NikiTo said:
I feel less positive about AMD than ago too. NV just “cares more”. Somebody could say that NV is not “carrying more” but it is “evil corporation trying to monopolize the whole world”. It depends of the view of a person and depends of his own hidden agendas.

No matter where the sympathy goes, NV has always been a very innovative company.
Still i think their RTX does more harm than good, but AMD gives me no option to make a point of flexibility > performance. >;(
But may change. Actually affordable RT GPUs for gamers is more a myth than ever before. Still hoping on Epic to push some progress.

I think in the beginning i could try dots around

I see that with this there is just ‘no problem to solve’, while with hexapod robotics there is.
Though, for robotic locomotion there is no need for AI. It works, but ‘traditional’ approach is 100x faster and more predictable IMO.
Also, if you remember Googles ‘walking’ AI ragdoll, result is more a laugh than useful, while interesting.

To me, AI is much more promising for ‘behavior’, on top of robotic control problems. Like the example of ‘playing games’, where agents learned to block each other way with moving obstacles to block doors. Only AI can do this, it's extremely promising for application both in games and real world robotics. And such thing you could model indeed with representing characters just as a point.
I'm convinced this can bring us true new games, despite doubts of others.

If this makes sense to you, you have to decide: Learn how to solve physics control problems (e.g. balancing ragdoll, lifting a box, stacking it up, etc.), AND learn ML for behaviour. Or do all things with just ML, which seems less effort so makes sense (but not sure if it''s indeed less effort, and catching two flies with only one hit is that easy).
Or you be smart: Focus on behavior, leave progress on low level control or animation to others, see how both sides evolve and join at some point. Just for example. What i mean is, you likely want to specialize on something narrow, to increase chance of success. The higher and wider the goal, the lesser the chance.

This really just said to make a point (similar to my exaggerated psychological logic about your ‘cycle of failure’ from before).

In fact, a hexapod is extremely simply to control robustly with a trivial state machine. Newton Physics developer did so in one of his demos. While this approach does not generalize to bipeds, you could just use AI behavior in top of that state machine to get robotic feel and impression, which is something that ofc. matters to ourselves, not only to make a showcase for others. It's not cheating at all, it's just picking the best tool for given problem.
On the other hand, if you control all joints with only ML, that's progress too, even if it can be done with a simpler and faster method. But because this simpler method exists, your progress is not enough to justify the use of ML, and you need to do some more as well then.

JoeJ said:
I see that with this there is just ‘no problem to solve’, while with hexapod robotics there is.

There is. But the problem is so small, so easy, it could be handwritten code, no NN.

JoeJ said:
Though, for robotic locomotion there is no need for AI. It works, but ‘traditional’ approach is 100x faster and more predictable IMO.

Absolutely correct. Two robots fighting in a close hand combat. One uses trained NNs, the other uses hand written code. The hand written code one will absolutely win. The hand written will have always the answers, and super fast. Imagine of a boxer. He is repeating the same movements thousands of times every day, until his brain remembers the movements and until they become a reflex. A reflex like when you close your eyes when somebody throws water in your face. It is super fast. Mega fast. No time to “think”.

Motor reflexes are hand written code. They appear early, are always there. Maybe come as a program from the genes, not from training.

NNs are slower. But they could compensate depending of the case. I know i am bad at boxing, therefore i am not fighting in a ring - survived. My slow NNs told me avoid the danger.

So, the AI robot could avoid close hand combat when it sees that the other robot has perfect reflexes, because it will “think” and would rather take a long stick and hit from the distance.

JoeJ said:
To me, AI is much more promising for ‘behavior’, on top of robotic control problems. Like the example of ‘playing games’, where agents learned to block each other way with moving obstacles to block doors. Only AI can do this, it's extremely promising for application both in games and real world robotics. And such thing you could model indeed with representing characters just as a point.

Hand written code is much faster. More effective in most of the situations. You are not thinking while walking or drinking. It is a reflex. You think when you have to speak.

We agree on that. In my opinion, a handwritten code must move the legs of a hexapod and AI must control the direction through the labyrinth.

I could go for -
hand codded reflexes. Ai for thinking.
Ai for everything, that learns to be fast on reflexes. Like this - 

This method is a compromise between productivity and final speed. It would take me ages to hand code the physics and Kinematics. And still it is a machine, so it is fast enough. Not as fast as hand written code, but could be not even noticeably slow.

And a super mega robot should be able to switch on will, between the two modes. Nothing new. In the last movie of Robocop they explored this concept - a thinking human is slow to react. They compared a thinking human to a machine with handwritten AI. The robot absolutely outperformed the human. This is why they added to Robocop a switch to can switch between reflex-killer-mode and reading-a-book-of-poetry-sitting-on-the-bank-in-a-park-mode. I am giving you examples with movies, not because i learn from movies. But in order you to see other people already worked a lot about some of the ideas. 

JoeJ said:
But because this simpler method exists, your progress is not enough to justify the use of ML, and you need to do some more as well then.

It matters - productivity!!!

Cheaper to develop by teaching it, than hiring programmers for Kinematics. Returning to the topic of the OP a bit. Price!! Focus on development price. In the car factories it is not the robots that replaced human workers. People are mad at robots. It was the boss who wanted to make more money faster who replaced human workers, by using robots. A human signs the contract to install robots in a car factory and a human signs the firing order for 1000 people to go home(and 10000 job places more that are related to the factory, will be gone too).

Don't blame the robot for what the human did. Robots are perfect at doing what the human told them to do.

And even I who seemed to to hype about AI here, am NOT gonna dream about conscience. An AI can replace a human and still to have not a conscience. We could never demystify conscience. In order an AI to replace a human, all the AI must do is repeat what the human does, cheaper, faster, better. Who talks about conscience, me not. 1000 years before science discover conscience in machines, humans will be replaced already by robots. And few humans who think they are gods will control all these robots. Still human race is alive no? While a single person remains alive, human race is alive too.

Try to talk about soul to a businessman who can win more with robots than with humans. Would laugh at your soul-talk. And if you insist, the smart businessman would try to fool you with talks like “i am saving the world here. Not becoming a billionaire. No no no. I am saving the world. More people need to eat more, need more cars and clothes faster. Overpopulation requires me to install robots. Overpopulation. Overpopulation. Better have more robots than more humans. More robots is not overpopulation. No no no. I save the world. Overpopulation.”

Overpopulation is overpopulation ONLY if it is with humans, not with robots, because robots make MONEY.

A factory with 5000 workers is more ecological than a factory with robots. Humans eat natural products, coming form earth. And poop natural waste going back to earth. Then becoming food again. Poop-food-poop-food, nature is so beautiful. 
What are 100 billions humans for the planet? - overPOOPulation.
Robots need lot of energy(IIRC brain consumes only 15 watts). 
El. energy for robots is produced in a less natural way than food-energy. Food fueling the muscles of the workers is a perfect most natural most ecological poop-reinvesting production. Batteries are fine, but super ineffective. Produce toxic waste in the making. Nowhere to put the waste. A human produces natural wastes. And when the human is broken in the end of his service, he will lay in the ground and recycle itself completely. Machines recycle too, they rust, but much slower.

And a rusting machine is a waste. A rotting human body is full of life, starting almost immediately after the death.

Anyways, i love robots. I would just not install mlns of them. I would use limited number of robots. Only when it is necessary, not when it makes more profit. But this will not happen. Robots will win, because humans are greedy for profit.

If hippies arise and attack the factories for robots, a human(it will be a human to give the order) will send 100 military robots to censor the hippies. Who would win - gaming AI configured to make only headshots or a modern sedentary human who can not take the dirt on the battlefield? Billionaire wins!
 

NikiTo said:
There is. But the problem is so small, so easy, it could be handwritten code, no NN.

The NN tutorial i did years ago was like this. The NN learned to solve a simple linear equation. It was about dots moving on the left side of the line, visually. Not impressive, but easy to understand. The equation was used to rate the ‘fitness’ of neurons AFAIK, but the neurons themselves did not know about the equation. Still they learned it this way.
Good example i guess. Easier than image labeling. But can't find the tutorial anymore. I think it's a good start. Likely after that i could learn image labeling and understand this too. I mean, ‘really understand’, not just getting the concept from talk or articles. That's what i think you should do first.

NikiTo said:
Two robots fighting in a close hand combat. One uses trained NNs, the other uses hand written code. The hand written code one will absolutely win.

Nah, that's surely complex enough ML is the much better choice. I mean really simple things like ‘walk from A to B'. That's simple. How to move limbs to walk, how to keep in balance while getting there. I can not do much more with my ‘traditional’ approach. Fighting is a skill that should be 'learned' i think.

MLs application is where hand written code is no longer possible, or takes ages to develop. You see that - ML is NOT used for things we were able to do before its rise, with few exceptions like reading scanned books of printed text. Btw, we used such software at work >20 years ago. It worked, but needed manual interaction when it was unsure about a word. Thus it was faster and cheaper to give the text to a secretary to type it in. That's one problem i see about your current project: A human can achieve the same goal likely just better and faster.

It matters - productivity!!!

Depends on your experience. If you lack that for things like skeleton hierarchies, dealing with 3D rotation, basic inverse kinematics, which is necessary even for the simple state machine, then ML might get you to the moving hexapos faster. Even if you have to learn about ML first. IDK.

NikiTo said:
So, the AI robot could avoid close hand combat when it sees that the other robot has perfect reflexes, because it will “think” and would rather take a long stick and hit from the distance.

I think you put too much into current ML here. You talk about real intelligence here.

NikiTo said:
Hand written code is much faster.

Not always, see ML imitating fluid sim being faster than simulation. Though that's not the same ofc.

NikiTo said:
We agree on that. In my opinion, a handwritten code must move the legs of a hexapod and AI must control the direction through the labyrinth.

Yeah, but we want smooth transition, so ML can take over as problem complexity increases, while simple analytical math model will always add better balancing than ‘learned experience’.
Not sure how easy it is to mix that.

NikiTo said:
In the last movie of Robocop

You should know that Hollywood or Anime rarely is helpful to science, not even for inspiration.
It's only useful for the philosophic aspect, but a scientist does not care about that while working. That's the job of society, not the scientist. IMO.

NikiTo said:
Don't blame the robot for what the human did.

I'll never blame a robot. A robot is only a tool of mankind. So far…

NikiTo said:
Anyways, i love robots.

But they don't love you. ; )

JoeJ said:
I mean, ‘really understand’, not just getting the concept from talk or articles. That's what i think you should do first.

There are two kind of lectures about NNs -

how to use NNs
what happens inside, why, how

You can already use NNs with the first one only and train an hexapod.

But i too want to know what happens inside. So far, i think i have the deep forward networks pretty understood. And i know what are RNNs to be used. But i want to learn RNNs in the deep as forward deep networks.

Definition of deep network is loose like for big data.

JoeJ said:
Nah, that's surely complex enough ML is the much better choice. I mean really simple things like ‘walk from A to B'. That's simple. How to move limbs to walk, how to keep in balance while getting there. I can not do much more with my ‘traditional’ approach. Fighting is a skill that should be 'learned' i think.

Boxing is pretty simple. Pure reflex is enough. You alone can code it manually.
MMA has more tricks allowed, you have work in a team. It's more codding.

I think if you don;t try to repeat Jackie Chan's style in a fight, and resort only to Kick Boxing, it is manually codeable. Kickboxing covers it it almost everything and is effective. In a real fight almost nobody uses legs. Legs are used in movies. Look at competitions of karate - mostly boxing. Legs are hard to use. They are used rarely when the goal it so win in a competition. And it makes less sense for a robot. Because a human kick has a meaning because it is stronger than a hand. It rarely reaches the target, but when it reaches it is a guaranteed win. There is old science of martial arts that evolved with time. I am not talking about a football kick. A well done, karate kick is a guaranteed win. It is not football. For a robot it makes no difference. Already using its machine-strength, a robot can punch stronger than a human can kick.

It makes no sense to a robot to fight like Jackie Chan. At least for me. I can not imagine a situation that can not be won only with hands.

So a robot could just box and destroy the opponent. And boxing is very simple.

I think many of the researches would try to revisit a problem like reading letters, regardless it is already working now. Because these researches know they could discover something, improve here and there. I have this feeling. Polishing an old problem could make you see some discovery.

JoeJ said:
I think you put too much into current ML here. You talk about real intelligence here.

I will not insist here, because i am not sure. Remember all these movies about martial arts where a single super master is kicking the ass of 50 opponents. This is nothing realistic. Number one tries to win, loses, number two tries and loses, number three, 10, 20, and more and more people come to ask for a beating.
 

Ai could measure the speed of the opponent by observing him kicking the ass of the first 10 opponents and then give up. In movies, it happens that 10 criminals just seen Superman bending a train rail…. still attack him LOL “He just bent an iron from the railroad, but i will still try to win hitting him with my wooden stick”. A robot would have obtained your full stats by observing your fight. And once it knows your stats, it could decide -

JoeJ said:
Not always, see ML imitating fluid sim being faster than simulation. Though that's not the same ofc.

I don;t know if this can be called faster, because it does not repeat the result. Not close. Looks nice, but not close.

JoeJ said:
Yeah, but we want smooth transition, so ML can take over as problem complexity increases, while simple analytical math model will always add better balancing than ‘learned experience’. Not sure how easy it is to mix that.

 Some times you need precision. NNs are not precise. they are closer to feeling, mood, taste. Not a precise exact result. 

JoeJ said:
You should know that Hollywood or Anime rarely is helpful to science, not even for inspiration.

I told you i don;t learn from movies. And for inspiration, for me, movies are great. 

If you are not inspired you are doing a task.

I know i do art when i code, because i get inspired. If inspiration moves you, it is art. If you make a logical reasoning like - “i need to program because i need to eat” it is a task. Without muse it is not art.

JoeJ said:
But they don't love you. ; )

I am a shy geek, nobody really loves me. Nikola Tesla never married. Rumors say probably died a virgin.

I am tired of repeating - if it acts as if it is alive, it is alive. The rest is religion.

if (input == “do you love me?”) {
  output  = “Yes! Yes yes yes!!!”;
}

See? It loves. This is more than enough. Add to ti GI and it will be awesome. The rest is religion.

When you are in love, how much of this love is thanks to well demystified chemical reactions and hormones? Love has no mystery. They teach it in marketing. It is a hollywood thing to sell movies, because love sells a lot.

NikiTo said:
I will not insist here, because i am not sure.

Nobody can be sure, because we don't know what ‘intelligence’ is. So i take that back anyways (even before reading your further response). The example of observing an enemy and classifying his skill can be learned too. I just assumed a narrower scope in my comment.

We would end up at discussing ‘can we achieve AGI?’, ‘do you believe we will?’, etc.

I thought about that while eating. And i came up with a new aspect from those new questions:

‘Can we increase human intelligence by combining multiple humans?’ I think it's a yes due to interaction and communication, likely the main reason we became so smart.
For machines it's even easier: More cores, more memory, networks, etc. Machines can be summed up without a loss in computational power.
AGI as a result of both those processes seems not unlikely to happen anymore to me. Before that, my question ‘can a creature build a machine that is smarter than the creature?’ got a no, because a yes sounds too much like a perpetuum mobile, violating laws of nature.

So i became a bit more willing to believe your SciFi visions about the future, maybe.

But lacking definitions, lacking understanding of ourselves and how our mind works, it all remains philosophy in any case. AI will pass Turing Test? We'll construct a harder test. It will never end.

I have one question for you: Do you want a independent species of intelligent machines?
Or do you want them to obey your orders?

NikiTo said:
I am tired of repeating - if it acts as if it is alive, it is alive. The rest is religion. if (input == “do you love me?”) { output = “Yes! Yes yes yes!!!”; }

Thanks for the answer. Ethically, that's not correct, NikiTo!
(Beside my doubts it would work for you to betray yourself so easily.)

Stop dealing with pointless stuff. If you wanna work faster, really stop looking at things as to why they take too ridiculous amount of time. It will lead you nowhere if you start being buroucratic, casting tenders on, “why we are slow”.

If you wanna know, why it takes time, consider how you spend your development time.

I will enter and try to help up, what is the one thing you miss in your current project?

What are three things you would like in your project/pipeline/assets?

Ano no, there are too many variables in a game that you could declare AI successfully delivering it by coding source code logic .

JoeJ said:
We would end up at discussing ‘can we achieve AGI?’, ‘do you believe we will?’, etc.

I changed my mind a bit about AGI. It is a loose term. It depends from one person to another.

If you are businessman and your harvesting machines can go to the field, harvest it and gets back without a human, the businessman would say - “see? it is more intelligent than the human workers i fired. They were always getting drunk and were falling asleep in the field.”

Who needs AGI, who cares about AGI? The important question is - “Does the robot win over a human in a fair competition?” Because if the robot outperforms the human in all of the aspects of the job, it is the most fair thing the robot to take the job of the human.

JoeJ said:
Machines can be summed up without a loss in computational power.

We have two teams working on a cure.
Team A is made up of 10 researches of IQ 140.
Team B is made up of one single researches of IQ 200.

Can team A do a better job than team B?

I don't think 100 idiots thinking together can outsmart a single genius.

JoeJ said:
‘Can we increase human intelligence by combining multiple humans?’ I think it's a yes due to interaction and communication, likely the main reason we became so smart.

I am not sure here either. If you finish faster, you are smarter. Take as example an exam in school, they give you 45 minutes to finish. Saying that - “My time ran out, i would have solved it if i had 5 minutes more” will not save you from a penalization. Because the nerd in the class actually finished in 45 minutes. This means the nerd is (most probably)smarter than you. Because he finished faster. And even an idiot would solve a very complicated puzzle given enough time to try all the possibilities.

Then the brute force of a computer can finish earlier than doing analytic search for the result.

You could argue that it is not fair to use brute force to get a result faster. But consider this -

Two robots solve a puzzle. One finishes faster than the other. He gets applauded. And gets a prize. Then after two weeks somebody rings on the door of the winning robot ant the robot opens the door and the person outside tells him - “Mr. Robot, we revised your power consumption during the test with the puzzle and we discovered that you was bruteforcing. Therefore we lament to communicate to you that we disqualify you and you have to return the price”

Does it really matter? Bruteforce or logical analysis? If you have an army of logical robots and the enemy has an army of bruteforcing robots, and they destroy you army by using better strategy in war, what matters more? Honors for having used logic or your head rolling for having lost the war?

It is all very philosophical.

Again - “If it acts like if it is smart, it is smart”
“if looks intelligent, it is intelligent”

“Not fair - he has a bigger head than mine, more brain inside, this is why he outsmarted me” - nobody will give a f. You was outsmarted by a gianthead. Stop whining!

If it is better than you at a job and replaced you for that job, it does not matter if it has an official certificate “AGI-approved” glued on its forehead or not. You starve it eats. Evolution tells you you are weaker, gonna disappear soon as a failure.

Anime quote -

A deity - “I can not believe that a simple human defeated me - the deity” and dissappears. (typical anime) Just as an example.

therefore -

An AGI robot - “I can not believe i was outsmarted by a bruteforcer” and dissappears.

Who cares about AGI?!?!? It is only words. If it has a database with all the possible answers to all the question you could possibly ask it, it can act intelligently. This is what matters. the rest is RACISM - “anallyzer vs bruteforcer vs databaser”

Is connecting wires to live biological neurons fair? It is not fair. It is not pure software based solution. It is not fair. But it will take your job, you starve to death.

I have to clear it again - only in case of a fair competition. If a robot steals your job because it robot sleeps with the boss, it is not valid.

JoeJ said:
Before that, my question ‘can a creature build a machine that is smarter than the creature?’ got a no, because a yes sounds too much like a perpetuum mobile, violating laws of nature.

It often happens a very smart child to be born from two alcoholic parents. The child has then more IQ than both of its parent summed together. It just that the child evolved in another path. Different than the parents.

This is a proof that you can create an evolutionary algorithm that could evolve more than you evolved and outsmart you.

JoeJ said:
So i became a bit more willing to believe your SciFi visions about the future, maybe.

It depends of you. Some researcher could create a robot race that colonizes the universe and you still could be complaining - "But it is just code/artificial neurons inside. But it has not a soul. But it can not die if you kill it" complain complain complain

Or you could open your mind a little bit to the idea. Only a little bit, because if you want too much to believe you could fall in some of the cults of some tech CEO. I htink you should open a bit more than your current state.

Reconsider again - "if it acts like a thing, it is that thing". And you will have to apply to this your own tastes - how much it has to repeat a human? Is it enough to talk to you, or you want it to touch you too? Do you need it to fart as a human or you could skip over this? It depends of your personal taste from here on.

In tat Series Devs they repeated a human perfectly atom by atom. Repeating even the state of the quantum particles. they were considering these clones not only alive, but equivalent to the original. It depends of you, but you are too pessimistic about it.

Hormones often impede humans to think clearly. Computer do not suffer from this. Is this fair? This is a huge advantage for computers. What about backups? A computer can have backups and replicate, is this a fair thing?

I will make you sit on a table in front of a computer and you two will have to solve a puzzle. Then you two will start. The computer will connect to the cloud, will generate 10bln random forests and get the answer in 2 seconds.

Then you would be like - “I complain! It is not fair. It used the cloud, bruteforce and made 10bln simulations in 5 seconds. I can not do that. the more i can simulate is one scenario at a time. Not fair. Not fair, it has advantage over me.”

Gasparov reaction when a computer beat him at his own game -
time in the link - look at his face. read his feeling. Does it really matter if it used SIMD or not inside? It beat a human.
https://youtu.be/3EQA679DFRg?t=664

JoeJ said:
I have one question for you: Do you want a independent species of intelligent machines? Or do you want them to obey your orders? NikiTo said: I am tired of repeating - if it acts as if it is alive, it is alive. The rest is religion. if (input == “do you love me?”) { output = “Yes! Yes yes yes!!!”; } Thanks for the answer. Ethically, that's not correct, NikiTo! (Beside my doubts it would work for you to betray yourself so easily.)

Everybody betrays himself one way or another in a different scales.

You can not prove to me you would not become the worst dictator in the history of the universe if you had the power. You can not prove it, because you was never tempted with such a power. You could say “i would be good, i would donate half my money to charity, then half my blood to sick pigeons” but this is all words. You can not prove it. You are not given the chance to prove it. And i would vote to not give you that power, just in case. Only i deserve that power. Only i can handle it. - is that the way you think about yourself? Do you think you would be a better billionaire than the ones we have now? You and me, we can only bla bla bla about morals, because at the moment of the truth, not even you know how would you react. Not even you know yourself enough. It is easy to claim - “i never stolen not a single cent" when you never worked in politics. then i give you a job in politics, you are given the opportunity to steal and surprise surprise you steal more than everybody before you. Not even you know it.

Churches are full of sinners who pretend to be saints. Politics are full of robbers who pretend to never steal.

Your moral-talk has no effect on me. You did not die on a cross to pardon the sins of your enemies. You has not proven to have morals in a practical way. Only words. and words are easy.

About AGI again - natural language is the last frontier for AI. It is curious how what looks the simplest task - chat bot is the hardest thing to do. It is even mythological and religious -

“”"""The oldest stories of golems date to early Judaism. … Adam was initially created as a golem … Like Adam, all golems are created from mud by those close to divinity, but no … golem is fully human. Early on, the main disability of the golem was its inability to speak. … Rava created a man. He sent the man to Rav Zeira. Rav Zeira spoke to him, but he did not answer. Rav Zeira said, "You were created by the sages; return to your dust". """""

It is not AGI if it can not speak. You can use it to define AGI - being able to fluidly use natural language. If it can not speak, it is worthless. From a religious book that dates much much before AI.

(Clarification - I know what AGI officially means, but what i want to say when i use “AGI” here is the thing you are after - the ultimate goal, the holly grail of AI. When you say "AGI", you mean the last task to solve in AI. The ultimate goal. After that goal is reached, it is only learning and becoming smarter and smarter. That same “AGI”. I know you write the acronym “AGI” but i you mean something loose.)

Resuming it - you need to relax a bit your expectation of intelligence. It is all very loose and philosophical, not well defined but already we can work with the loose term. We already can work with unclear terms and goals. If we relax a bit. Lower the guard a bit. If an AI tells you a very good joke, would you laugh? Or you will hold your mouth with your hand forcing you to not laugh, because it has not a soul. Ok, it has not a soul. I agree, but it can act as if it has a soul. NNs can fake complicated and sophisticated moods, flavors. Ofc it has not a soul. But you still could enjoy the music it makes or the paintings it creates. You still can use it to deliver you pizza. After all, it is easier to define intelligence and conscience than defining a soul. Leave that “it has not a soul” behind a bit.

NikiTo said:
I don't think 100 idiots thinking together can outsmart a single genius.

Because dumbness adds up as well :D

NikiTo said:
Then the brute force of a computer can finish earlier than doing analytic search for the result.

Ill posed: Analytic solutions don't involve a search, but brute force might. AFAIK.
In many cases an iterative numerical solution is faster than analytic one, though. Or the latter is just too complicated to be widely used.

NikiTo said:
Who cares about AGI?

I wonder about that too. I thought you would?

Maybe we want it because nothing is ever fast enough for us anymore, not even evolution. So we sacrifice ourselves in favor of synthetic replacement just to proof we can progress faster.

Toooo much text, NikiTo. You're dawdling. No wonder your project takes too long ; )

Maybe John Carmack works on AGI now because he got tired about finishing stuff.