FAQ Search
Memberlist Usergroups
Profile
  Forum Statistics Register
 Log in to check your private messages
Log in to check your private messages
Moonpod Homepage Starscape Information Mr. Robot Information Free Game Downloads Starscape Highscore Table
Best RTS
Goto page Previous  1, 2, 3, 4, 5, 6, 7  Next
Post new topic   Reply to topic    Discussion Pod Forum Index -> Battlescape View previous topic :: View next topic  
 Author
Message
icarus
Troll
Troll


Joined: 01 Mar 2004

Location: Olympia Washington



PostPosted: Wed Sep 29, 2004 8:54 pm    Post subject: Reply with quote

actulay a normal coumputer could never outthink a human
but a neural net* coumputer desigend to emulate a normal brain could beacuse it is desigend to work like a human brain

and the coumputer going sentint isint to far feched but is unlikly whith todays coumputers
but more's law dictates that coumputers will increse in speed at a rapid rate countuinisly (i forgot excatly how fast)


i prodict a future in whith the hole virus beacoming sentint thing will become comin place
its all beacuse of something i call order from chaos
"if you have an infanite number of monkeys sonner or latter you will get shakespeers compleet works
the internet is the monkeys and sakespeers coplete works are the sentint cyber entaty

*also caled a posatronic brain in asmov's books
Back to top
View user's profile Visit poster's website
Poo Bear
Pod Team
Pod Team


Joined: 14 Oct 2002
Posts: 4121
Location: Sheffield, UK



PostPosted: Thu Sep 30, 2004 5:41 am    Post subject: Reply with quote

Neural Net programs are light years away from getting anywhere close to a real human brain. We cannot even implement what we already know about real neurons anyway, never mind what we don't know. A neuron takes 1->many electrical signals and may or may not fire off a signal of its own - sounds simple enough.

So far we know a neuron can send out at least 50 different types of signal, presumably in response to different input states. Each signal can be "tweaked" to travel a different distance after leaving the neuron. There are other mechanisms in the brain releasing chemicals that affect the neurons function.

Now a human brain contains about 8 billion of these, our DNA (which took 100's of millions of years to evolve) sets up the initial connections for us, up to the age of about 4years a huge number of new connections are made defining our "personality", then 1000's more continue to be made and broken for the rest of our lives as we learn and forget.

Conclusion
1. we don't know how brains work even at a basic level.
2. we don't understand how the connections are made day to day or the mechanism by which the blueprints were laid down in our DNA.
3. if you ignore the fact we don't know what we are doing, is it even possible to build anything this complicated with this many connections any time soon - no.


Sorry - I hate it when books, tv and films talk about this utopian vision of the future with faster than light space ships, intelligent computers, super human robots, immortality, no need for money, everyone working together to colonise the planets, yadda, yadda. I prefer a darker more realistic vision of prejudice, figthing, chaos and struggle, I guess i'm a realist (sadly).
Back to top
View user's profile Visit poster's website
icarus
Troll
Troll


Joined: 01 Mar 2004

Location: Olympia Washington



PostPosted: Thu Sep 30, 2004 6:12 am    Post subject: Reply with quote

no the future will have sentint coumputers hyperdrive and supper robots but life will still suck
Back to top
View user's profile Visit poster's website
Johnh



Joined: 06 Sep 2003
Posts: 160



PostPosted: Thu Sep 30, 2004 1:55 pm    Post subject: Reply with quote

Actually, i dont' think a "positronic" brain makes much sense. Positrons are positive electrons (i.e. anti-electrons). Hence they are annihilated whenever they come in contact with an electrong, producing massive amounts of energy.

Another thing to consider: We don't even know WHAT sentience/consciousness really is. Let alone being able to emulate it. As far as we know consciousness may just be an illusion produced by an abstraction of the many moles of molecules reacting in just the right ways. Religions might also be correct in claiming that your consciousness comes from "the soul" instead of the brain. In which case we would need a whole new branch of science dedicated to whatever your soul is made from, and another dedicated to determining the laws by which your 'soul' must live by. Call it "Spiritual Chemistry" and "Spiritual Physics".

Anyways, I don't know anything about Neural net programs, but they are still programs. We like to presume that humans do not follow a set program, but a computer must always adhere to it's programming. Of course we don't know if we are really any different from computers. That's a really scary thought.
Back to top
View user's profile
icarus
Troll
Troll


Joined: 01 Mar 2004

Location: Olympia Washington



PostPosted: Thu Sep 30, 2004 3:01 pm    Post subject: Reply with quote

well thats what i thought asmov ment when he sayed posatronic brain i geuss i wass wrong

and neural nets are not programs thay are the actual coumputter (hardwhare not softwhare)
Back to top
View user's profile Visit poster's website
Dunbar



Joined: 22 Sep 2003
Posts: 21



PostPosted: Thu Sep 30, 2004 8:00 pm    Post subject: Reply with quote

For the nerds out there:

Quote:
A positronic brain is a fictional technological device, originally conceived by science fiction writer Isaac Asimov. Its role is to serve as a central computer for a robot, and, in some unspecified way, to provide it with a form of consciousness recognisable to humans. When Asimov wrote his first robot stories in 1939/1940, the positron was a newly discovered particle and so the buzz word positronic, coined by analogy with electronic, added a contemporary gloss of popular science to the concept.

Asimov remained vague about the technical details except to assert that the brain's substructure was formed from an alloy of platinum and iridium. Asimov relied on the reader's knowledge of the capacity of positrons and electrons to be formed in pairs and to annihilate each other, in order to convey the impression that such pair creation and destruction could serve as a metaphor for the evanescence of thought. The focus of Asimov's stories was directed more towards the software of robots (such as the Three Laws of Robotics) than the hardware in which it was implemented.


Johnh makes a good point about defining consciousness. One example of a test for artificial intelligence is the Turing Test, proposed by Alan Turing. In this test a human judge engages in a natural language conversation with two other parties, one a human and the other a machine; if the judge cannot reliably tell which is which, then the machine is said to pass the test. Turing originally proposed the test in order to replace the emotionally charged and for him meaningless question "Can machines think?" with a more well-defined one.

There are problems with this test of course; for a more information look up "turing test" in Wikipedia. There's also information on neural nets, such as:

Quote:
Most researchers today would agree that artificial neural networks are quite different from the brain in terms of structure. Like the brain, however, a neural net is a massively parallel collection of small and simple processing units where the interconnections form a large part of the network's intelligence; however, in terms of scale, a brain is massively larger than a neural network, and the units used in a neural network are typically far simpler than neurons. Nevertheless, certain functions that seem exclusive to the brain such as learning, have been replicated on a simpler scale, with neural networks.


So it is a kind of hardware.
Back to top
View user's profile
Poo Bear
Pod Team
Pod Team


Joined: 14 Oct 2002
Posts: 4121
Location: Sheffield, UK



PostPosted: Fri Oct 01, 2004 6:16 am    Post subject: Reply with quote

icarus wrote:
and neural nets are not programs thay are the actual coumputter (hardwhare not softwhare)


Wrong - neural networks (like most things) are conceptual ideas which can happily exist as a piece of software on your PC, as a dedicated hardwired purpose built machine or even as a temporary configuration within a connection machine*. Just google for "backpropagation network" which is a specific type of neural net and you will see plenty of examples to download.


*a connection machine is a type of super computer that can rewire itself electronically on demand to perform a given parallel operation at incredible speed.
Back to top
View user's profile Visit poster's website
VonBraun



Joined: 29 Apr 2004
Posts: 42



PostPosted: Fri Oct 01, 2004 12:28 pm    Post subject: Reply with quote

It is almost impossible to mimic a human brain, as no brain is perfect. Not even Cloned brains would be perfect (if they are ever used(Stupid UK Cloning and GM rules GRRRRRR... Sad )
Back to top
View user's profile
Johnh



Joined: 06 Sep 2003
Posts: 160



PostPosted: Fri Oct 01, 2004 2:35 pm    Post subject: Reply with quote

Presuming I do well enough in my Computer Science classes, I might be learning about neural networks in a few years =)
Back to top
View user's profile
grayrest



Joined: 22 Jun 2003
Posts: 7



PostPosted: Sat Oct 09, 2004 2:16 am    Post subject: Reply with quote

I've never gotten into AI -- I've never been impressed by neural nets and that seems like the best thing going -- and I haven't done any game programming, but I have played a whole lot of strategy games. A confession as I begin: I'm imagining TA the entire time I'm writing this.

It's been established that game AI programming is difficult. This parallels other areas of computer science, like speech. The solution in these other areas is to create a huge database to take care of the hard part and build a rule set to take care of the rest.

In RTS games, I see the tactics as being a fairly well defined rule set. There have been games -- Myth stands out in my mind -- which do incredible jobs with the AI but don't have any strategy programming, only tactics. Building up a library of tactics (encircling, melee tanking, ranged unit retreat, flank, pincer) isn't easy, but seems like it would be possible to build a rule based system that can competently play tactics.

The problem comes in the larger scale. Like Go, there are simply too many possiblities to iterate through all the possible positions and pick the best, especially in real time. I've wondered for years (got the idea around 2000) why game programmers don't extract information from the human players and send these stats to a central server. This would create a feedback loop and, theoretically, a perpetually challenging virtual opponent.

Looking back after a college education, I suppose this would be a neural net, now that I think about it again... Regardless, even a simple build order in TA would have dramatically increased the difficulty of the AI. Other low hanging fruit would be unit building tendencies by minutes into the game. Both of these would be map-dependent to be most effective. There are more complicated possibilities: preferred routes and play style when ahead versus when behind, play style breakdown. I'm sure there are more.

I know I'm not the first to come up with this. What I haven't figured out is why nobody has done this. Is it because it's difficult to implement? or just too much effort to justify as part of the game development cycle?
Back to top
View user's profile
limulus



Joined: 05 Aug 2004
Posts: 14



PostPosted: Mon Oct 11, 2004 6:25 am    Post subject: Reply with quote

There is no such thing as AI in any game that has ever been released so far.

What games have is an IF...THEN set of rules, some games have a bad ruleset (see C&C 1 where one of the rules for the Computer seems to be: "Build units all the time but only send them to the enemy's base in groups of 1-4 units so they can easily be killed") and some have a good one (like Starcraft, where the Computer really seems to use your "weak points" to his advantage), but no game has anything more than a rule set. To tell you the truth, I usually require something around 2-4 hours when playing a new strategy game until I have figured out a way to defeat the CPU every time.

A true (intelligent) AI would be one that could then adapt to your strategies. Say in game X you have found a weakness in the CPU ruleset: The player never defends its storage buildings. You start attacking the storage buildings and they get destroyed. After that, not only does the CPU build new buildings, but it remembers you attacking them before, assumes you are going to try again and uses this knowledge to its advantage. That would be true AI. Of course, this can also be accomplished with a complex set of rules, but then intelligence is nothing but a VERY complex ruleset.
Back to top
View user's profile
r000000b
Starscape Jedi
Starscape Jedi


Joined: 10 Jan 2004
Posts: 63
Location: Staffordshire



PostPosted: Mon Oct 11, 2004 7:12 am    Post subject: Reply with quote

probably because its too hard, really really hard,


Quote:
The solution in these other areas is to create a huge database to take care of the hard part and build a rule set to take care of the rest


getting the data is easy its the 'rules' that are hard. Same with computer vision.


Some thoughts that go through my head when im thinking about using a tactic:

* Is it appropriate given the placement (and types) of units and the map terrain
* Did it work last time?
* Can I disguise it?
* Is it still viable if he realises what I'm up to
* Is it going to give me a significant advantage if it works
* How can I best capitalise on it if it does work
* What is going to happen if it goes wrong
* How can reduce my losses involved
* Can I change my mind if I have to
* When is the point of no-return
* How many units do I need?
* is it worth assembling a larger force or should I go now with surprise

Its hard to quantify such informaition (what is estimated surprise) and measure it let alone make decisions based on it.

If you could create a good representation of the game (hard) state, then you could learn routes through a game state space (hard) that lead to more favourable games states (hard). Even that would have to be heuristicly driven (hard). You cant exhaustivly search game state space, I wonder if there is any milage in a more unit up rather than commangder down approach using unit agents to communicate and coordinate simple activities at a low level that could lead to emergent 'big' tactics. Interesting stuff for a researcher but off the critical path of a game developer i guess.


PS.

Neural nets are overated, they dont have any magic that allows them to do anything a traditional program cant. But they do sound cool and have a place in scifi. In most applications running a NN is a last resort, understanding a trained net is hard. I suppose in situations where you have a lot of data and dont know what is going on then you may as well run it through one, but there are other ways to mine data
Back to top
View user's profile
r000000b
Starscape Jedi
Starscape Jedi


Joined: 10 Jan 2004
Posts: 63
Location: Staffordshire



PostPosted: Mon Oct 11, 2004 7:31 am    Post subject: Reply with quote

Quote:
but then intelligence is nothing but a VERY complex ruleset


Computation is only a candidate for AI. We dont know exactly what intelegence is, however with computation we can simulate anything to arbitry precision, so we can probably make things that seem as intelegent as we want. (they may not actulaly be intelegent)


Quote:
What games have is an IF...THEN set of rules

Its the dots in IF...THEN... that hare the hard bit, so is the ordering of the rules and the applicability (generalness) of the rules. You also need ways of updating the rules and their ordering. I went to a seminar not to long ago about training this kind of pattern matching systems using a genetic algorithm, as I recall evaluating the importance and contributoin of a rule is very hard and chagning them sanely is hard.
Back to top
View user's profile
zwzsg



Joined: 01 Nov 2004
Posts: 15



PostPosted: Mon Nov 01, 2004 2:05 pm    Post subject: Reply with quote



VonBraun wrote:
My faverite games (at this point in time) are
TA:Final Frontier
[...]

Yay! Another TA:FF fan! I like it alot too! Especially when used along with OTA units, even a krogoth look tiny compared to a piledriver!




And TA is still my favorite RTS, because for reason that I don't understand, there's still no RTS that has copied its simple & beautiful, powerfull mechanics. Things like shift enqueing order, guarding factories with F.A.R.K. to make them build faster, planes that dogfight over your base yet are fully controllable, reclaimable wreckage, ......



Oh, and those screenshots use some 3rd party stuff like 3D replayer, the Original TA is 2D and without those fancy starships.


Oh and about AIs, I really don't want an AI that learn and play like an human, because anyway, games AI and human works completly differently, they don't even have the same scope (the AI must entertain the player, send him units to blow up, but not put the player in a situation where he can't win, while the human player just use every trick he can to win). A dumb and scripted AI, if correctly done, can be much fun! Just give us cleverly made mission with lots of in-mission twist and special events, and multiplayer, then I'm happy even if the AI isn't top-notch.
Back to top
View user's profile Visit poster's website
Demi Virus



Joined: 27 Oct 2004
Posts: 20
Location: Chicago



PostPosted: Sat Nov 06, 2004 11:18 am    Post subject: Reply with quote

For the old Windows 3.1. It has to be Metal Marines
Back to top
View user's profile AIM Address
Display posts from previous:   
Post new topic   Reply to topic    Discussion Pod Forum Index -> Battlescape All times are GMT
Goto page Previous  1, 2, 3, 4, 5, 6, 7  Next
Page 4 of 7

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB © 2001, 2005 phpBB Group