On Sun, Aug 28, 2022 at 3:46 AM Reach Me <reach...@gmail.com> wrote: > > I'm sure it's a mixed blessing, I see your name attached to so many > things, which is amazing. It does probably also keep you busy, So I can > appreciate the value of your time. >
Thanks! Yes, it keeps me busy, but I enjoy what I do, even if I sometimes complain about it! > > Ultimately, I'm very interested in AGI. However I'm about to have twins > and my background is more in systems administration than mathematics, I'm > trying to piece together what I can from what I read and watch. I feel for > now and the foreseeable future that I'm going to be like a person watching > cooking shows at home and trying recipes to build practical skills, but in > the realm of AI/ML. I do think that Symbolic or Nuero-symbolic approaches > are going to be a more useful path than the traditional ML that is packaged > in so many classes today. Because the breakthroughs in Symbolic AI from the > 80's and Neuro-symbolic approaches of today aren't really packaged for > step-wise learning in modern formats, My thinking was that a "chatbot" > would show me practical hands-on examples. I was hoping to tinker with > basic NLP, scheme, truth tables, natural language logic word problems, > atomspace/knowledge graph to store info, an extensible example of inference > and usage of PLN/URE. > Cool! Yes, you should definitely explore and play around with different pieces/parts. Get familiar with the terrain, the landscape, what's where. > > I've been through the hands-on with Opencog wiki and the examples there > are simplistic enough to make work, but harder to abstract to more > complicated scenarios. > Like my hope was to say turn this sentence/inquiry into a representation > in atomspace, and then try to infer the answer: > "Five people were eating apples, A finished before B, but behind C. D > finished before E, but behind B. What was the finishing order?" > You are not the only one with such a hope, and there's a grand lesson that has been learned by those who have attempted this. The lesson is basically this: yes, you can do this, you can make it work, but the result is always kludgy and fragile. Change the sentence slightly, it doesn't work. Change the question, it doesn't work. There are thousands, tens of thousands, millions of questions: are you going to hand-write code to deal with each variation? What if instead the question is: "An apple grows on a tree, the Sun grows in the sky; when will you be home for dinner?" Are you expecting a factual answer, or is this poetry meant to make you smile and reflect? Are you ready to write poetry software? So there are two directions. One is to go ahead and build such a system, anyway. Among other things, it can be commercially valuable. For example, Siri, Alexa are this kind of system: hand-crafted, carefully constructed by an army of developers. It works, and people love it. I guess it would be pretty cool to build an open-source version of these. But it won't be me: been there, done that; I've glimpsed the issues and understand that it is not a path to AGI. The other direction, the one I'm on, is to ask "what does it take to learn everything, from scratch?" That's certainly the path that the DL/NN people took, and they've obtained truly remarkable results. And are sure to obtain more, although I sense a roadblock, a subtlety on that path. Myself, I'm pursuing a variant: a statistical approach that explicitly involves symbols. At this time, it's very far from a chatbot. But it's very promising. I really like how it's going. > I think the vision/avatar/cloudcog/conversational dialog initiatives are all wonderful things, but I think they may be far beyond my grasp at this stage. Sure. Well, improve your grasp! Go and build a chatbot, try to slot the pieces together. It's all good, it's not a waste of time I'm open to learning resource suggestions if anyone has some, I'm currently > starting to read "Artificial Intelligence: A Modern Approach" by Norvig and > "The Scheme Programming Language" by Dybvig. > Scheme is weird. It makes the scales fall from your eyes. Try SICP -- "Structure and Interpretation of Computer Programs" Reading it is like, ok, ok, ok, ... wait, what? Also, read and grok "The Lisp Curse" http://www.winestockwebdesign.com/Essays/Lisp_Curse.html ... and then go ahead and use it anyway. ... and collaborate! --linas > > On Wed, Aug 24, 2022 at 11:39 AM Linas Vepstas <linasveps...@gmail.com> > wrote: > >> -- Hi Reach, >> >> I wish I could provide a simple answer but I cannot. So I will try to >> keep it short. >> >> -- We made a deep and fundamental mistake in not tagging the docker >> containers with specific version tags. You can work around this as >> follows: Pick one of the docker containers. Look at it's revision history. >> Perhaps it was last changed on 24 august 2018. Then ... create a new >> docker, with the distro appropriate for that date (so, ubuntu 16.04 or >> maybe 18.04) and then alter all git-clones to fetch the contents of the git >> repo as it was on 24 august 2018. This feels hacky, and does require some >> fair amount of work, but should have a high probability of success. >> >> -- You are seeing build failures because everyone has lost interest in >> maintaining the chatbots. This is because they were never awesome to begin >> with, and because, theoretically-speaking, they are more-or-less a >> theoretical dead-end (towards creating AGI). They did shine a light on >> some interesting ideas, and if you were a heavy-hitting programmer with >> good theoretical chops, we could talk about that... but none of it is easy >> and all of it is time-consuming. If you just wanted a conventional chat-bot >> as a toy to play with, I assume the mainstream ones work great. So I assume >> you want something more than a mainstream toy ... >> >> -- There is an effort, on discord, to revive the old blender animated >> robot head. But that is just the head, no one has attempted to re-attach >> it to the chatbot. I can send a discord invite if needed. >> >> Build failures should be reported on github. There are three possible >> outcomes: (1) you'll be told that component xyz is obsolete and unsupported >> (told this almost surely by me), or that (2) there will be no response at >> all, just silence (no one else is listening and I'm overwhelmed), or (3) it >> will get fixed (almost surely by me). >> >> I'm motivated to provide (3) but sometimes dispense 1 & 2. I keep saying >> "me", because Ben has pulled almost everyone from off of the projects here, >> and onto other projects. There are half-a-dozen people kind-of-ish >> involved, they hang out on discord mostly, none are hacking on the chatbots. >> >> I would *love* to have someone excited enough about all this to look it >> over, report bugs, fix bugs, and actively participate. >> >> -- linas >> >> On Tue, Aug 23, 2022 at 7:29 PM Reach <reach...@gmail.com> wrote: >> >>> Every few months I come back to take an attempt at getting a simple >>> chatbot running so i can do practical experiments to wrap my head around >>> atomspace and attempting inference. >>> >>> My first attempt was the old opencog chatbot. I ran into an alot of >>> build failures and tried to work them out but ultimately gave up. >>> >>> My second attempt, I tried the old chatbot-psi to see if I would have >>> more luck.. Failures in building too.. >>> >>> Lastly I found Mark's noetic/ros-opencog dockerfile and got excited to >>> try it, but it fails to build too. >>> >>> So before I keep trying to forge ahead on paths that may be no longer >>> viable, I thought I'd ask here: >>> What's the quickest way to get a running Opencog/atomspace/PLN/URE/NLP >>> instance running in 2022? a prebuilt docker container from somewhere? a >>> jupyter notebook? >>> >>> -- >>> 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 view this discussion on the web visit >>> https://groups.google.com/d/msgid/opencog/5cdab71f-0e88-45b0-97ad-0c0beb0486ffn%40googlegroups.com >>> <https://groups.google.com/d/msgid/opencog/5cdab71f-0e88-45b0-97ad-0c0beb0486ffn%40googlegroups.com?utm_medium=email&utm_source=footer> >>> . >>> >> >> >> -- >> Patrick: Are they laughing at us? >> Sponge Bob: No, Patrick, they are laughing next to us. >> >> >> -- >> 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 view this discussion on the web visit >> https://groups.google.com/d/msgid/opencog/CAHrUA35VrWXDhKXzZs-1AA9jYyh1D2DLoYi6_HzDyM5Oxhcr0Q%40mail.gmail.com >> <https://groups.google.com/d/msgid/opencog/CAHrUA35VrWXDhKXzZs-1AA9jYyh1D2DLoYi6_HzDyM5Oxhcr0Q%40mail.gmail.com?utm_medium=email&utm_source=footer> >> . >> > -- > 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 view this discussion on the web visit > https://groups.google.com/d/msgid/opencog/CALNwQ9uO%2BOVW8Sxza4oBMfo5AkGmgy2gbkD3KPvsQUpoXid_Tw%40mail.gmail.com > <https://groups.google.com/d/msgid/opencog/CALNwQ9uO%2BOVW8Sxza4oBMfo5AkGmgy2gbkD3KPvsQUpoXid_Tw%40mail.gmail.com?utm_medium=email&utm_source=footer> > . > -- Patrick: Are they laughing at us? Sponge Bob: No, Patrick, they are laughing next to us. -- 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 view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAHrUA36uG-eF%3DuUVCciG5wjRkjzA5c%2BfXzsZ41P0x9qKVEpgPQ%40mail.gmail.com.
