Private AI development?

[DennisGorelik]

Dennis Gorelik:
"Neural net should be implemented in software (for instance based on RDBMS)."

Really? Wouldn't you lose what makes ANN so great - the massive parallel processing power??

Sorry, but massive parallel processing power is not the most important feature of ANN.

The two most important features of ANN are:
1) Ability to change weight of its links.
2) Ability to <a href="http://www.dennisgorelik.com/ai/DecisionMaking.htm">make decisions</a> based on the weights of the links.

 

[G71]

Please, correct me if I'm wrong, but if you implement limited syntax in your system then you stuck with limitations of your language and your system cannot get full access to the huge collection of texts created by human civilization.

Every language has some syntax and there is always the possibility that some sort of NL_to_MySyntax conversion tool will be developed later. I'm currently unable to process NL well (no one seems to be able to do it well) and I'm not even very interested in that. I'm interested mainly in the AI algorithms. The key initial step is just to pass the information to the system in a format the system can understand. When you have a really good system, it's not a problem to get lots of students preparing the input data in the needed format. That huge collection you mentioned contains lots of useless/incorrect/old/impertinent/naive/duplicated (etc..) data. An AI system can act intelligently even with very limited knowledge base + I can get lots of data from Cyc-like systems if needed. Getting the meaning from regular sentences (automatic learning from the Internet for example) is a huge problem I do not have time to deal with. Most of the systems which are trying to understand the data see its future in XML like technologies. In my system, lots of the needed metadata is provided just by the syntax.

 

[G71]

Sorry, but massive parallel processing power is not the most important feature of ANN.

I do not consider myself to be an expert on neural networks but I took some ANN classes and one of the first things teachers usually mentioned was that the 2 major features of neural networks are massive parallel processing and adaptivity.

 

[G71]

If you really want to develop <a href="http://www.dennisgorelik.com/ai/StrongAI.htm">strong AI</a> you need to freeze all your projects which are not directly related to the strong AI project.

It's not that easy for me. I work on several projects which I consider very important. I currently spend most of my private dev time working on application for serious medical research. People are dying because researchers do not have as good tools as they could/should. It's hard for me to ignore that. I know that AGI would help them too but there is a big difference between what it takes to finish an AGI project vs relatively small (yet powerful) domain specific tools.

 

[posi]

Doing more than reading? Are YOU developing your own AI system? A system where you can change anything you want any time you want?

I'm developing my own AI system that sounds alot like what you're on about. I've just setup my own website to start some form of development, aicentral.net. My email is sheer@wave.co.nz if you want to talk some more. I've worked on AI theory for the last six years and am now ready to make my system. I've done all the formulas etc to what I think is to be an 'intelligent system', and am presently trying to develop my own programming language to make it.

 

[G71]

posi,
Where is the description of the language?

 

[kula]

2 nodes make a line, 3 nodes make a triangel. When using a 3d engine the maximum number of nodes to make a sphere were set to 90, although it did not look perfect. How many nodes do you need to make a perfect sphere? Size does not matter. (It has to follow the laws of physics and it has to be 3 dimensional.)

I doubt that all of you can answer that question correctly. Can you build an AI that can?

Depends on what the sphere was made of and the size of each individual atom/particle/wavelength that it was constructed from.

Maybe a quantum computer could be made to calculate a perfect sphere, but as soon as it was measured.....poof, big bang. Maybe thats what happened to god.

kula

 

[G71]

Baal, I hope you already woke up from that micro-world crap and realized that a perfect 3D simulation of the real world is practically impossible because of various limitations (our hardware, our knowledge, time constraints, etc) and more importantly, it's not needed at all in order to develop a powerful AI system. If you need a 3D world for your AI then use one of the existing 3D engines (=recommended at this point) or build a real-world robot. Limitations of the current 3D engines are not that important and the AI system should be able to adapt to the level of details of its current world. Even just the text-world is a great environment for intelligent systems. BTW it would be great if you can finally post some of the promised screenshots of your AI system. I guess it's not a big deal to hit the PrintScreen and post it. FYI, I'm not moving forward with my AI project now, even though I have very clear ideas for coding new modules. I'm now working on a non-AI private project (with a few other hot shot developers), going for big money. It's kind of unusual opportunity which I did not want to miss. I'm not sure when I'm gonna get back to my AI stuff at this point. I just know that I will get back to it unless the other project (or something) kills me. ;-)

 

[Baal Zebul]

Frankly i see difficulties using computer vision in a computer simulation. Did you think that i was going to have a camera in front of the monitor that would record the simulation and send back to the computer?

The latest advances that i have made will be very difficult to build in a simulation. It is however possible but also unneccesary since they come for "free" in the real world.

Jiri, everything you write makes me happy because it gives me an insight of how limited your knowledge are on the human brain. Nobody that understands how a human processes information would write in the style that you do.

 

[G71]

Did you think that i was going to have a camera in front of the monitor that would record the simulation and send back to the computer?

No. Why camera? You would already have a 2D stream. I would be more concerned about the processing of the 2D stream data (and doing it in real time) and about the 3D simulation itself. You were originally talking about a complex 3D world simulation with many real-world-like features (many intelligent life forms living/working there, dealing with global disasters, detailed simulation of laws of physics, etc).

everything you write makes me happy because it gives me an insight of how limited your knowledge are on the human brain. Nobody that understands how a human processes information would write in the style that you do.

OK my favorite all-knowing kid ;-), if it makes you so happy then lets assume that you know a LOT more about human brain than me ;-). What you should realize is that we are talking about AI which is supposed to run on current (or near future) computers. These systems will not process information the same way as human brains. Many things need to be done very differently and we often do not really need to know how it's exactly handled by our brains in order to develop powerful AI systems. For this purpose, it's good enough if we just understand some of the high level principles of our thinking. When you talk with senior AI developers who understand some of the high level stuff differently then I recommend you to go into details and make sure that the misunderstanding is not on your side. When it comes to the detailed level of our AI systems then you need to know more about technology than about human brains.

Just curious, what was the biggest software project you have actually finished? I noticed certain patters in what you wrote on various places which make me wonder about you ability to get non-trivial projects done. You may want to put some thoughts on your development methodology. Can you give us a brief overview of the progress/status of your AI project (any link?)? Since I'm currently busy with non-AI projects, I would like to see the others making some progress with their AI. In that sense, I'm the exact opposite of you. I want you (and many others who have something to do with AI development) to understand things well and succeeded. What was the URL of your AI web site? Since you have reportedly spent more than a year working on your AI (which is, as you put it, “more important than life” to you), one would expect that you finally have some real stuff there. As you know, I'm not interested to read the general AI garbage stuff which can be found on thousands of pages on the Internet or about components which currently cannot be developed because of technology limitations. I’m interested in practical algorithms, functional user-AI interfaces and technical details about real (and currently doable) AI projects. Do you have something like the user manual for your AI? Any details about reasoning of your AI? Anything testable demo (finally)? Let me guess: you cannot really provide anything because:

- “You do not like the competition” (If you are as clever and knowledgeable as you think then you do not need to worry about that. ;-)) and/or

- "You are bored" (that’s a good one ;-))) and/or

- "You know too much and others (or just I) know soooo little" (When I get less busy I may create a web page with quotes from folks like you and make the funny stuff obvious enough for them to see.. Now, seriously, stop comparing yourself with others and do some real work on your AI) and/or

- "it involves some huge secrets” - potential “Terminator movie threat” etc so you are basically saving the world when not providing any practical details about your amazing project... ;-)

Or do you have something else to add to the “Baal’s top excuse list” this time? I wonder what kind of magic prevents you from pressing the print screen button and posting a screenshot. Sorry for pushing you ;-))).. It may help you in certain sense. ;-))

 

[Baal Zebul]

No. Why camera? You would already have a 2D stream. I would be more concerned about the processing of the 2D stream data (and doing it in real time) and about the 3D simulation itself. You were originally talking about a complex 3D world simulation with many real-world-like features (many intelligent life forms living/working there, dealing with global disasters, detailed simulation of laws of physics, etc).

Do you know what an Isometric view is?
If (etc 20) ALF's are simulated then ... I see it as possible but very unneccesary to use the same visual stream for every ALF, first retrieve the D3DVector for 20 ALF's then calculate their field of view.
Are you suggesting to use not only the Isometric view, but also a seperate camera for every ALF? Or are you suggesting that i should use a text-based feedback as i intended?
Do you want multiple viewports? The most viewports that i have worked on simultaniously are 4 (a world builder)
20 different viewports plus one main window with the Isometric view. It would be rather messy, and i doubt that it would be optimized.

btw, that is not what i intend to be working on.

- “You do not like the competition” (If you are as clever and knowledgeable as you think then you do not need to worry about that. ;-)) and/or

Of course i do not like competition.

- "You know too much and others (or just I) know soooo little" (When I get less busy I may create a web page with quotes from folks like you and make the funny stuff obvious enough for them to see.. Now, seriously, stop comparing yourself with others and do some real work on your AI) and/or

Once again you demonstrate stupidity. Have you finally relized that you will never invent real AI? who are you trying to impress? is it your family that you wish to matter to?, the others at this forum?, perhaps your friends?, maybe some boss at work? Let me give you a little advice, this is so minor that you would not matter more, you would matter less. You might be more "famous", but you would be famous in the wrong fields. I reckon that your parents raised you better and that it is the environment that has changed your parameter thresholds, leaving you so dire in need of recognition that you even would try to matter in fields that no one cares about, a field were only those that were too stupid to ever succed end up.
When you do know who you are then you might have a chance at understanding AI.

 

[G71]

Do you know what an Isometric view is?

Yes, I'm familiar with all the axonometric views. But even if you avoid the deformation of objects using the isometric view, you still have a bunch of tough problems to deal with (unless the world contains only primitive types of objects).

Are you suggesting to use not only the Isometric view, but also a seperate camera for every ALF? Or are you suggesting that i should use a text-based feedback as i intended? Do you want multiple viewports?

What I have seen strongly suggests that you have a serious problem with algorithms you want to use for reasoning and planning. If you did not make major (and well tested) improvements in those areas then forget the 3D simulation of that world. Focus on problem solving of a single ALF in relatively-easy-to-develop environments which can provide rich scenarios for complex problem solving. Use text as the main input/output format. Support timelines and n-dimensional arrays with pointers on various objects. For example, if it's supposed to play chess then it can create a 2D array (the board) and pointers would point simple objects like "black king", "nothing", "white queen" etc. The color and type would be stored separately in a dynamic list of properties (inherited from a parent objects) for a particular object. Teach it game rules, known real world rules, various types of relationships and hierarchies, roles associated with various processes, deductive/inductive/abductive reasoning etc. Let it play with some math, with some basic coding, tell it about itself, yourself and others, show it what can be done with MIDI, let it understand how jokes work etc. Let it solve some sticky theoretical problems. When it works well (not just in your mind, but in the real test environment) then start to think about providing other, harder-to-build, worlds or build a real world robot. How the system thinks is just much more important than a particular type of complex environment it interacts with.

Have you finally relized that you will never invent real AI?

Considering how wrong you were with a bunch of other things, I take this as a compliment. ;-)

Once again you demonstrate stupidity... who are you trying to impress?

It might be beyond your understanding because the recognition is one of the key driving forces in your mind, but just relax and try to understand that some people have simply no need to impress the others. Believe it or not, I'm one of those people. What I said was for you and people like you. The people who keep making big claims and who still have empty hands after while. Look what you wrote a year ago. The stuff you currently talk about pretty much shows where you are with you AI. From my point of view, it looks like you nearly wasted a year. Time can fly very quickly when playing with AI. you need to really think what's worth to focus on. You may look back 5 years from now and realize that you are still far from finishing the very first fully functional version of your AI project. I know several people who ended like that. Be careful because you seem to be very well on that way.

 

[Baal Zebul]

Yes, I'm familiar with all the axonometric views. But even if you avoid the deformation of objects using the isometric view, you still have a bunch of tough problems to deal with (unless the world contains only primitive types of objects).

I am a bit confused here, how could possibly the type of camera i choose affect the complexity of an object. So if i have a chase camera then my objects can be much more complex?
However the "complexity" of the objects (the polycount) will affect the fps count and since the isometric camera have a larger field of view the camera will see much more of the high-poly object which naturally will affect the fps count negativly. But, if you have a isometric view then you will not need that high-poly objects since they will not look better any way. Instead you can focus on the shape of the objects and making trees for instance highly complex with many branches and leaves. Of course, the "looks" of the tree close up might not be as good as you might want but the complexity will be because when you look at them at a further distance they still look good.

What I have seen strongly suggests that you have a serious problem with algorithms you want to use for reasoning and planning. If you did not make major (and well tested) improvements in those areas then forget the 3D simulation of that world

I am confused here too because i just gave you facts about how 3d environments work. I have 4 years experience on 3d game developing and i have worked on games for almost every day during those 4 years. But yet you are trying to say that i have no idea how it works, despite that i just illustrated the negative outcome of your suggestion.

Considering how wrong you were with a bunch of other things, I take this as a compliment. ;-)

You can take it any way you wish. But do remember that what you just said contradicts what you said next.

It might be beyond your understanding because the recognition is one of the key driving forces in your mind, but just relax and try to understand that some people have simply no need to impress the others.

You could not be more wrong.


It is actually pretty funny because what i have written in my latest document seems to convince people pretty well.

1. Jiri, please do explain why people see at as fit to "risk" and even lose their lives in the line of duty as firemen, police officers, soldiers.
2. Please explain why a soldier throws himself at a grenade to save his plutons life.

I can say already now that you have no clue but please try, i look forward to your reply

 

[G71]

how could possibly the type of camera i choose affect the complexity of an object. So if i have a chase camera then my objects can be much more complex?

Not complexity, other things, including readability of some object parameters can be affected.

isometric camera have a larger field of view

Isometric view makes 3D objects to appear bigger so it's more likely that less stuff will fit the view (comparing to the regular 3D view).

I am confused here too because i just gave you facts about how 3d environments work.

The list of important details you did not mention is long and it seems to me that you are not even aware of some of the challenges. But it does not really make sense to discuss it this way. If you want to discuss it then provide a link to a decent documentation (including details about ALF-3D interaction). Then we can systematically discuss it. Without that, the discussion is kind of mess.

I have 4 years experience on 3d game developing and i have worked on games for almost every day during those 4 years.

Yeah, that clearly defines you: every day.. 4 years.. And then, when I was interested in details then the best thing you had to show was some old screenshot form a never-compelted 3D "game", generated by a simple-to-use 3D-world generator developed by others. Where do you think your AI will be in 4 years? Long ago, you were talking about having crystal clear AI ideas and about being done by now. You were confident that you know all the needed stuff as well as you are confident at this point. Think about that a bit. Think about those people who, back then, (unlike me) did not see a problem in your ideas. Make sure you get feedback from right folks. BTW experience with 3D games is helpful but it's not the key to the strong AI development. IF it was, the world would be different now. Now, try to really focus on the next sentence. Forget 3D and focus on advanced reasoning algorithms. The brain is more important than the look. If you think the brain is good enough then tell me what kind of testing was done.

what i have written in my latest document seems to convince people pretty well

Even if the best folks in the field review your well done documentation of the project (let's say you have it) and say that they are convinced that it's the strong AI we were all waiting for then it still does not mean that it's really the thing. In the history of AI, many things looked promising (especially on paper) and later proved to be dead ends.

1. Jiri, please do explain why people see at as fit to "risk" and even lose their lives in the line of duty as firemen, police officers, soldiers.
2. Please explain why a soldier throws himself at a grenade to save his plutons life.

By making the decision, they are satisfying their own need to do what feels right to them ATM. And that's based on their system of values which can be altered by many ways. Those who are intentionally committing the worst crimes are doing the same thing.

 

[Baal Zebul]

Isometric view makes 3D objects to appear bigger so it's more likely that less stuff will fit the view (comparing to the regular 3D view).

You use the isometric camera as a "top" camera, you could naturally use it as a normal camera and raise the zoom to get the result that you just described but i have never used the isometric camera that way.
You set your zoom of the isometric camera and you set your far plane (which is the distance from which the isometric camera is viewing the world)


I worked on 5 games during those 4 years.

By making the decision, they are satisfying their own need to do what feels right to them ATM. And that's based on their system of values which can be altered by many ways. Those who are intentionally committing the worst crimes are doing the same thing.

You are right but you understand the problem very generally. what you said about system values are of course correct but you do lack the most important detail.

 

[G71]

You use the isometric camera as a "top" camera

Then maybe you should consider 2D world and some simple icons representing objects on the screen (just to provide something visible if you think it's necessary). Internally, you can support a dynamic (possibly complex) list of properties etc. One of the reasons why I chose text-oriented AI is the simple way of introducing new concepts to the AI. It can make valid thoughts about new concepts even when having very limited knowledge about it because that little piece of info might be sufficient for a number of particular thoughts. In 3D, new concepts = lots of work which users likely cannot do and lots of details which may not necessarily be important. Introduction of new concepts is something you really want to keep simple because there is a huge number of concepts to learn in order to understand our world and none of us knows everything. You do not want to have the dev team involved whenever there is something new for the AI to learn (like new types of objects and their relationship with the other objects and world events) + when you simulate the nD world then you can easily hit the limits of your hardware.