BinarySurfer

Thank you Infinite Recursion. Finally, someone voiced his or her opinion about the future advancements in robotics.

Duck

AI is not compariable with any other area of technology, eg, stealth technology, that is not a valid argument.


Anyway I will argue no more.


I've spent years thinking about how to create 'real' artificial intelligence (and years working on AI systems in my job). And although it would be impossible to implement and no computer will ever have enough processing power/memory to run it, I think I have a solution that 'might' work. Anyway this is it.

You would have to create a simulator of the universe, that simulates down to atomic level (or even further), it would have to simulate all the rules of physics, and all the rules of nature. Then into this simulator you would create 'planet-earth/sun/moon/etc' replicas.

Then with your universe created, and all the rules of physics/nature implemented, you would add the building blocks of life, cells/dna and stuff, then let the simlutor run for the equivalent of billions of years and intelligent beings should evolve.

Real intelligence cannot exist unless it is in context.



EDIT: Infinite Recursion hardly voiced opinions about future advancements in AI, he just said it is possible.

BinarySurfer

That's enough for me. Infinite Recursion is the only one to voice an opinion on this argument. The possibility statement is enough.

kruptof

the intelligence part of the things is the issue for me, what do mean by intellegence:

crunching large numbers in milliseconds, playing chess, predicting events given some random (or not so random ) is intellegence, but getting the shopping and having a connversation would not be considered intellegent, or is intellegence mimicing human behaviour or such as being able to make decisions which a toddler can make. well a person who can have a conversation get the shopping and decide when he is feeling pain but cannot do any mathical stuff or other things that us humans consider to be smart, would not be considered as intellegent.

__________________

Infinite Recursion

Thank you BinarySurfer.

In case I was not clear, I was not comparing AI with stealth technology, merely pointing out the potential secrecy of projects involved in such areas prior to being released to the public's knowledge (unclassified).

There are several approaches to the development of AI. Yours seems to be geared towards evolution, which I would agree the world/environment would need to be defined. However, this could be a predefined rule set given to the AI from which it has the ability to learn, reason, comprehend, solve problems, form its own conclusions, or adapt by adjusting its methods based on information derived from prior trials-and-errors or other scenarios.

I agree with your statement of "no computer will ever have enough processing power/memory". A single computer would never be able to do this by itself, especially if failover is a factor. Think in terms of parallel processing, high performance / distributed computing etc... Imagine a Beowulf Cluster of supercomputers which each box processing >1 trillion calculations a second.

For reference, Check out the Cray XT4.
http://www.cray.com/downloads/Cray_XT4_Datasheet.pdf

In the end... necessity, inspires invention.

__________________
http://jasonpowers.net

"There are a thousand hacking at the branches of evil to one who is striking at the root."

DaWei

AI is perhaps a misnomer. The underlying intelligence lies in the application of technology at the behest of the intelligent applier. That said, the advances have been great.

One example is in OCR (and other pattern-recognition tasks). The ability of your $100 printer (and its software) exceeds greatly the abilities we were able to offer the USPS in the mid-80s, even with million-dollar machines.

Another area is gaming. Backgammon AI, for instance, using neural networks, plays at world-class level. I'm sure you're more aware of those kinds of abilities in chess, as it gets more exposure.

If Duck's definition of "AI" is some entity that duplicates human-like abilities without outside intervention, then he should say so. It would greatly restrict the area of discussion, leaving him in a more tenable position. His idea of "AI" seems to be one of replicating God, an endeavor I would term "silly".

__________________
Abstraction doesn't make it impossible to write bad code; it makes it possible to write superior code.
Contributor's Corner: Grumpy on C++ Exceptions DaWei on Pointers

Duck

There isn't really a solid definition of intelligence, ask google, AI definition etc.

But for a computer to be intelligent (by my definition) it needs to be able to learn, and make independent decisions.

You see as DaWei explained (earlier), computers can only follow predefined operations/routines, it cannot learn new ones. Eg, if X is input then output Y, well what if Q was input...etc. Following predefined routines involves no intelligence on the computers part, and leaves the computer with extremely limited abilities.

DaWei

Actually, computers can learn by experience. That means the acquisition of new numbers to crunch, said numbers being acquired by self-evaluation of one's performance, as measured by suitable criteria. It's something successful humans do, too. For a very simple example, see my Tic-Tac-Toe game. For a much more complex example, see the aforementioned Backgammon AI. The backgammon AI is so successful that it has caused masters to reevaluate their thinking on some of the opening-roll moves that were held nearly sacred for decades. The machine thought differently; the machine has been proved to be correct. The foundation of the machine's ability was laid by its builders. From there, it is free to depart, within those foundational constraints.

__________________
Abstraction doesn't make it impossible to write bad code; it makes it possible to write superior code.
Contributor's Corner: Grumpy on C++ Exceptions DaWei on Pointers

Duck

Ok appologies.

I'm sure you know how stated things work, but...

OCR basically works by comparing the character with hundreads of variations of each character until a match is found - or not found as is often the case.

Board games there are a finite amount of moves therefore everyone can be tried until one which 'works' is found.

Neural networks only tune a predefined operation, they do not facilitate the learning of new operations.

But of course depending on how you define intelligence these things can be classed as examples of limited intelligence.

Duck

No computers can't learn by experience. They can tune themselves, as I mentioned previously (neural networks), BUT they can only tune themselves on a 'test set' of data that a human being tells them is good, or a 'digital pre-definition' of the end result, eg, in your Tic-Tac-Toe game you may have defined the conditions for win and loose, therefore if the cpu loses it could change its perfered moves for the next game based on that predefined outcome.

Therefore they can't even tune themselves without pre-definitions.

Therefore a computer can't learn, which is a fundermental requirement of being intelligent (by my definition).