Hi, Thanks for the thoughtful reply. ... but its also clear that we have very different concerns and goals.
All the bullet-points you make are entirely valid for non-AI systems architected for commercial use. They're kind of useless if the goal is to do AI. The only thing I'm really interested in is the AI. Nothing in your list is useful or helpful for that. So, to pick an example: you mentioned neo4j-nlp. Cute, ... sure. Read through the "feature matrix" they post there. Zero useful features. I mean, great, if you've got a specific commercial product in mind, but useless for AI. Arguably more-than-useless; its a distraction, a time-waste. And perhaps this is why opencog remains in the back-waters -- no one is trying to do anything like this -- they're pursuing other goals. Which - again, that's nice if that's what you want to do, but for me, I'm interested in the science of creating a thinking machine that can actually think. There are approximately zero tools out there for this. I went on a hunt for such a decade ago: https://linas.org/agi.html and found .. zero tools. Here I am, ten years later, haven't updated the list, because there are still ... zero useful tools. So opencog is my best attempt to cobble together something non-useless. Progress is slow. -- Linas On Tue, Nov 6, 2018 at 5:41 PM Amirouche Boubekki < amirouche.boube...@gmail.com> wrote: > I will respond in a single mail to all the topics you raised in the other > emails. > > a) thanks for the kind words, even if I could implement tinkerpop4 in c++ > and all the requirements listed in the README, I would not do it, because I > am sure the future is not c++ (nor Java). > > b) from my opinionated point of view, atomspace is not the kind of > database I want to work with to develop an intelligent system that should > handle the mixed workload of both analytics and user-facing features. The > requirements to load everything into ram is a no-go for me. What do you do > when there is no more RAM? I mean you can have 12TB or 100TB maybe but you > can not store commoncrawl <http://commoncrawl.org/>. > > c) Like I said previously, I think what you need is foundationdb > <http://foundationdb.org/> > > d) tinkerpop4 was surprising because they mention several times RDF which > is the formalism I chose to work with. > > e) I think it's not a good idea to mix ACL with data, you end up with a > very confusing machinery and ACL is a very complex topic. No modern > database embed ACL machinery and leave that to upper layers. > > f) triple store for the win > > g) gremlin is what inspired my work, I will look into tinkerpop4 closely > even if I don't buy the compiler thing, yet. Basically, I don't believe the > gremlin runtime/compiler can automagically guess what indices to use before > knowing about the queries. Just take a simple TF-IDF or BM25 full-text > search, every time you change the formula you need to update the indices. > This is not feasible to know upfront all the indices you will need, that is > why I recommend a flexible storage engine like foundationdb. Where you can > build a triple store (and graph or hypergraph) and go to the lower level > key-value storage to speed up specific queries. Even if, you index all the > things, like I do and like RDF data store do, you need at some point a > clever trick to handle sorted range scan (everything between 42 and 1337 in > order) or more complex queries like geospatial search or 2d + time. Also, > AFAIK, there is no way to index segments (to answer queries like segment > overlap with segment e.g. "give all events that happened between 2017 and > 2018) in a key-value store without relying on an out-of-key-value-store > data structure which means it breaks transactions semantic and you end up > with some part of the data that is eventually consistent... Maybe > PostgreSQL can help in this regard. > > h) gremlin gives false hope (at least from my readings) that will handle > all the graph algorithms you need. You can not implement an optimal A* > algorithm or shortest path only in gremlin and even if you could the > gremlin engine AFAIK can not stream results. Gremlin is mostly for > neighborhood search. For deep traversal that must be guided, you must issue > several (gremlin) small queries ie. there is no point of a gremlin compiler. > > i) being able to stream results is very important because in a real-world > scenario, where the computation of a solution might take an infinite amount > of time. It must be possible for the user to ask for partial results and > input new data to narrow the currently occurring search maybe that's what > https://en.wikipedia.org/wiki/Beam_search is. Interactive / > conversational (re)search. > > j) https://github.com/graphaware/neo4j-nlp that can be of interest. > > k) again regarding, gremlin, do they plan to handle bigger than RAM > results? I am not aware of such work. AFAIK the result must be able to stay > in RAM. > > All the best! > > > Le ven. 2 nov. 2018 à 23:07, Linas Vepstas <linasveps...@gmail.com> a > écrit : > >> I skimmed it in 30 seconds. We should probably/almost certainly build a >> tinker-pop style interface onto the atomspace. I think its very very >> doable, I think many valuable lessons could be learned. It would make the >> atomspace better. It would be an excellent project for a strong developer >> to morph into a master developer. >> >> At the moment, I have no clue at all what sort of interesting algos such >> an API allows. What could we do, that we can't already do? What would be >> the killer app that makes it important to have this API? A skim of the >> paper doesn't say. >> >> Let me rephrase: the atomspace does not yet have a tinkerpop-like API, >> because nothing we've done so far needed this API. Is this because we >> haven't been imaginative enough? Too busy doing other things? What really >> cool thing could we build with this? >> >> Amirouche -- this means you. Care to take a stab at actually implementing >> this? Care to recruit and convince someone else to do this? Or at least, >> for starters, care to take a stab at answering the earlier questions? >> Cause its always worth knowing why something should be done, before going >> too far down the road of doing it... >> >> --linas >> >> >> On Fri, Nov 2, 2018 at 3:20 PM Amirouche Boubekki < >> amirouche.boube...@gmail.com> wrote: >> >>> https://zenodo.org/record/1476234/files/forth-kind.pdf?download=1 >>> >>> -- >>> 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 opencog+unsubscr...@googlegroups.com. >>> To post to this group, send email to opencog@googlegroups.com. >>> 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/09200c55-4f0d-4197-8835-a21859a11442%40googlegroups.com >>> <https://groups.google.com/d/msgid/opencog/09200c55-4f0d-4197-8835-a21859a11442%40googlegroups.com?utm_medium=email&utm_source=footer> >>> . >>> For more options, visit https://groups.google.com/d/optout. >>> >> >> >> -- >> cassette tapes - analog TV - film cameras - you >> >> -- >> 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 opencog+unsubscr...@googlegroups.com. >> To post to this group, send email to opencog@googlegroups.com. >> 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/CAHrUA343VopPUYX%3DE-yedCO5HvOdO-0xLmGhnBa18RXt%3DERwHg%40mail.gmail.com >> <https://groups.google.com/d/msgid/opencog/CAHrUA343VopPUYX%3DE-yedCO5HvOdO-0xLmGhnBa18RXt%3DERwHg%40mail.gmail.com?utm_medium=email&utm_source=footer> >> . >> For more options, visit https://groups.google.com/d/optout. >> > -- > 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 opencog+unsubscr...@googlegroups.com. > To post to this group, send email to opencog@googlegroups.com. > 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/CAL7_Mo9V0mt3jX1q6tuv7zL_Kc8TgaN5myt62UDCMDH_B5KZMw%40mail.gmail.com > <https://groups.google.com/d/msgid/opencog/CAL7_Mo9V0mt3jX1q6tuv7zL_Kc8TgaN5myt62UDCMDH_B5KZMw%40mail.gmail.com?utm_medium=email&utm_source=footer> > . > For more options, visit https://groups.google.com/d/optout. > -- cassette tapes - analog TV - film cameras - you -- 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 opencog+unsubscr...@googlegroups.com. To post to this group, send email to opencog@googlegroups.com. 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/CAHrUA37BquK14qagUYMvR40EDsm2ueJ4cJbS_wkeQ9AzXBO4ig%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
