Hi Alex, I think Ivan Vodisek has already given much of the correct answer, but I will amplify and say, yes its correct. Some important inline comments below.
On Sun, Oct 1, 2017 at 6:20 AM, Alex <[email protected]> wrote: > Hi! > > I have heard that AtomSpace is big hypergraph and the names of the > representation elements - nodes and links - suggest exactly that. > > So - knowledge can be represented as nodes and links, i.e. as hypergraph - > let call it the "Type1" labelled hypergraph - semantic graph. > > But I guess that the same AtomSpace/OpenCog knowledge can be represented > as mathematical formulas, e.g. as term logic formulas. Those formulas can > have attributes that correspond to the probabilities but they are formulas > anyway. Each formula is the word or sentence in some formal language and > therefor it has syntactic graph representation as well. Let call it the > "Type2" labelled hypergraph - syntactic graph. > > The question is - what is the connection between Type1 (Semantic) and > Type2 (syntactic) graphs? I guess, there can be established semantic rules > that allow on to construct Type1 graph from the Type2 graphs and back. > What is the right type for the representation? I guess, Type2 graphs are > more appropriate for the formal reasoning (e.g. sequent calculus). > They're both graphs. You answer your own question by saying MMT below, so I don't understand why you asked that question. > > I have this question because there is this formalism - MMT > https://uniformal.github.io/ - Meta-Meta-Theory that tries to unify all > the (in)formal knowledge in one foundation free framework. There are wealth > of literature how Florian Rabe with his collaborators try to encode every > big formalism (Axiomatic Set Theory, Constructive Type Theory (Coq > culture), Higher Order Logic (Isabelle/HOL culture) in one modular > language. So - result can be the formalism, that allow to express every > formula of every formalism in one common language and - of course - that > means, that every formula can have Type2 graph assigned to it! And that > means that we can encode in one graph database all the possible formulas > (all the possible non-multimedia knowledge). > I have never heard of MMT before, but I believe that the opencog representation is probably going to be very similar to what MMT does. It would be an excellent exercise for you (or for someone) to compare the two, and see where they differ, how they differ. I would very happily take the best ideas fro MMT and put them in opencog, or find some way to collaborate with the MMT community -- I think we are working on the same general ideas. > MMT is largely completed work, so - there remains the technical work only > - one can take the best open source graph database (JanusGraphs is the > best) > This is not enough. The MMT landing page already lists several dozen things that JanusGraphs does not do. Likewise, I have not looked at JanusGraph carefully, but I'm certain that opencog does many things it does not do. Ou focus is NOT to be "just" a graph storage system, but a graph storage system with many additional services (the MMT page lists many of these) Our big ones are: * the pattern matcher * the pattern miner * the rule engine Our smaller ones are: * a sparse matrix subsystem * a parsing (categorial grammar) subsystem The atomspace does have a postgres backend, and perhaps it would be excellent to add a JanusGraph backend. I don't have time to do this myself; we need programmer volunteers to do this. > and encode this knowledge and attain the most universal knowledge base > possible, that is certainly more expressible than OpenCog (that currently > uses (probabilistic) term logic). > False statement; you misunderstand opencog, or are envisioning it incorrectly. It is probably safe to say that atomese is mor advanced than Janusgraph+MMT today, but I might be wrong, because I never heard of these two before today. I might be wrong, but I doubt it. > > One should add that each formula syntactic graph (Type2 graph) can have > associated semantic graph representations (Type1 graphs) - there can be > more semantic representations for the one syntactic one. Sadly, this > relationship between syntactic-semantic graphs is very little researched > field. There are, of course, research about semantic graphs themselves > (every knowledge representation with graphs do this), but about connection > between syntactic and semantic graphs there is only one work of which I am > aware of: http://homepages.inf.ed.ac.uk/ldixon/papers/dixon-camcad-09.pdf > - about logical graphs. > Perhaps we need more tutorials and wiki pages about how to do this. We've been doing logical graphs like what Dixon discusses since about forever. It's all there in atomese, and its all "old hat", old news to us. > > So - we can have knowledge system that have the best from the both worlds: > - The knowledge representation forma is taken from the MMT > - The knowledge representation techniqye is taken from JanusGraph - the > best that the industry can provide. > I am pretty sure that atomese today already meets these goals, and goes far beyond this proposal. I might be wrong, but I doubt it. > > That can be the future of knowledge representation systems. Sometimes I am > very, very suscipicous about efforts of building custom knowledge bases. > There are necessary so many resources to implement technicalities that I > can not believe that custom knowledge base can compete with universal, > industrial quality graph database. My guess is - if industrial graph > databases had been around at the time of inception of cognitive > architectures (Soar, Clarion, Cog, etc.) the all the cognitive architecture > would be built around/using the industrial quality graph databases. > Atomese already has a way of creating a JanusGraph plugin. Perhaps JanusGraph has some features that our current plugin API does not support -- and for this, it would be excellent to expand/improve our plugin API. It would be interesting for e to get into that level of detail. > > So - one graph database can host both types of graphs - both syntactic and > semantic graphs and also this graph database can host reasoning and > self-development procedures (which are programs, which can be represented > as syntactic trees and saved in the same graph database as the remaining > knowledge) for self-(re)evolution. So - big, big self-evolving system or > hypergraphs that lives in the industrial grade graph database. Maybe this > can be the start of AGI? > Yes, that's part of the vision of opencog. > I call my system MOC - MetaOmegaCog (MetaOmega stands for > meta-meta-meta...) > > What are your thoughts about such plans? > Come study atomese more carefully, and perhaps we can talk about how to do this. --linas > > > -- > 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/c436ad44-7d70-4af2-a3b9-cae81d6d2788%40googlegroups.com > <https://groups.google.com/d/msgid/opencog/c436ad44-7d70-4af2-a3b9-cae81d6d2788%40googlegroups.com?utm_medium=email&utm_source=footer> > . > For more options, visit https://groups.google.com/d/optout. > -- *"The problem is not that artificial intelligence will get too smart and take over the world," computer scientist Pedro Domingos writes, "the problem is that it's too stupid and already has." * -- 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/CAHrUA36WeFWO3npsCUYfiQp6jocHCG41bbqhbqYCm2pJKYyO1Q%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
