On Sun, Apr 26, 2020 at 1:30 PM Jack Park <[email protected]> wrote:
> I'm trying hard not to be an advocate of anything; rather, I'm in > exploration mode. > > Tell you what: I'd like to see (have not yet found) some solid examples of > AtomSpace - full-on knowledge graph implementations, whatever. Something to > be able to follow through one or two complete examples which are more than > toy exercises. > The latest blog entry? https://blog.opencog.org/ You're not going to find "complete examples" anywhere -- if you can't master the toy examples, a "complete example" will not make anything easier. That said -- this all is open source, so --- The mozi biochem annotation code -- https://github.com/MOZI-AI/annotation-scheme It handles datasets with some 10-50 million protein and gene encodings. You would have to write to the developers to get sample datasets, I'm not sure if they are proprietary or what. There used to be the old Hanson Robotics behavior code; it has fallen away but it had a bunch of scripts for reacting via facial expressions, to visual cues (people entering the room, making hand-gestures). It implemented something called "behavior trees" (see wikipedia) -- what's left of it is here https://github.com/opencog/opencog/tree/master/opencog/eva but I'm not sure where the ROS integration has moved to. It used to be here: https://github.com/opencog/ros-behavior-scripting -- the repo is still there, but the contents were gutted and moved somewhere else, not sure where. Maybe some hanson robotics tree. The robot control stuff was here: https://github.com/opencog/blender_api -- with some effort, the original pipeline can be made to work again. Somewhere there's an AIML import module: you can import everything that AIML does, and run it in a compatibility mode. The goal of doing this was to also attach vision and motor control. That never got far, because AIML sucked. AIML had maybe 50K or 100K stimulous-response pairs, so that was maybe under 1M atoms, not sure. Here: https://github.com/opencog/opencog/tree/master/opencog/nlp/aiml -- it worked in real-time, enough to get the robot to trade shows and on TV. Then a switch to ChatScript, but chatscript is far more complex than AIML, so they used something called "ghost" -- https://github.com/opencog/opencog/tree/master/opencog/ghost -- the chatscript also has about 20K-ish conversational interactions, gambits, responses, so maybe 200K atoms ?? no clue, actually. Also this: https://github.com/opencog/loving-ai-ghost and this: https://github.com/opencog/ghost_bridge and this: https://github.com/opencog/loving-ai The robot stuff and chatbots are abandoned for a variety of reasons. Partly because chatbots suck and partly because there's no money. But also because chatbots got little/nothing to do with the "hard problem" of AI, so its boring to Ben, and to me, and to most anyone else working on this stuff. Again -- that's why there's a focus on learning. That leaves the genomics/proteomics stuff, which is interesting because you can actually do data mining that other bio systems cannot do. -- Linas > > On Sun, Apr 26, 2020 at 11:04 AM Linas Vepstas <[email protected]> > wrote: > >> About opencrux... below, >> >> On Sun, Apr 26, 2020 at 10:24 AM Jack Park <[email protected]> >> wrote: >> >>> That's a useful observation. I wonder if it has anything to do with the >>> fact that they give appearances more of offering a platform aimed at >>> solving real-world problems as compared to a platform for language modeling >>> research? >>> >>> Their landing page reads rather differently from that of the AtomSpace >>> landing page. They make results-oriented promises; I don't see that on the >>> AtomSpace landing page. >>> >>> Their landing page appears to be the work of skilled marketing types; >>> the AtomSpace landing page appears to be the work of, well, not skilled >>> marketing types. Their top nav bar says Products, Solutions, Use Cases, >>> Community, ...; AtomSpace is a MediaWiki, one very familiar to developers, >>> but not to business oriented people. >>> >>> What if AtomSpace ignored all the cool buzz words and opened with >>> problems it can solve, ways people can start using it right out of the box >>> without building it, tuning it, etc? >>> >>> Would you, as a scientist, be able to live with wall-street-oriented >>> thinkers taking over your pet projects and packaging them up for far >>> less-skilled consumers? >>> >>> I think this is an issue faced by a lot of us. >>> >> >> Quite right. We've never had anyone interested in marketing participate >> in the project. Marketing is a skill, and unlike open source, there is no >> "open marketing", so you have to pay these people actual $$$ to get slick >> content. >> >>> >>> I studied Grakn carefully; the front story is one, for me, of some >>> attraction to the apparent simplicity of the system; the back story, for >>> me, is that Grakn appears optimized in ways which get in the way of >>> allowing me to push it in ways I believe it should be pushed; a problem of >>> ontological commitments I cannot undo, so I just walk away. >>> >>> For a really interesting case for comparison, take a look at >>> https://opencrux.com/ >>> >> >> Well, care to be specific? I suppose it's possible to put a >> prolog/datalog API on top of the atomspace. I'm sufficiently removed from >> that world that I would not want to even begin without a lot of >> arm-twisting, but I can consult. >> >> I'm not sure what to say about other things. Bitemporality? We have a >> concept of a space-time server, optimized for dealing with time .. and >> space coordinates. It's neglected... >> >> Document-graph? Sure, cause why not? Seems easy to me... >> >> Back-ends other than postgres? That's interesting. It's not "hard" - its >> not conceptually hard, and also the infrastructure/API is already in place. >> But it does require some fair amount of slogging. One would have to be >> enthusiastic about it. >> >> Datalog queries? I'm certain that anything you can say in datalog, you >> can say in Atomese, and so its a matter of writing a converter/translator >> that accepts datalog as input and spews atomese as output. How hard is >> this? No clue. You've actually messed with this kind of stuff, you'd know >> better. Again -- everything that is a "symbol" in prolog is an atom in >> atomese. The atomspace is pretty much nothing more than a glorified symbol >> table. >> >> ... and this seems to be the key insight that the opencrux developers >> have made: symbol tables are (ad-hoc, in-RAM, poorly-designed) databases. >> Rip out the ad-hoc symbol table of programming language XYZ, replace it >> with a real, actual database, and wow ... off we go on a wild ride. I'm >> trying to think of what programming languages XYZ this could be the most >> interesting for... where you'd get the most bang-for-the-buck. Maybe prolog >> -- maybe that is the lesson from opencux ? >> >> Care to suggest an easily-hackable version of prolog/datalog on which >> this experiment could be done? Just for the heck of it? >> >> I mean, you could do this trick for python or javascript ... I don't >> think the python community would accept it, it would be too crazy and weird >> for them. The javascript folks might .. but they already have some pretty >> decent infrastructure already, so they don't need something this low-level. >> They've done this integration at a higher level, already. (programming in >> javascipt is far more mind-expanding than python. Python shuts down your >> world-view, narrows your thinking. Blinds you to possibilities. javascript >> does the opposite.) >> >> --linas >> >>> >>> On Sat, Apr 25, 2020 at 7:48 PM Linas Vepstas <[email protected]> >>> wrote: >>> >>>> <snip> >>>> I dislike promoting competitors, but the grakn.ai system has taken >>>> some baby-steps in this general direction. I'm envious that they are far >>>> more popular/funded/suported/used than the atomspace. >>>> >>>> --linas >>>> >>>> -- >>> You received this message because you are subscribed to the Google >>> Groups "link-grammar" group. >>> To unsubscribe from this group and stop receiving emails from it, send >>> an email to [email protected]. >>> To view this discussion on the web visit >>> https://groups.google.com/d/msgid/link-grammar/CAH6s0fydKLSFN40GykJ5FNR0d_DKqPi0yrAyVU%2B4wpFd6aig9Q%40mail.gmail.com >>> <https://groups.google.com/d/msgid/link-grammar/CAH6s0fydKLSFN40GykJ5FNR0d_DKqPi0yrAyVU%2B4wpFd6aig9Q%40mail.gmail.com?utm_medium=email&utm_source=footer> >>> . >>> >> >> >> -- >> cassette tapes - analog TV - film cameras - you >> >> -- >> You received this message because you are subscribed to a topic in the >> Google Groups "link-grammar" group. >> To unsubscribe from this topic, visit >> https://groups.google.com/d/topic/link-grammar/FMGJq5YhbME/unsubscribe. >> To unsubscribe from this group and all its topics, send an email to >> [email protected]. >> To view this discussion on the web visit >> https://groups.google.com/d/msgid/link-grammar/CAHrUA36GQq2B6nrFcQxN7DqQ8gZJH1DqvvaUkBp54rcR2b89Mw%40mail.gmail.com >> <https://groups.google.com/d/msgid/link-grammar/CAHrUA36GQq2B6nrFcQxN7DqQ8gZJH1DqvvaUkBp54rcR2b89Mw%40mail.gmail.com?utm_medium=email&utm_source=footer> >> . >> > -- > You received this message because you are subscribed to the Google Groups > "link-grammar" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To view this discussion on the web visit > https://groups.google.com/d/msgid/link-grammar/CAH6s0fxp6nFOVBvBLuDkW-i0nazm2FwgucGe5DHs_Enur83Gcw%40mail.gmail.com > <https://groups.google.com/d/msgid/link-grammar/CAH6s0fxp6nFOVBvBLuDkW-i0nazm2FwgucGe5DHs_Enur83Gcw%40mail.gmail.com?utm_medium=email&utm_source=footer> > . > -- 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 [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAHrUA36dnu2voXYUizOf83VJjju88MormL-eDe-bFQrrxuAP0A%40mail.gmail.com.
