No. There are statistical commonalties across all text and relationships of words (which are symbols) to all other words and groups of words' relationships to the other words and groups of words. The mathematical knowledge structure behind the text representation, the text linguistical representation is like a GUI, is quite detailed and extravagant. There is much there especially as the quantity of text increases and non-fictional forms of text are detected and processed. The text-only AGI extracts knowledge from the text and communicates using it as a medium. You have to be careful as there are gaps and fuzzy areas and the AGI could be somewhat air-headed in certain areas of knowledge as it could misinterpret text especially fiction and come up with crazy ideas and myths. It would continually struggle with reality. But for it to be more than just a glorified chat-bot the core engine needs more than just inter word relationship processing.
Text is easy, other types of input are really tough like video - difficult to tame so text is good for development and testing. Supposedly a real AGI would be able to deal with any input and mine knowledge out of it but I'm sure in practical reality there is a lot of AI filtering of the data going on just like with sensory input on humans. John > From: a [mailto:[EMAIL PROTECTED] > It's impossible for a human reading a book written in an exotic foreign > language, so you are going too far. It's like cracking a Rijndael > encrypted file with a 1000000000000000-bit key size, but worse. Infinite > possible interpretations. > > John G. Rose wrote: > > This is how I "envision" it when a "text only" AGI is fed its world > view as > > text only. The more text it processes the more a spatial physical > system > > would emerge internally assuming it is fed text that describes > physical > > things. Certain relational laws are common across text that describe > and > > reflect the physical world as we know it. There might be limits to > what the > > AGI could construct from this information alone but basic Newtonian > physics > > systems could be constructed. If you fed it more advanced physics > textbooks > > it should be able to construct Newtonian+ systems - branch out from > the > > basics. It's "handles" to the physical world would be text based or > > internally constructed representational entities, which BTW would be > text > > based i.e. numerical representations in base 256 or base n, binary in > > physical memory. Theoretically it could construct bitmap visual > scenes, or > > estimate what they would look like if it was told to "show" what > visual > > imagery would look like to someone with eyes. It could figure out what > color > > is, shading, textures, and ultimately 3D space with motion - depending > on > > the AGI algorithms programmed into it that is... But if it was not fed > > enough text containing physical interrelationships its physics and > projected > > bitmaps would be distorted. There would have to be enough information > in the > > text or it would have to be smart enough to derive from minimal > information > > for it to be accurate. > > > > Now naturally it might be better to ground it from the get-go with > spatial > > physics but for development and testing purposes having it figure that > out > > would be challenging to build. > > ----- 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=8660244&id_secret=52942046-4e4138
