Page 1 of 2 [ 26 posts ]  Go to page 1, 2  Next

ahayes
Veteran
Veteran

User avatar

Joined: 2 Dec 2006
Gender: Male
Posts: 9,506

14 Jan 2007, 4:48 pm

no, I'm not interested in politicians



andy1976uk
Veteran
Veteran

User avatar

Joined: 9 Dec 2006
Gender: Male
Posts: 1,122

14 Jan 2007, 6:27 pm

Atomika wrote:
The one huge obstacle with creating "strong" AI is that it will take computer programers several hundred years to program as much information into a computer possessed by a 2 year old human.

But, in MIT and in other places, they are experimenting with neural-nets and they are programming computers to "learn" information to get around this obstacle.



That approach in addition to employing genetic algorithms seems like they best approach IMO.




Quote:
I think in MIT they have a robot that can actually play air-hockey and can distinguish between objects. Also, in Japan they have are working on a robot that has geniune emotions.


Have you seen the Robocup contest?

http://www.robocup.org/




Quote:
I hear that it likes the smell of alcohol.


Hmmm, they've created Bender from Futurama!

Image


Maybe human-like robots aren't that far in the future after all :lol:



l05tin5pac3
Tufted Titmouse
Tufted Titmouse

User avatar

Joined: 23 Mar 2009
Age: 45
Gender: Male
Posts: 42
Location: Austria

24 Mar 2009, 2:39 pm

Anyone into AI still here? I noticed the last post is some time ago...

I am researcher in AI. Especially interested in strong AI. I have started a thread on aspiesforfreedom dot com but didn't get much feedback. at least not much useful feedback.



ruveyn
Veteran
Veteran

User avatar

Joined: 21 Sep 2008
Age: 87
Gender: Male
Posts: 31,502
Location: New Jersey

25 Mar 2009, 6:23 am

Mythical wrote:
I've always been interested in AI, but only recently actually started to do any research on it..

Parents can be so annoying sometimes :(


I worked in the field for over 20 years. There is no such thing as artificial intelligence. There are rule based computational schemes but anyway you slice it all you have are Turing Machines in one guise or another.

There are some useful pattern recognition heuristics in use (I developed a few of those myself). That is NOT intelligence. Creative Intelligence does not work algorithmically. A computer can check the proof of a theorem for correctness. It cannot figure out what theorems to try and prove. That requires a living brain.

ruveyn



l05tin5pac3
Tufted Titmouse
Tufted Titmouse

User avatar

Joined: 23 Mar 2009
Age: 45
Gender: Male
Posts: 42
Location: Austria

25 Mar 2009, 7:25 am

ruveyn wrote:
There is no such thing as artificial intelligence. There are rule based computational schemes but anyway you slice it all you have are Turing Machines in one guise or another.


for those who don't know:
- a turing machine is a universal machine, which can theoretically compute everything.
- a class 4 (classification after stephen wolfram) cellular automaton is an algorithm which can generate turing machines and even other class 4 automata inside itself by random

@ruveyn
do you want to say that a "living brain" has something / can do something a turing machine cannot do?
even if you were right, can you claim the same about a class 4 cellular automaton?
are you a dualist?



UrchinStar47
Sea Gull
Sea Gull

User avatar

Joined: 23 Feb 2009
Age: 37
Gender: Male
Posts: 216

26 Mar 2009, 12:43 pm

l05tin5pac3 wrote:
for those who don't know:
- a turing machine is a universal machine, which can theoretically compute everything.

A Turing machine is in itself a purely theoretical device, particularly an Universal Turing machine that you seem to be reffering to. There are mathematical equivilents to touring machines as well.

Still, an UTM is only capable of calculating solutions to finite algoritms. It will also halt on finding an undecidable problem.



ruveyn
Veteran
Veteran

User avatar

Joined: 21 Sep 2008
Age: 87
Gender: Male
Posts: 31,502
Location: New Jersey

28 Mar 2009, 11:32 am

l05tin5pac3 wrote:
do you want to say that a "living brain" has something / can do something a turing machine cannot do?
even if you were right, can you claim the same about a class 4 cellular automaton?
are you a dualist?


Dualist? Watch your language! A finite state cellular automata array can be simulated by a UTM so it computes the same class of functions as a UTM which, by Church's Thesis is the class of recursive functions.

ruveyn



l05tin5pac3
Tufted Titmouse
Tufted Titmouse

User avatar

Joined: 23 Mar 2009
Age: 45
Gender: Male
Posts: 42
Location: Austria

29 Mar 2009, 3:39 pm

ok, sorry about the dualist offence. won't happen again. but then, what makes you think a thing like a human brain cannot be reproduced in any other way than it's biological form?

ruveyn wrote:
A finite state cellular automata array can be simulated by a UTM so it computes the same class of functions as a UTM which, by Church's Thesis is the class of recursive functions.


people here seem to know stuff like this. still you didn't answer my question. but so as not to make it solely your problem, I will also state two theories now:

my "weak AI theorem":
a human brain, digitalized by modeling every single one of its neurons with every single one of its connections using relatively simple artificial neurons (the weights of the connections deduced by observing every single dendrite for some days), connected with a biological body instead of the original brain, will not be distinguishable from the original and - even more importantly - will not be able to tell the difference between itself and the original during a short period of time (maybe hours). Given a retrogradient chemical weight modifying possibility it may even be undistinguishable for a longer time.

my "strong AI theorem":
given a digital genetic code which encodes a virtual body including a huge (~10^12 neurons with ~10^15 connections) frequency based ("spiking" / "pulsed") neural net which can develop (fractal dimensioned) sub-structures (like our cortex and ganglia) and a complex virtual environment, and inter-dependencies between this environment and the virtual creature (natural enemies, changing weather, food, ...) it is possible to "breed" an intelligent creature. the more similar the environment and the creature to our real world, the more similar to a human it can get.

I know this is currently impractical and I know those theorems lack a great amount of refinement, but I guess you get the idea. I think "consciousness" and all this ill defined stuff emerge from "correctly" dimensioned neural networks, and not only biological ones. I believe the proteomics to play only a secondary role in cognition and neural action which a couple of experiments (remember aplysia?) seem to support.

looking forward to your counter arguments!



l05tin5pac3
Tufted Titmouse
Tufted Titmouse

User avatar

Joined: 23 Mar 2009
Age: 45
Gender: Male
Posts: 42
Location: Austria

31 Mar 2009, 4:23 pm

one more theory before the thread passes away into oblivion again:

a cortex like structure (4 layers of artificial frequency based inhibitory and excitatory neurons) in combination with an amorphous hopfield-net-like nucleus, some pre-classified sensory input and a couple of output possibilities, constructed in a certain way (how is my secret) so that it will develop a rich self-representation, can be trained to understand natural language and will develop consciousness and a personality as well as drives, emotions and so on if trained accordingly (as you would "train" your own child).

and even more radical:
i think this is already practical. I think two relatively standard IBM stations with about 1024 processors each plus a little grid with about 100k machines should do the job just fine - in real time!

bluntly put: I think we already have the equipment for a strong, human like AI since more than 10 years. we are just (yet) too stupid to do it. even too stupid to try to do it.



Alec-Teal
Butterfly
Butterfly

User avatar

Joined: 9 Apr 2009
Age: 30
Gender: Male
Posts: 12

11 Apr 2009, 6:03 pm

i find AI facinating, there are so many different types though
AI can be defined as:
- anything that reacts not randomly to some sort of event or stimuli (source me :) )

now unfortunatly if the AI was able to dynamically code itself
- it could create stuff smarter then itself -

- and thus something like terminator might actually happen, that would suck

But ai thats learns is different

lets say you had some sort of bot, that got hurt if it touched an object hotter then 50d Celcius
and was able to recognise objects taht had 8 rotational symetry phases, oreanted perpendicularly to the robots local xz plain

have i toched this before
- no
-touch it
-did i get burned
-yes
-remember not to touch that
-no
-rember they are safe to touch
-yes
- was it safe to touch last time
-yes
-do i want to touch it again (or check it)
-yes
-check it
-did i get burned
-yes
-rember not to touch that
-no
-its still safe
-no
-find something else
-no
- do i want / need to check it again
-yes
-check it
-did i get burned
-yes
-rember not to touch that
-no
-its still safe
-no
- find something else


Alec

BTW this AI would be smart if its job was to touch stuff, the part about rotational symetry at the top is to make it easier about how it would see things, and not think everything was different due to shape

-alec

[edit]
this was all indented nicly
just associate the questions to the nearest yes/no
as they dont span / cant span outside each other



Ichinin
Veteran
Veteran

User avatar

Joined: 3 Apr 2009
Gender: Male
Posts: 3,653
Location: A cold place with lots of blondes.

11 Apr 2009, 6:52 pm

I have had a bit of intrest in AI-related things, like pathfinding. I tried looking at it from another perspective and realised that if you pre-filtered the game matrix (2d one for simplicity) you would end up with a pre-calculated walkable path that required little computing power, and the processing power could be spent on selecting the shortest/fastest path instead.

Some day i would love to make my own game, this is one reason why i dived into the whole pathfinding thing.

There is an intresting toy here in Gamasutra: look for the link "Pathdemo" on the page:

http://www.gamasutra.com/features/19970801/pathfinding.htm

(it is a bit old and written in Delphi, but it works in Windows XP, i dont know if it works on Vista and newer systems.)

Anyway, it allows you to draw obstacles in a 2D matrix and select a start/stop point for the selected algorithm(there are many) to try to find its way through.


_________________
"It is far better to grasp the Universe as it really is than to persist in delusion, however satisfying and reassuring" (Carl Sagan)