I have it all figured out... Almost.  I do not have AGI complexity all figured 
out.  But I do believe that I can reduce the complexity enough to create an AGI 
program that can demonstrate minimal competency that goes beyond any genuine 
AGI program that is around today. But I might be wrong.  I won't know until I 
try it.  It should take me a year to run my experiments. Now let's say that a 
year from now I am saying the same sort of thing.  I have it all figured out, I 
have really learned a lot in the last year but AGI programming is really 
complicated and I haven't been able to get the kind of results I had been 
expecting.  Well, that will be a negative indicator but researchers should be 
given some leeway.  That kind of leeway should be accorded to part-time amateur 
researchers as well.  (It should especially be granted to part time amateur 
researchers.) But on the other hand I should be able to report something about 
the negative results that I got.  I shouldn't carry on with hubris and 
hyperbole when faced with an outstanding setback.  Now if 2 years goes by and I 
am still saying the same thing then my credibility should start declining.  The 
people who are claiming that my credibility should be going down now do not get 
how science works.  But a researcher has to demonstrate enough integrity to 
report honestly on negative results. Minimal competency in AGI means that the 
program will be able to demonstrate some genuine learning and it will be able 
to embark on unique paths.  The only way you could get two copies of the same 
program to be exactly the same is to duplicate the hardware and run time 
conditions almost exactly.  Also, if the program worked then it could be 
adapted to other IO modalities in a fairly short time. I am going to write a 
text-based AGI program.  No, it won't have deep knowledge into what things look 
like.  However I believe that the basis of intelligence is not directly 
dependent on multiple IO modalities although I do have some opinions about the 
variations of data-events that need to appear in the IO Data Environment. Is 
there going to be a problem in minimal competency learning because the 
text-based AGI program won't have any iconic grounding?  No.  The program will 
have text-based grounding, information about relevant concepts that can be 
found in other statements.  The fact that the grounding problem can be 
abstracted (and reduced) down to a question about packets of data should be 
enough to get you to think outside the box on this one.  The  iconic grounding 
theory was originally proposed as a possible solution to AI / AGI nearly 30 
years ago but in the time since it was first proposed there has not been any 
outstanding evidence that it represented a viable solution to the problem. What 
about the basic question?  When we think about something a little novel we 
usually refer to a visual or other sensory experiences to make a determination 
about it.  Well that is a reasonable question to ask but it is not a profound 
one.  There are people who were born blind so right away you should be a little 
wary of asserting the necessity of the visual confirmation in general thought.  
Secondly, we do not use that method for every kind of novelty that we think of. 
So this does leave me with one interesting question. Does the knowledge about 
every object in a text-based AGI program have to be associated with visual 
descriptors?  That could prove quite inconvenient.  The answer is no.  The 
worse case is that a concept might have to be designated as a material object 
or as a material object of fantasy.  That would allow the program to infer that 
it had some visual features.  So if someone introduced the idea that  box had a 
distinctive (kind of) shape for the first time, the new synthesis would not 
cause the program to start smoking from the iconic fantasy of the skeptics' 
childhood. Jim Bromer                                           


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to