From - Wed Jan 14 11:37:06 1998 Return-Path: Received: from relay1.UU.NET by mrco.carleton.ca (4.1/SMI-4.0) id AA04776; Thu, 24 Dec 92 03:31:26 EST Received: from nyx.cs.du.edu by relay1.UU.NET with SMTP (5.61/UUNET-internet-primary) id AA10827; Thu, 24 Dec 92 01:29:26 -0500 Received: by nyx.cs.du.edu (4.1/SMI-4.1) id AA14736; Wed, 23 Dec 92 23:30:11 MST From: ahawks@nyx.cs.du.edu (andy) Message-Id: <9212240630.AA14736@nyx.cs.du.edu> X-Disclaimer: Nyx is a public access Unix system run by the University of Denver. The University has neither control over nor responsibility for the opinions or correct identity of users. Subject: FutureCulture Digest #161 To: future-digest@nyx.cs.du.edu Date: Wed, 23 Dec 92 23:30:10 MST X-Mailer: ELM [version 2.3 PL11] Status: R ______________________________________________________________________ |______________ / | | / | | u t u r e <___________ u l t u r e | _______________________________________________________________________| Issue #161 Wednesday, December 23rd 1992 Today's Topics: --------------- A holographic approach to AIA memory extropians & leri-l Re: MAIA: A practical project __________________________________________________________________________ Date: Wed, 23 Dec 92 00:02:40 PST From: Al Sargent Subject: Re: MAIA: A practical project This MAIA sounds really good. It sounds like a pretty manageable project, since you don't need to worry about data security and network transport issues. But, I do have a few questions. ~~ First, let's think about the comprehender. What does it need to do? First, ~~ it must receive textual information. This can be accomplished by feeding it ~~ text from various mail-lists. After she has the text(s), MAIA must have some ~~ way to comprehend these texts. In my opinion, the most efficient and most ~~ flexible means would be via scripts or schemas (a la Schank). Through the use ~~ of primitive scripts that MAIA would be 'born' with, she could understand such ~~ things as agency, action, location, motion, goals, objects, etc. (speak ~~ english, please, Mark?) Okay. In other words, this means that if she receives ~~ a message from Andy (ahawks) saying '...and I think that _x_', MAIA would then ~~ understand that ahawks (Andy) holds the optinion that _x_. By representing ~~ this inference in the form of a script, we conserve space and may find our way ~~ halfway around the frame problem. Wouldn't this conversion of all this text to scripts or schemas -- even if it was cut down by the MAIA's owner -- take up huge amounts of memory? Would be able to filter out things that are not facts, but basically opinions -- for example the first sentence of this message? Also, would you have some sort of thesarus, that be able to take the word 'Information America' (I think that was it) and know that the writer intended to say 'Amix', which was the case with some messages a day or so ago. How much initial commonsense knowledge would be optimal for the MAIA to have? And finally, do you think your MAIA could ever, given enough time to learn (and programming effort on your part) pass the Turing Test? (The Turing Test was proposed by Alan Turing back in the 40s, and it was a process by which a person would type any statement or question to a terminal, and receive an answer back from the comupter. If the person could not tell whether the agent at the other end was another person or a computer, then the computer is said to have passed the Turing Test. It hasn't happened... yet.) Al ______________________________ Date: Wed, 23 Dec 92 02:44 CDT From: Institute for Intelligent Systems Subject: Re: MAIA: A practical project Regarding the memory issue: Since all information will be stored in script form, much information can be contained in a single script (and scripts may contain scripts, allowing for recursion, see Schank or Dresher for detailed methods of script memory systems). For example, a RESTAURANT script would contain all information pertaining to restaurants, including examples, customers, prices, types, menus, waitrons, etc. etc. etc. Each of these topics could themselves be scripts. Scripts are a very efficient way of storing a large amount of (somewhat) nonlinear, relational, taxonomic, causal, and hierarchical info. Regarding thesaurus like capabilities and innate (born) knowledge: through experience the MAIA could infer simlies, metaphors, and multiple instantiations of the same topic or referent (again, see Dresher for an account of this). As far as common-sense, lexical, or world knowledge innate to the MAIA, not much is necessary (again...see Dresher). Since the MAIAhas the capability of creating new scripts (frames, schemas, whatever), she can _learn_ whatever knowledge is necessary from a very limited set of intial frames. As far as understanding words, I have a copy of a program called ANN, written in BASIC (back when line numbers were still necessary, yet!), that has a vocabulary of about 20,000 words, and a verbal IQ of about 120. ANN has the capability to learn new words as well. (Based on neural net techniques and spreading activation). A similar program, called CCReader (in LISP) is comparable and just as add-on-able. Would a program such as this pass the TT? No. Because I'm biased - the TT is not, and was never intended to be, an actual test of machine intelligence; rather, it was a thought experiment to demonstrate just how difficult it would be to program a computer to perform in a human manner. Even if it were subjected to the TT, it would probably not pass, because there are always trick questions that could trip up the computer. Ask it about a Zen koan or the such...be self-referential with it in an absurd manner. My answer to the TT: if it works, don't fix it. I could think of a few friends who wouldn't pass the TT. Sorry, I forgot the rest of the issues, since it's 3am (my fave posting time), and this damn VMS editor wont quote from email. Grrrrr!) Anyway, there. -fugue (aka Mark Langston) langston@memstvx1.memst.edu "Secrecy is the beginning of tyranny." fugue@mindvox.phantom.com ______________________________ Date: Wed, 23 Dec 92 03:48 CDT From: Institute for Intelligent Systems Subject: A holographic approach to AIA memory Okay. Final post for the night. I swear. (And I didn't even have any coffee today! (Do you know what an accomplishment that is for me?)) One of the main concerns I hear echoed about such intelligent agents roaming (or just observing) the net is PROCESSING and SPACE constraints. There may be a way around both of these. The idea I am about to discuss is related in similar fashion in Nov. '92 BYTE (Information Theatre), and in several forms by Horgan & Tienson, and Horgan (email me for exact ref's ..I'm at home, the ref's arent). Until now the discussion of any kind of memory for an intelligent agent has assumed a large, explicit database for storage of information. This database has taken one of two forms: resident on a single computer, or distributed among many computers. Either way, the assumption is: the sucker's gonna get LARGE. There are several problems with such storage techniques: it takes a lot of space to store such a database, it takes complicated methods to search such a database, such a search takes a long time and a lot of cpu time. There may be a way around this. Imagine every unique term in a database (atom to you lisp programmers, datum to statisticians or programmers, predicate to logicians, word to writers etc.) as a dimension in a hyperspace. E.g., an encyclopaedia may exist in a hyperspace of 30,000+ dimensions. If each term in a database is defined as a unique dimension in a hyperspace, any given collection of terms (a paper, letter, idea, concept, email message, book, etc.) may be described as a vector in this hyperspace. Furthermore, computations may be performed over this hyperspace to find distances or other relations between vectors. These relations can be construed as similarity of content, similarity of context, etc. E.g., vector A and vector B may be very close in this hyperspace; in reality vector A represents a sales pitch for UNIX and vector B represents a bug report for UNIX. The two are closely semantically related. If an agent's memory is construed as a collection of vectors in a hyperspace, symbolic computations may then be performed on the hyperspace without the actual database being present. In other words, an agent's memory could be represented as vectors in a high-dimensional matrix, and the database could be stored elsewhere. What are the benefits to such a memory system? There are many. Search time would be cut down several orders of magnitude, since random sampling of vectors could quickly extract needed information. Second, all information contained in memory would be represented by a fixed-length vector. An email message and PacBell's phone book would both exist as vectos of the same length. Third, the agent would only need to reference the actual database once the information has been found. This is akin to using an index to find a specific topic or page. Fourth, Any changes to the actual database requires only a change of a vector, and does not affect any symbolic operations for search or retrieval. Fifth, such an holographic representational system allows for storage of natural-language text, instead of its conversion to unwieldly predicate calculus or some such. Lastly, the holographic memory allows for computation over representation via vectors at many granularities. In plain english, using such a technique one can just as easily pinpoint a specific word as one could a set of items concerning a specific topic. Just as easily, and just as accurately. And probably faster. What does this mean for AIAs and their counterparts? Instead of having a ton of informational luggage to lug around, the agent can simply carry (or keep) the set of vectors and a matrix representation, and then send a specific request (consisting of one or more vectors) to the actual database to get the actual information. I know this may seem a bit vague, and I'm just as sure I'll be prodded into elucidating it in the coming weeks. But sleep finally caught up with me, and I better jump at the chance before it goes out for beer and pretzels without me. 'Til then, Fugue (aka Mark C. Langston) langston@memstvx1.memst.edu "Secrecy is the beginning of tyranny." fugue@mindvox.phantom.com ______________________________ Date: Wed, 23 Dec 1992 14:57 CST From: Scotto mark carranza: >I'd like to initiate a query into the improvement of this medium, the mail >list, certainly my use of it, perhaps even this mail list, although I think >that might just be asking for trouble, I need more tact and wisdom. Well, while I've never been able to provide anything resembling tact or wisdom, I can certainly supply a bit of sarcasm and useless pomposity. Thus, we begin to answer yer questions... >- Are there folks out there who use this _good_? This is an ambiguous statement. Do you mean people who use a mailing list *as* a "good", or do you mean people who use a mailing list "well"? I'm going to operate under the assumption that you mean the latter, and to that I would say: Yes, there are folks out there who use lists well. More on this in a moment. >- Is there a mailing list FAQ? Again, a little ambiguous. I've never seen a FAQ about running a list, but I can presume such FAQs might exist; I know of two mailing lists devoted to the topic of running mailing lists and using LISTSERV software, and they might just have FAQs. There are also files available that contain giant lists of mailing lists, but I don't think that's what you're looking for... >- Why isn't this easier? How can communicating well be simpler? Once again, ambiguous. :) Why isn't this easier than *what*? Easier than talking face to face? The question of how can communicating well be simpler is a question that has plagued us since *well* before the dawn of the Internet. I have a feeling that some of these problems are merely the standard linguistic problems inherent in communication finally rearing their ugly heads in cyberspace. :) The Internet is an unusual forum for information transfer, and there are, I have found, two avenues of approach to this: one says that we are here for the dissemination and absorption of as much pure information as possible, and the other says that we are here to create a new community in cyberspace -- a sort of bizarre sociological experiment into the ways and means people might relate without their bodies around to help. I've found that certain mailing lists -- Extropians, for example -- are extremely high on information transfer and relatively unaware or, more likely, disinterested in the notion that that place is a new sociological medium. That's pretty much, IMO, why so many flame wars break out so often over there -- on that list, miscommunication is highly annoying, because it garbles the transfer of information. On other lists -- Leri-L, for example -- the sociological elements take priority. Information is being transferred, certainly, but what is more important is how the individual people choose to transfer that info: how they choose to communicate. Each person there seems to be aware of the notion that regular rules for holding a conversation don't need to apply in this environment, and "tweaking the model" is a favorite pasttime. I see FutureCulture as being somewhere in the middle. There is a tendency toward deliberate trading of facts, but individual styles quite often tend to pop up. So...how to make this simpler? I don't think that's a topic that can be reasonably addressed. You will never be able to make it simple enough to allow for pure communication until you somehow transcend the linguistic mechanisms of communication (this could happen, of course, beyond some singularity or another, but that's sort of off-topic, I suppose...:)). The more reasonable approach is to find the environment that most suits what you feel comfortable with as far as methods of communication. There are refinements to be made everywhere, but I don't believe they can happen under the guise of "making communication simpler," but rather, understanding the full scope and limitations of the medium at hand. >- What could lead to more tightly coupled interaction as these group > focus/business/net/commune type posts are pointing to? Well, that depends. On a purely mechanical level, not everyone on the Internet has the time and access to allow for such grand ideas as merging this list with that one or creating one big list that has all the other lists on it, etc, etc. I think a *much* more effective way of achieving tighter interaction is by doing what a few people do already: sign up for damn near *every*thing that seems to be of interest, see how you can personally draw links between them, and devote energy to bringing them together on the personal level. Paco has pointed out to me that the distinctions between mailing lists on the Internet are quite useful -- the kind of stuff Leri-L generates will never ever be appropriate fodder on Extropians, etc. But that doesn't rule out my being on both lists and trying to incorporate ideas from both lists into my personal philosophy. I'm not sure how to make these lists "interact"; I'm not even sure that that's what you were asking, of course. :) >- How easy it is to start a list, to maintain it? Hmmm. I'm sort of an oddball in this department. I started Leri-L by simply asking our local sysadmin to set it up and allow me to run it. He did, not knowing at the time what a strain it would be on his simple little vax machine; our school had never run a mailing list before. But I get the feeling that if you're one of those crafty programmer types with any kind of interesting access, you can easily set one of these jobbies up. Maintaining it manually takes a bit of time, but there are several programs available -- LISTSERV is the only one I can think of by name, but I know Andy has some kind of automated thing going, and Leri will soon -- that can handle the grunt work for you. I find that "maintaining" the list also takes into account, on a much less mechanical level, keeping it "on topic" if such a thing is desired, etc. And that all gets back to stuff I talked about above. >- What are the dynamics of group formation, persistence, and interchange? >- Information flows, ideas squirt out here like diahrrea after breakfast, > but where is it sticking, where does intelligence persist? Where does intelligence persist? My friend, if I knew anyone who could effectively answer that question, I'd be a much happier person. :) I won't address that part of the question. As far as group formation, this still all goes back to the above delineation. Your group will form and persist depending upon the nature of *how* you intend to communicate, much more so than on *what* you intend to communicate. There's quite a bit of intersection of topic between Extropians and FutureCulture, but the method of discourse is *so* varied as to make them incompatible. Hope this helps a bit. I can elaborate at will, of course...:) ______________________________ From: Petrisse Briel Subject: extropians & leri-l Date: Wed, 23 Dec 92 18:07:10 PST What are the stated intentions of Extropians and Leri-l? They sound inter- esting... ----------- All recent anecdotes about the exceptions to the idea that viruses generally do not kill hardware are duly noted and accounted for. I consider myself lucky that the PET disappeared. Kevin _________________________________________________________________________ | | | That's all for today! | | To send a message to the list: future@nyx.cs.du.edu | | To subscribe/unsubscribe/change format: future-request@nyx.cs.du.edu | | All other requests: future-request@nyx.cs.du.edu | | List Maintainer is: (andy [aka hawkeye]) ahawks@nyx.cs.du.edu | |_________________________________________________________________________| | | | The opinions expressed in FutureCulture are those of the individual | | author only. | |_________________________________________________________________________|