Trinary; The flawless Base-3 Number system.

[Dinoguy1000]

Ooh la la, I'll have to put that on my userpage pronto.

Hehe, yeah, I tend to get bored alot and go around stealing other people's userboxes...

I'll go ahead and move the thread then...

[Deniz Oezmen]

Ooh la la, I'll have to put that on my userpage pronto.

Do as you please.

Hehe, yeah, I tend to get bored alot and go around stealing other people's userboxes...

You must have been bored for quite some time.

So ... let's see if I can get something productive done ...

[Darkstar]

What I don't get is: How did you "create" a base-3 system? Base-n systems are well known since... I don't know, the middle ages or something? I remember doing arbitrary base-n calculations in school when I was about 10.

What's so revolutionary about your implementation of base-3? I guess you can do calculations a bit more "efficient" (i.e. it takes less cycles to multiply 2 numbers, for example), but at the cost of added complexity in the CPU ALU. I don't think it'll be worth it. And if you want to completely ditch backward-compatibility with all existing systems, all hardware manufacturers will laugh at you.

But I'm interested anyway

You said you'll answer questions. So here is what I'm particularily interested in:
*Is your computer a von Neumann architecture or Harvard, or something completely new?
*Or are you simply talking about gates (like and/or/not/nand/add/...) which you "changed" from accepting and delivering binary to trinary?
*What is that trinary transistor you're talking about?
*How do you "encode" the 3 states on a wire? 0V, 1V, 2V voltage levels? or more like 0V, -5V, +5V? Or do you use currents/frequency modulation/... to represent them?

Oh, and btw, simply re-designing computers to work in base-3 instead of base-2 won't magically create the "most efficient programming language ever". If your language is turing-complete, it is exactly as powerful as C (or any other significant language out there).

As a sidenote, there are actually already "thing"s in hardware-land which use base-n calculations right now. For example WiFi uses QPSK (quad phase shift keying), which can be seen as working in base-4. And Hamming codes (used on CDs and in image encoding for example) work in almost any base, for example Hamming(11,7) is base 128 code...

[Atrusino]

Ah, the very reason that I have almost no time to post here.. This system will, in time, allow a computer to say 'Why?'
The system itself has possibilities, but as you said.. Changing to use just the system alone is impractical.. The costs along would be rediculous.. However, my research has mostly been going into that big question.. "Why?" Once I figure out how to make that work.. AI will no longer be a thing of the future.

[Darkstar]

I could make my C64 ask "why?" over 15 years ago:

Code: Select all

10 print "Why?"
20 goto 10

Ok, jokes aside

The thing is: Everything you develop with/for your base-3 system, I can write a program to "emulate" or "simulate" whatever your system does on a "standard" computer. So if you say "AI will be a reality with trinary", I can then implement the same on a "standard" computer, even if it's a bit slower because you can do whatever you do more efficient in base-3.

Since you didn't answer my questions I have some more, now that you brought up AI.
* How does your AI system differ from previous AI/logic programming languages? Especially how do you want to overcome the difficulties which basically stopped research with Lisp, Prolog and SHRDLU more than 10 years ago? (i.e. deduction from a database, efficient storage of "knowledge" data, deduction of new deduction rules from old ones, etc.)

Edit: Oops, didn't notice that I used the wrong tags there...

[Atrusino]

The only AI that I've ever seen in the works doesn't truly 'understand' anything that it isn't programmed to undestand. An example is the way bots work in a video game.. They know to resolve certain paths to locate a player and when they locate them, they know to use a certain weapon to attempt to kill them. But the 'AI' there doesn't understand why it has to do that. I'm talking about an actual conscience that is capable of understanding the nature of a problem and not just the problem.

[Captain]

The only AI that I've ever seen in the works doesn't truly 'understand' anything that it isn't programmed to undestand. An example is the way bots work in a video game.. They know to resolve certain paths to locate a player and when they locate them, they know to use a certain weapon to attempt to kill them. But the 'AI' there doesn't understand why it has to do that. I'm talking about an actual conscience that is capable of understanding the nature of a problem and not just the problem.

Blah blah you're avoiding questions and basically talking out of your ass at this point. Either tell us how you intend to solve this AI problem, and why you would need a trinary system to solve it (which you don't), or go bore someone else.

[Darkstar]

Blah blah you're avoiding questions and basically talking out of your ass at this point. Either tell us how you intend to solve this AI problem, and why you would need a trinary system to solve it (which you don't), or go bore someone else.

Exactly what I wanted to say just now. I already thought so right after the first post, but I wanted to give him another chance I thought maybe he was so busy coding/developing his trinary system that he didn't have much time to post here.

But now it's pretty obvious that he has no clue what he's talking about.

@Atrusino: Good luck in getting anyone to buy your shiny new invention. And I'm already keen on seeing and communicating with your 'trinary' AI. I hope it'll be smarter than Eliza, and, who knows, maybe it even passes the Turing-Test

[Deniz Oezmen]

Wait ... all of this was actually intended to be serious?

(I read it with a pair of big irony tags in mind ...)

[Dinoguy1000]

Hmm... If you want to get close to true AI, you should probably have a look at neural nets (unless my understanding of how they work and what they're supposed to do is flawed).

[Deniz Oezmen]

Neural nets are one approach for all kinds of decision-making systems, even "learning" ones, but they have their share of practical problems as well; just as an example: NNs tend to become relatively large in order to gain meaningful results for non-trivial problems. This leads to difficult learning phases, where highly complex optimization problems have to be solved to effectively train the system parameters. In practice, this can often only be done approximatively, without a guarantee of optimality.

I'm no expert in AI, but the pattern recognition folks usually tend to look for systems with fewer parameters and a structure that is more closely adapted to the problem at hand (like hidden markov models with underlying language models in speech recognition, for example). NNs do generally not have a problem-specific structure, thus they need fine-tuned input data and much computing capacity.

(That is not to say that NNs are a bad idea. As most techniques they have their use and have been -- and probably still are -- a topic of research. I'm curious what the future will bring, but there is still a variety of other approaches.)

Anyway, much of the AI euphoria of the past decades has seriously cooled off when people began to realize that the sheer computing power needed (regardless of the base used ) is not the only hurdle to building a "comprehending system" (whatever that may be), let alone the prospect of creating a piece of soft- or hardware with some sort of "conciousness". Unfortunately, there's much more work to be done than just finding the right rules and applying them.

[Dinoguy1000]

Aah, thanks for explaining them for me.

As far as recent research into AI is concerned, it is my understanding that the concept of a "universal" AI - one that can effectively solve any problem that a biological system can (once again, my definition/understanding may be whack) - has been largely given up in favor of pseudo-AIs that are designed to solve specific problems - like what you explained (a couple of examples would be the autopilot software installed in many commercial and military aircraft, or similar software currently in development for civilian (I think?) land vehicles).

[Deniz Oezmen]

As I said, I'm no expert on the field of AI research. However, I believe you are right. As far as I know, the terms applied to these two different paradigms are "strong AI" and "weak AI".

Of course, there is much philosophical dissent about whether a weak AI should really be called an AI. It all boils down to the problem of defining intelligence and our view of ourselves. At first, it seems unreasonable that a mere (more or less) brute-force problem solver like Deep Blue should be called truly "intelligent", as it just follows a certain programming and reacts to its input in a defined way, but does not "know" what it does. But what would happen if we someday realized that our own thoughts were also the result of a deterministic (biological) programming and that knowledge was just a state of memory? Hilarity would ensue ...

Well, I remain sceptical of our chances to ever answer this question, so we can leave this to the philosophers. Nevertheless, I find many of the theoretical problems that stem from the research field of weak AI to be very interesting and also practically relevant -- like the autopliot software you mentioned, or speech recognition, translation systems and many others.

[Dinoguy1000]

One thought I've actually had more than once would be the possibility of training a weak AI system to be able to identify unknown file formats. I have no idea how accurate it would be, or how much work it would take, but it's still interesting to think about.

[Atrusino]

Sorry it took me so long to get back..

Blah blah you're avoiding questions and basically talking out of your ass at this point.

somewhat, but not quite..

Either tell us how you intend to solve this AI problem, and why you would need a trinary system to solve it (which you don't), or go bore someone else.

AI isn't a problem because it doesn't really exist yet.. There are scattered forms of a self-thinking program, but they are all reather limited at the moment. as to why you need a base-3 system? you don't really 'need' it per se, but it would make things faster as well as more efficient in the long run. AI will always be a bit limited if you run it in a base-2 environment.

Exactly what I wanted to say just now. I already thought so right after the first post, but I wanted to give him another chance I thought maybe he was so busy coding/developing his trinary system that he didn't have much time to post here.[/quite]

As to the last.. Been rather busy doing a lot, heh.. As to the first, eh.. I'm not all there in the head anyway often don't write things exactly how I want them to turn out. Ragardless, I'm trying to salvage that mistake.

But now it's pretty obvious that he has no clue what he's talking about.

Have it your way. There are a lot of parts that I don't quite understand, but I do certainyl udnerstand the transistor I have to create to get the thing to work anyway.

Good luck in getting anyone to buy your shiny new invention

The invention has just begun.. I've got a lot more work to go and the advancements on the standard mathmatical system to the theory that is now present as well as the transistor that makes it all tick is the pavement that will become a long road.

As far as recent research into AI is concerned, it is my understanding that the concept of a "universal" AI - one that can effectively solve any problem that a biological system can (once again, my definition/understanding may be whack) - has been largely given up in favor of pseudo-AIs that are designed to solve specific problems - like what you explained (a couple of examples would be the autopilot software installed in many commercial and military aircraft, or similar software currently in development for civilian (I think?) land vehicles).

Why have such limited software?

I have no idea how accurate it would be, or how much work it would take, but it's still interesting to think about.

Hopefully, I'll have a computer solving simple word problems by the end of next year.. The complicated proccess that is actually involved there is enough to get myself a reasonable understanding of what needs to happen for a computer to actually 'recognize' something. Fromt here on, the possibilities of file recognition or anything of the like are quite real.


As far as dodging the original question..
This is a base-3 number system inplemented in a 3-state environment like you'd see it in a computer.

2-6-18-81
1-3-9--27
0-0-0--0

What I've got is a little different and as per the term "Base-3" I shouldn't have called it that. I had to do some research myself on the base-3 system to understand what it meant. I'm a bit of a theorist and do tend to forget certain alleys of math and I certainly don't know everything.

-Atrusino

Edit: I've got a schematic for the transistory ready to ship off to the patent office as well. Fun stuff, though I'm still rather new to it.