store the links external from the objects themselves. The set of links are then shared across all people who have entered anything.
These systems have no particular intelligence other than recognizing a tag having the value "xyz123" - and building a link between all objects sharing the value of that tag. There is absolutely no intelligence in the process.
However, it does potentially allow you to "discover" valuable new links when you look at the complete set of objects entered by the other people who share that tag. It could be said that you are linked via a Vannevar Trail linking to all people who claim to have an interest in "xyz123", and from there you can see all their links, that are again linked to all other users sharing that persons complete set of tags.
You may find something that you could not possibly find any other way. Note that not even a Google search might have found this since it is a human who found it (perhaps by accident) and assigned this tag to it. That particular tag may not even occur in the object - it is the mind of the person who viewed the object that assigned the tag "xyz123" to it - not a program.
The two websites I am referring to are of course:
FlickR (for a base of pictures/photos)
del.icio.us (for a base of bookmarks)
An interesting graphical interface to FlickR can be found at FlickR Graph. You will need reasonable computer power to use this feature. The interface is compelling and fun to watch. You can check out the display merely by checking ot see who is online and looking at their photo collection.
Wednesday, February 02, 2005
On Intelligence - Jeff Hawkins
An interesting text by the creator of the original Palmpilot and the Treo smart phone has hit the stands. [ISBN 0-8050-7456-2]
Jeff Hawkins is now the Director of the Redwood Neuroscience Institute
Jeff is attempting to create an overall unifying theory about how the brain works. He suggests that "intelligence" lies within the "neocortex" - a six layer structure in the brain.
The neocortex:
In Vannevar Bush's article, the trails were associative indexes, the association done not by machine, but by humans, who excel in establishing associative relationships. An interesting question is whether Vannevar Bush perhaps thought of the trails as hierarchical? Certainly they were a separate named collection of associations, each association being a connection between parts of two separate documents, perhaps interspersed with original text. There is some suggestion that he might have seen these as hierarchical since he talks about a trail about English Crossbows including streams on "strength of materials", "history", etc.
Jeff Hawkins is now the Director of the Redwood Neuroscience Institute
Jeff is attempting to create an overall unifying theory about how the brain works. He suggests that "intelligence" lies within the "neocortex" - a six layer structure in the brain.
The neocortex:
- stores sequences of patterns
- recalls patterns auto-associatively
- stores patterns in an invariant form
- stores patterns in a hiearchy
In Vannevar Bush's article, the trails were associative indexes, the association done not by machine, but by humans, who excel in establishing associative relationships. An interesting question is whether Vannevar Bush perhaps thought of the trails as hierarchical? Certainly they were a separate named collection of associations, each association being a connection between parts of two separate documents, perhaps interspersed with original text. There is some suggestion that he might have seen these as hierarchical since he talks about a trail about English Crossbows including streams on "strength of materials", "history", etc.
Wednesday, February 25, 2004
Grid Computing
Although not trail related.. it is about connecting millions of computers in the largest computing net on earth.
This is a news release about Grid Computing surpassing the 2.5 million computer mark. News Release
I have a main computer hooked up to this grid and am approaching a full year of computer time donated to the GRID.
This is a news release about Grid Computing surpassing the 2.5 million computer mark. News Release
I have a main computer hooked up to this grid and am approaching a full year of computer time donated to the GRID.
Thursday, January 29, 2004
How to Create a Vannevar Trail
Vannevar Bush envisioned some sort of apparatus in a desk that would electronically allow linking of documents (represented by images on indexed roles of microfilm). It is clear that this linking could not involve the original images, as they were read-only on microfilm. It was also clear that whatever form these links would take, they were not only separate from the original documents, they could be shared with other people, as well as being intermixed with original text by the author.
I decide to call these things (or special type of document if you will) a Vannevar Trail [VT for short].
I also want to explore how XML technology might be employed to implement the Memex concept. In particular, the most general form of XLink - where the links are stored external to any documents seemed like a good candidate technology to implement a VT.
I decide to call these things (or special type of document if you will) a Vannevar Trail [VT for short].
I also want to explore how XML technology might be employed to implement the Memex concept. In particular, the most general form of XLink - where the links are stored external to any documents seemed like a good candidate technology to implement a VT.
Subscribe to:
Posts (Atom)