Digg

The Digg Crew wants to hear your thoughts!
Please take our short survey about Digg and potential feature ideas.

Pursuing the Next Level of Artificial Intelligence

nytimes.com — Like a good gambler, Daphne Koller, a researcher at Stanford whose work has led to advances in artificial intelligence, sees the world as a web of probabilities.

Bury

maheshee11 212 days ago, made popular 210 days ago

360 diggs
digg it

28 Comments
Who Dugg It?

swatward, on 05/05/2008, -5/+1AI doesn't seem possible at our levels of programming. Something can follow a series of complex instructions, but that's not really artificial intelligence.
- Kamill85, on 05/05/2008, -2/+6I disagree, it's possible if this basic set of instructions is able to create new ones, randomly, to serve its own real or imagined needs.
  - veloscaper, on 05/05/2008, -6/+3you really don't know anything about computers. There is no real random.
    - Kamill85, on 05/05/2008, -1/+6Human brain is a computer too, so if real random doesn't exist, our brain does not have it either.
      
      On the side note: I was talking about software part of the problem, hardware solutions such as random generators etc. are another story.
      - veloscaper, on 05/05/2008, -6/+2our brains are not a computer.
        
        random number generators is software, and btw they are pseudo random.
      - Awap, on 05/05/2008, -1/+5There are hardware random number generators. Basically they take noise from a naturally chaotic source and feed it through algorithms to get the random distribution they want. For instance, a CCD that is in a completely dark enclosure will give output that is almost completely random. Feeding that through a hash algorithm will give you a nice big number that is honest-to-god random, not pseudorandom.
        
        Many high end computers have such a device built in for doing things like creating cryptographic keys and one-time nonces. SGI even demonstrated a system that took digital video of multiple lava lamps as its entropy source. Even on a commodity computer /dev/random takes entropy from things like bus timings and other external events, so that the numbers don't follow any pseudorandom pattern. /dev/urandom gives unblocking random numbers; when the entropy runs low, it reverts to a pseudorandom pattern rather than blocking for more entropy. This works on any unix-like system, including a mac. I would guess that there is an equivalent on windows too.
      - iiiears, on 05/05/2008, -0/+2Awap don't confuse us with information this is digg.
        /sar
    - Godlike, on 05/05/2008, -0/+2Games like WoW buy their random numbers and use a tolerance.
      
      For example, in WoW, if you pull 4 numbers in a row that are less than 20% of the possible maximum result of the roll then the game rerolls the 4th number until it is above 20% of the possible total. If it so happens that the next 3 results are less than 40% of the maximum result then the game rerolls until you are within 60% of the maximum. A single result within the tolerance ends the entire process.
      
      While it may not be possible to have truly random numbers, you can have sets of numbers and rules within a system that make things both random and sane.
      
      You might say it's still not truly random, but when you apply random things to any system they are no longer just random. You have to work within any system regardless if it is physical reality, a computer game, or an artificial intelligence everything is bound by a system that can be guided to generate fairness or sanity.
    - Culyt, on 05/05/2008, -0/+4Actually there is real random, there is a website using quantum random number generator. Thats about as random as you will ever get in the universe, if its not then there is no 'real random' anywhere and its just a theoretical concept.
      
      With that said, you don't need 'real random' to randomly generate intelligence, and decent psudo-random should be good enough. The main problem with psudo-random number generators is that you can guess in advance the next number by knowing the current number and the mathematical formula, this is a problem for encryption and things like gambling this isn't a problem for evolutionary programming. evolutionary programming also uses brute force so bad numbers will simply be ignored. In fact you could probably evolve using only brute force, but it would take a whole lot longer.
  - Godlike, on 05/05/2008, -0/+1The big thing here is that it shouldn't ever be random. A truly intelligent being can make a choice based on any factors; we should be looking at the base ideology that would guide such a creature and not trying to fake the seeming of intellect with chaos.
- JQP123, on 05/05/2008, -2/+5"AI doesn't seem possible at our levels of programming."
  
  It's not just the "level of programming". It's the fundamental design of what we call a "computer". All available evidence suggests that "intelligence" is something more than a binary logic algorithm.
  
  The computer as we know it is just an inert tool --- a binary logic record and playback device. Recently, it was shown that a simple, general purpose, binary logic computer capable of solving any problem can be built using only 6 mechanical switches and possibly driven by a hand crank. Such a mechanical device would obviously operate very slowly but in theory, it is just as "intelligent" as any modern computer since it is capable of running the same binary logic program.
  
  http://digg.com/hardware/Simple_Turing_machine_sho ...
  - Genady, on 05/05/2008, -1/+2Actually the innovation of the Memristor will change the idea of a microprocessor. If Memristor really does lead to a more analog microprocessor then I'm more inclined to think that AI can arise.
    - JQP123, on 05/05/2008, -1/+2An analog computer would definitely help ... mainly because it offers a better approximation of the real world. By design, a binary digital computer only understands absolutes. Everything must be either "black" or "white" whereas in the real world, most things are actually some relativistic analog shade of gray.
      
      Still, I think a real intelligent computer will have to be somehow based on organic chemistry. A brain in a box if you will --- just feed it, provide some input/stimuli and watch it learn and grow.
      - intangible, on 05/05/2008, -0/+1Sounds like a science fair project end for my pet monkey.
  - NanoStuff, on 05/05/2008, -1/+1"All available evidence suggests that "intelligence" is something more than a binary logic algorithm."
    __
    
    There must be some crosstalk from your parallel universe into mine. Over here, all available evidence suggests the opposite.
    
    A more popular article in this regard is with is http://www.seedmagazine.com/news/2008/03/out_of_th ...
    
    I quote "Consciousness is a binary code", near the bottom of the page. From the Blue Brain director, the largest computational brain project.
    
    "but in theory, it is just as "intelligent" as any modern computer since it is capable of running the same binary logic program."
    __
    
    That's a bad theory. Intelligence is not the sum of computational capability but the way in which it is used. Computers can be as intelligent and otherwise thoughtful as humans, and much more so after that, and any Turing-complete programming language is capable of supporting the necessary task. The problem is entirely the software.
  - Culyt, on 05/05/2008, -0/+2EVERYTHING is binary, or rather can be expressed in binary. Taking an analog wave and converting it to binary might incur some loss, but it doesn't need to be a lot, as long as it isn't a very bad conversion it should be adequate. Think of scanning a painting with a high resolution scanner, the original painting was painted by a person who couldn't possibly pain at 1200 dpi or even higher, its impossible. Instead he makes brush strokes that are viewed from a foot or more away, he makes something closer to a vector image rather than a bitmap one. The high resolution scan can pick up specs of dust, cracks in the paint on the paints, even atoms if you where to use an atomic force microscope. In a similar way your brain could actually be scanned on the atomic level and simulated on it, rather than being an AI simulation it would be the worlds most advanced physics simulation. But from what we know of intelligence this would be overkill, there is nothing that is not really understood about how the brain functions, we don't know the overall picture but we know enough about the details that its not such a problem, we know how nurons fire, we don't know why when you have enough of them you get intelligence or sentience.
    
    The blue brain project recently took a rats brain, choose a neocortex, poked each nuron in that neocortex with a probe, took the electric charge of each nuron, maped the nuron connections (thousands of them for each nuron, millions of neurons) into a large specialized super computer and ran a successful simulation of the rats neocortex, one of the primary building blocks of intelligence. They think they will have a simulation of whole rats brain within 2 years and they are planning to link it up to a robotic rat for the hell of it. They will have enough computer power to do a whole person in 10 years. And shortly after that the computer power will be on a single desktop processor. When we have that level of computing power we don't even need to understand the science behind it, we can just randomly generate computer code until it responds correctly and perhapses what comes out will be able to tell us how it works (or kill us all). Right now we program neural networks and evolve them, but we could just let the neural networks them selfs evolve, or maybe we will get something entirely different.
    - JQP123, on 05/05/2008, -0/+1"... but we could just let the neural networks them selfs evolve..."
      
      Only if you (or someone, a programmer) can specifiy in binary code exactly what "evolve" means and provide a detailed algorithm for exactly how it should be carried out. In other words, it's not really "self evolving", it's just a dumb computer blindly executing yet another algorithm.
      - NanoStuff, on 05/05/2008, -0/+1You really have to understand that there is no limit to the descriptive precision of algorithms. Any natural phenomena has an accompanying algorithm, whether or not we have yet figured out what it is. We could conceivably evolve intelligence from what you call a 'dumb algorithm' by old fashioned natural selection, but this isn't a realistic proposition given the requirements for processing power. Keep in mind that a digital processor is nevertheless no less analog than anything else in nature despite it's higher level binary operation, much like the brain.
      - JQP123, on 05/05/2008, -0/+1"You really have to understand that there is no limit to the descriptive precision of algorithms."
        
        Practical limits are imposed by external constraints such as time and money. Noone has ever produced even a moderately compelling example of "intelligence" using a binary logic computer. This doesn't prove that it can't be done ... but it certainly doesn't offer too much hope that we're on the verge of any sort of major breakthrough either.
      - NanoStuff, on 05/05/2008, -0/+1"This doesn't prove that it can't be done"
        
        Well then, there you go :)
        
        A compelling example of intelligence would be intelligence itself, so it cannot be said we're not making progress because we don't have intelligence. We are making much progress and eventually we will have it. The most direct and easiest route would be plain old brain scanning, the IBM approach. Since it would arguably be wrong to enslave virtual humans to perform our redundant tasks, the system's humanity could then be gutted to reduce it to some simpler residual intelligence. That could be the hard part and it's a very blurry line, however according to IBM, we won't have to wait very long to find out.
Shadowgamers, on 05/05/2008, -8/+3I knew it. It was the Jews who created the Cylons...
youtellme8, on 05/05/2008, -6/+1We should be careful. Artificial intelligence would be the end of us.
Futile, on 05/05/2008, -8/+1Judgment Day is near.
pixelperfect, on 05/05/2008, -0/+1I imagine if AI does come it will be from a "fluke", as in - not planned. How can one create something more complex than it's self? It would be hard enough to create something 90% as complex. This won't stop Hollywood from making a million more AI movies till the next millennium.
- Genady, on 05/05/2008, -1/+1Oh it's easy to create something more complex than oneself. First there's intercourse. 9 months or so after that you beat the child daily until it needs therapy. Much more complex.
spootmonkey, on 05/05/2008, -1/+1Why does everyone think HAL when they think of AI research? There is almost no serious research being done on Hard-AI because it has failed too often to get any results.
- Amablue, on 05/05/2008, -0/+2"failed" as in failed to work properly, or "failed" as in it destroyed the ship it was on?
caponumen, on 05/06/2008, -0/+1LOL, a very great probability of me pushing your real world bots down the stairs.
Login or register to add a comment

Create a new account or login to join in the conversation on Digg. You'll also be able to Digg stories to help promote things you like.

Pursuing the Next Level of Artificial Intelligence

Login or register to add a comment