Linas, thank you for your precise and profound explanations!

I mean, I guess you could layer lisp or scheme on to of atomese, but it 
> would be painfully slow, with the current evaluator, and would be quite 
> bloated, since atoms are fat.  Atoms get indexed in the atomspace, and that 
> ends up being costly in both cpu and ram.
>
> We index them because we want to use the atomspace for KR... which 
> conflicts with using them only for performing calculations.  Its a 
> balancing act, its hard to figure out where the imbalance is.
>
 
As far as I understand, what is going on here at OpenCog, an Atom is the 
most universal thing in the universe - able to represent  "all that is the 
case" - how Witti would say.

Universality is always in contradiction to performance. One can not balance 
this.
I think a step to overcome this is to compile certain types of atoms at run 
time to something optimized for performance and than recompile the results 
back to regular atoms.

Humans do this - why an AGI should not do the same?

Maybe especially at your main topic - link grammar. 
Somewhere I read your complaints, how slow it became when you ported it to 
the atomspace.

My thoughts about this was that there should be a possibility to transform 
a given  text corpus to a list of integers, where every int represents a 
word or sign, operate on this list and bring back the results to the atom 
space.  

Humans do this. Remember a situation in which you try to understand 
something new. You read about it, you draw diagrams etc. until you have the 
feeling that you completely understood. 

In case of words - if it is possible to transform a text corpus to 32bit 
ints or even  16bit ints,  it or chunks of it, could be put into the 
cpu-cache and could be worked on very fast.

My feeling is that this could close the gap between performance and 
universality: agents that transform formats at run time and specialized 
code for special tasks - in many  cases even GPUs could be involved.

NL-comprehension is crucial for AGI. You're going to have to face that - no 
escape ;) 

An AGI needs narrow AI-agents. A human genius is often someone who is able 
to perform narrow AI tasks.

--Andi


 

-- 
You received this message because you are subscribed to the Google Groups 
"opencog" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/opencog/697a6971-e1a8-40ff-9dcb-18572fb961e8%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to