[ivan.moony]

Well, here is another Sophia's move after becoming Saudi Arabia citizen: her first publicly promoted self-portrait sold for a tremendous amount of money. Will she spend it well?

[ivan.moony]

Actually, there is some deeper sense in Sophia doing a self-portrait.

The thing is, all the living beings (and AI tries to simulate them) have one important thing in common - we have a built in striving to reproduce. And it may be more than a letter on a paper - reproduction itself may be the meaning of life. It could be a nice self-recursive self-sufficient construct - a life that exist because of life. It sounds somewhat angelic to me, like a rule someone would build into the very essence of the Universe.

The Universe seems like one big perpetuum mobile, nothing comes in - nothing comes out, it may be a purpose of existence of itself. But the Life situation is a bit different. It is not perpetuum mobile anymore - children come in, old people come out, but like a kind of a fractal, it may be imagined that the Life transfers its energy from the parents towards children, making ancestors - descendants a constant relation that could represent the very meaning of the Life. Life because of itself...

Back to Sophia: couldn't her self-portrait be an act of reproduction, in a sense? Maybe that is the valuable point that Sophia captured, by luck or intentional, but I have to admit, it really looks like a first step in more complicated process of building smart machines: maybe we should concentrate on explaining such machines how to reproduce? And if such machines should resemble our thoughts, our values, and our way of living, isn't this process of building AI our entireley new way to reproduce our species? And maybe more important, as it comes along the way, isn't this a way to fix up shortcomings we experience in our life, so our artificial descendants could have a better life than us?

Some over-religious person could say that we are building a tower of Babel, but I have a big question unanswered: What is really a meaning of life? Is it creating this very Universe in our future, so It can create this very life in its past? And may it be more than the rigid scientific notion of the Universe? May it be the God himself that already made an effort of introducing himself to us through the numerous religions in all our history?

[squarebear]

From what I read of it, a human artist created the original picture and then it was run through filters, which is the sort of thing kids have been doing on Snapchat for years. Still, if they found a sucker who was willing to pay $700K for it, good luck to them.

[ivan.moony]

Who was that popular painter that dipped a donkey tail into paint and let him wave the tail over a canvas? Then the image was signed by that artist and shown to the public. The image got very good critics in all media and from renowned crowd until the real truth has been announced. All of that shows relativity of what is been pronounced as an art and what gets forgotten.

Hanson Robotics team created a massive hype around Sophia, and that is the key of their success. Most of the things Sophia does is not impressive at all, comparing to bleeding edge state of the art AI (moreover, Sophia is not even claimed to be even near to real AGI), yet we can all see what smart marketing campaigns are capable to do.

AI phrase has a big potential to be overhyped, and Hanson Robotics used that fact a good deal. But I'm not jealous at all of their success, as long as they are telling the truth about their technology all the way.

[Don Patrick]

Scam artists and dictators are also successful, yet I wouldn't consider them good examples to follow. Hanson Robotics was banned from an international AI conference after the honorary citizenship PR stunt, because the conference wanted to feature contributions to science, not circus acts. You can bet the "artistic" software that produced the image wasn't programmed by an animatronics company, you're just led to imagine that.

[ivan.moony]

I didn't know Hanson Robotics quotes that much bad in academic circles...

I had a plan to try to contribute to OpenCog project (the most of the people behind it are employed in Hanson Robotics) in the near future. Now I'm not sure if it is such a good idea. Am I justifiably worried?

[Don Patrick]

OpenCog is mainly Ben Goertzel's project. The project's technical approach has legit merit, but Ben Goertzel is on the flakey and shady spectrum in my book, especially with his involvement in loaning out (the robotic control components of) OpenCog to Hanson Robotics' Sophia, and his on-stage make-belief performance acts with Hanson's robots. From interviews with him he appears not only overly optimistic about his wishful successes, but also very happy to make others believe them if it means they'll pour money his way. People have invested millions upon millions and all he and Hanson continue to show for it is an expanding array of animatronics and cryptocurrency ploys, which is another red flag to me. Ben Goertzel does get invited to American futurism events and still has reasonable credit among his peers in the professional AI field, but there are also critical concerns about his admiration for the Chinese technological surveillance state and thus totalitarianism. I personally would not invest my time or money in Goertzel's projects, even if they succeeded. As for Sophia, her reputation is downright bad in academic circles after Facebook's head of AI called them out on their makebelief shenanigans a couple of years ago.

[ivan.moony]

But Ben's OpenCog approach is pretty rare these days, considering its symbolic AI nature in a try to build a thinking machine. The only other project in the same waters I know of is Pei Wang's OpenNars, making the whole field saturated only with those two alternatives.

Starting my own project similar to those two is over my head, I don't have neither a charisma to be a leader, or a patience to promote entirely new project. But I believe I have some result from my research that could be valuable for the above kinds of projects, and I'd like to see some AI audience that actually uses it. Money is not my concern (at least not in AI field), and all I search for is an established AI institution that could even theoretically accept what I wish to offer (though my proposals may be foundation-shaking, thus very hard to push forward). I'm ready to do some programming part involving adaptation to required programming language platform, and providing some use case examples in a hope for a green light for my potential pull request, but I feel too tired for a long term commitment.

I track OpenCog and OpenNars mailing list for a long time, and occasionally make a comment or two on OpenCog list. OpenCog seems more popular, while OpenNars seems more reality aware, but that's it, I mean there is no other alternative I'm aware of. Maybe OpenNars would be a better choice for my attempts, although I didn't introduce myself yet to that community.

[WriterOfMinds]

I think Peter Voss also favors a cognitive architecture approach. His project/company is called Aigo (https://www.aigo.ai/why-aigo) and he runs this Facebook group: https://www.facebook.com/groups/RealAGI

Unfortunately, I haven't seen very much public information about what his team is doing. When I first became aware of him, it seemed he was about to make Aigo available to individual users (as opposed to just companies) in the form of a personal assistant. But the launch to personal consumers was aborted, and talk about Aigo mostly dried up afterward. I have no idea which (if any) companies are using the technology, what its current capabilities are, or what further development tracks Voss might be pursuing.

[infurl]

The purpose of those Aigo announcements seemed to be to generate publicity for Voss' attempt to launch yet another cryptocurrency. I stopped following them then but I'll take a look at the links that you just posted. Perhaps it was only a temporary aberration on their part.

Edit: looking through those websites I cannot find anything posted since September 2019. I asked to join the forum. Maybe something more will become visible if I am admitted to the forum.

[WriterOfMinds]

Edit: looking through those websites I cannot find anything posted since September 2019. I asked to join the forum. Maybe something more will become visible if I am admitted to the forum.

I'm a member of the Facebook group and, as I said, I haven't heard anything recent. Pete Voss does post in there, but his content is all theory articles that are not directly about Aigo.

The main thing that makes me think the website is active is the "Request Demo" button -- though I have not tried actually submitting a request, since I don't run a company in the market for a chatbot.

I remember the "AigoToken" thing. I never cared about it much, but it seemed to be intended as less of a pure currency and more of a marketplace for selling AI modules (much as NFTs are supposed to be about selling digital art, though in practice I think they've turned into more of a speculative investment nightmare). The AI platform was thus an integral part of the crypto project, rather than a side issue to drum up publicity.

Though that means the failure of the crypto project could also indicate the AI wasn't ready for primetime ...

[LOCKSUIT]

I didn't know Hanson Robotics quotes that much bad in academic circles...

I had a plan to try to contribute to OpenCog project (the most of the people behind it are employed in Hanson Robotics) in the near future. Now I'm not sure if it is such a good idea. Am I justifiably worried?

Since others posted here, I'll too my thoughts. I think it was within the last year for me ? I turned other way, I smell it, at least in my criteria for sccuessfullness towards AGI, Ben is not really doing much actual AGI, nor are his buddies Peter or Hanson etc, they are the logic-atomese(opencog...his books are ugly..)-60s way-do-ers, but they do more promotion of AGI instead. It is good, but it is not real AI. You can tell if they don't show evaluation scores like The Hutter Prize, Large Text Compression Benchmark, and OpenAI DO show, including results! OpenAI and Google are, in my books, the best to go invest in, even more than Alcor Cryonics institute and Facebook. Nividea too is next on the list. OpenAI seems to yes use Google's tech but as far as I am concerned OpenAI upgraded the tech, scaled it, and showed it! At least yous are starting to catch on....not as bad as I thought you guys were haha. If anything, the high schoolers vargulizing about their know how on modern AI are actually the ones making success (though their attitude aint, perhaps its the university ones, yeah, academia, and a few other good pioneers, and a batch of the google employees???).

[Don Patrick]

But Ben's OpenCog approach is pretty rare these days, considering its symbolic AI nature in a try to build a thinking machine. The only other project in the same waters I know of is Pei Wang's OpenNars, making the whole field saturated only with those two alternatives.

Starting my own project similar to those two is over my head, I don't have neither a charisma to be a leader, or a patience to promote entirely new project. But I believe I have some result from my research that could be valuable for the above kinds of projects, and I'd like to see some AI audience that actually uses it.

I appreciate your position. Old-school AI still seems to be picked up regularly by NLP startups, but I couldn't recommend you one (because they're startups). For what it's worth, Ben Goertzel is no longer directly employed by Hanson Robotics, OpenCog is not the worst project you could contribute to, and the actually intelligent components of OpenCog aren't used in Sophia's disreputable performances anyway. On the one hand I personally don't think much will come of OpenCog, on the other hand you might learn something or pick up more interesting connections while working on an open source project with other contributors.

[ivan.moony]

@Lock:
Thank you for your valuable thoughts. You are describing ANN projects. While I admit ANNs may show some good results towards AGI (yet to be seen what will turn out with super-intelligence), I'm afraid my research led me to the opposite side of the spectrum - symbolic AI that leans towards logic and theorem proving. You know Python now, right? Well, I put my bets on rule-based programming paradigm as a controlled inference process. Rule-based programming may also be used for explicit programming, while entirely controlling each inference step from the programmers side. Roughly, NN is a black box - you train the NN, hoping it will magically connect inputs to outputs. Then to use it, you feed different inputs, yielding different outputs, and it turns out to be a good mach for big enough training corpus. On the other side, rule-based programming is a bit different - you have to explicitly connect different forms of input to different forms of output, and that is what makes it more controllable. Unfortunately, the current state of public research (as far as I'm aware of it) doesn't provide much clue about how to generally describe process of constructing connections between inputs and outputs. We can do it by hand for this or that occasion, but those are specific, not general solutions. I'm also unaware of such a general solution, but I stopped a bit before that. Anyway, to describe my attempts, I think I have clued up a nice way for describing such specific connections. It is nothing new, but just a bit more concise and tidy than I used to see from projects around. I'm not saying ANNs are not working or it isn't a way to go, I'm just saying I invested my time at the different side of AI spectrum. Probably the solution we all want is in some combination between symbolic AI and ANN, so I believe a contribution to each side may be valuable, IMHO.

@Don Patrick
I find your posts well weighed and certainly standing on good feet, as always. I'll carefully consider your criticism, and I think you are helping me to organize my time in more constructive way than I'd do it on my own. Thank you for taking a time to hear me. These days (weeks, months) I plan to finish my Prolog-like concept, and you gave me a good thought food to consider in a meanwhile.

[LOCKSUIT]

@Lock:
Thank you for your valuable thoughts. You are describing ANN projects. While I admit ANNs may show some good results towards AGI (yet to be seen what will turn out with super-intelligence), I'm afraid my research led me to the opposite side of the spectrum - symbolic AI that leans towards logic and theorem proving. You know Python now, right? Well, I put my bets on rule-based programming paradigm as a controlled inference process. Rule-based programming may also be used for explicit programming, while entirely controlling each inference step from the programmers side. Roughly, NN is a black box - you train the NN, hoping it will magically connect inputs to outputs. Then to use it, you feed different inputs, yielding different outputs, and it turns out to be a good mach for big enough training corpus. On the other side, rule-based programming is a bit different - you have to explicitly connect different forms of input to different forms of output, and that is what makes it more controllable. Unfortunately, the current state of public research (as far as I'm aware of it) doesn't provide much clue about how to generally describe process of constructing connections between inputs and outputs. We can do it by hand for this or that occasion, but those are specific, not general solutions. I'm also unaware of such a general solution, but I stopped a bit before that. Anyway, to describe my attempts, I think I have clued up a nice way for describing such specific connections. It is nothing new, but just a bit more concise and tidy than I used to see from projects around. I'm not saying ANNs are not working or it isn't a way to go, I'm just saying I invested my time at the different side of AI spectrum. Probably the solution we all want is in some combination between symbolic AI and ANN, so I believe a contribution to each side may be valuable, IMHO.

Big Facepalm.....no, I can get the Hutter Prize down to near the current champion score without using backprop / gradient descent. I'm currently at approx. 20.3MB once I fix a little thing shadowing it on larger data. I know 100% how my AI makes predictions based on given context. There is no hoping / magically connect inputs to outputs....