"YKY (Yan King Yin)" <[EMAIL PROTECTED]> wrote: On 4/19/07, Russell Wallace <[EMAIL PROTECTED]> wrote: > Oh, I agree with your emphasis on producing something useful. > > I think the ability to generate 3D images and video on demand will be very > useful in areas like movies, video games, mechanical engineering, > architecture; my vision of what an AGI should look like could be described as > a smart CAD program. > > Now, I agree answering commonsense queries would also be useful, and by all > means let that also be a feature. But what will happen to a project that > focuses _only_ on text I/O? I can answer that: short-term progress using > tricks akin to Eliza, Google etc will be so much easier than chewing on the > hard problems of spatial reasoning etc, that the pressure to use the tricks > will be overwhelming. I don't have the self-discipline that would be required > to stick to the hard long-term path in the absence of any visual output. I > don't believe anyone has. So while such a project may produce improved > versions of Eliza, Google etc, it can never produce AGI.
Now I see your point... But video generation is a creative process that is best done with the planning / acting part of the AGI. Right now I tend to focus on the knowledge maintenance part. Let's say AGI = knowing + acting (eg, Cyc focuses on knowledge, whereas Soar and ACT-R focus on acting.) Very often you *cannot* directly translate a representation in memory to a video sequence. For example, with "there is a book on the table", one can ask: 1. wood or glass tabletop? 2. hardcover or paperback book? 3. is the book open? etc... It is actually a creative process to render what is in the 'mind' to a visual form. In a sense it is similar to the task of NL generation. A second example is to "imagine a certain girl". Now the AGI may have to fill in a lot of visual details (face, clothing, etc) that aren't really there in the memory. And if the AGI draws a stick figure to represent her, that is a creative process. I guess it's good for the AGI to have the ability to draw things or render things in video, but this function may not be implementable as a direct translation... YKY I'll meet you half way on the video point, and take up the unstated challenge by YKY. On the video front, as I said, but will rephrase maybe somewhat differently. I believe for a full fledged autonomous AGI some form of embodiment in a virtual or real world is necessary. That being said, I dont think full-movie quality is necessary by any means, neat yes, but unnecessary. I believe something along hte lines of a fully acting Second Life /The Sims character, that can talk and move and manipulate all opbjects would qualify. This gives the ability to have a Character be told "Go to the Door" or "Get me a soda" and be able to physcially watch and see them do that. This woudl be lacking in a totally text based AI, unless you had it print positions coordinates or something awful like that. Is it a requirement then for the AGI? Well it really depends. Most things can be learned and used via Text only.. Most questions and conversions can be had, but some physical actions cannot adequatly be represented. So I believe you should have both, and should start with the text based option, to learn and interact with a basic agents. Then another group of graphics programmers could easily give you a basic 3d environ, and start adding in mechanical functions for you. Very often you *cannot* directly translate a representation in memory to a video sequence. For example, with "there is a book on the table", one can ask: 1. wood or glass tabletop? 2. hardcover or paperback book? 3. is the book open? etc... It is actually a creative process to render what is in the 'mind' to a visual form. In a sense it is similar to the task of NL generation. For this example, and this is something I have thought about, I want to have the Agent be smart, and since it has limited ability to directly "see" the virtual environment, and the lack of richness there, but given a room and a scene with a table and a book on the table, I would expect an Agent to know the context of objects enough to ask these kind of questions. So I imagine putting a Agent in the world, then it going: AI: Look at the book. AI: What color is the book? AI: Is the book a paperback or hardback? AI: What is the title of the book? AI: Look at the table. AI: What is the table made of? This should make up for the lack of richness and maybe force the user to make a more creative scene, and should show knowledge of the world by asking pertinent questions about the objects it interacts with. I've already tested this out somewhat with the Person object. By having a certain number of Person Frames you start to see patterns such as has-a-brother slot being used very often. So the AGI when meeting a new person would reasonably ask if the person had a brother, without being programmed directly to look for this. Im not sure howextendable this example is, and it doesnt flow together correctly in a conversation, it jsut randomly asks the questions currently. James Ratcliff _______________________________________ James Ratcliff - http://falazar.com Looking for something... --------------------------------- Ahhh...imagining that irresistible "new car" smell? Check outnew cars at Yahoo! Autos. ----- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=231415&user_secret=fabd7936
