Google AI Project

[1EZduzit]

Originally posted by: Moonbeam
In 2007 IBM will have a computer with 1/16 the computing capacity of the human brain.

My cheapo Texas Instrument calculator already has more computing power then my brain. Of course, I have the better sense of humor.

 

[JEDIYoda]

Originally posted by: CraigOley

Originally posted by: JEDIYoda

Originally posted by: CraigOley

Originally posted by: JEDIYoda
okay so you guys adreess this nonesense while I go out and have my dylithium Crsytals re-energized!!

"nonsense"? Try researching something before you say it is not true.

Actually within the parameters of realistically evolving into some sort of AI is just bogus and at best absurd.

All absurd conjecture with no basis for fact as of yet....

If I am wrong correct me using facts that are from reliable sources....

Thank You!!

Hmm. Have you ever played a video game? They have pretty good AIs there...

Actually I have never played anything close to a bot in a video game that learns to play you....
Video games is a horrible example and what you would call AI is not even close to being true.
Those bots are programmed to respond to scenarios that have been prethought out and pre programmed in to the game itself....

Sorry bad example Craig....

 

[CraigOley]

Maybe you should define your idea of AI because it is a lot different than the one taught in most US Computer Science programs.

 

[Meuge]

Originally posted by: Moonbeam
In 2007 IBM will have a computer with 1/16 the computing capacity of the human brain.

I wonder how they estimated it.

Oh, yes, I know - by pulling the figure out of their asses.

The brain is very far from being a digital computer equivalent. It's less like a CPU, and more like a GPU... with hardwired pathways, which can perform certain operations with extreme speed... interconnected under a master clock.

The master clock of the human brain operates at 20-60Hz, depending on person and circumstances... so I think that computers have surpassed that a long time ago.

When you show me a computer that can recognize a chair it hasn't seen before as being a generic "chair"... when a computer can perceive the "gestalt" of the information it can absorb, we can start comparing it to the human brain. In the meantime, our digital computers are little more than glorified abaci.

 

[Dissipate]

I'm sorry but as a computer science major I am very skeptical.

Computers at this point are purely mechanical devices. You plug in A, B, C and D and you get out E. Even seeminlgy complex, fancy shmancy object oriented programming languages are really in the end just performing extremely simplistic operations.

I didn't know the true extent of the simplicity of computers until I took a class in assembly. It was at that point I realized how ridiculously simple each instruction the computer processes really is.

Advances in computer science now (as they have always been) are basically mathematical advances. Every algorithm that a computer runs can be defined and explored in a mathematical manner (especially since the instruction set is finite).

In other words, any true advances in software are going to be made at the purely mathematical level. That is why they call it software engineering. A software engineer is just taking algorithms that have been known about for a long time and just tweaking/customizing them and putting them in the right order to get something useful out.

If you are going to invent your own algorithm you might as well get your chalkboard out because that is an exercise in mathematics. For instance, if I am going to develop a new encryption algorithm then that is going to be all math. Hence, any AI that is going to be produced on a standard computer is going to have to be within the bounds of mathematics. Does AI exist within pure math? If it did, that would be one hell of an equation. I think the most interesting investigations along those lines would be Conway's Game of Life and other celluar automatons. But those investigations are still extremely simplistic simulations, and calling the results 'life' is really stretching it.

I'm sorry to say but you simply cannot produce strong AI with today's old fashioned computing machines. That would be like trying to write a masterpiece novel using only a couple words. Any strong AI that is going to be developed is going to have to seriously diverge from the standard computing methodology.

 

[rchiu]

Originally posted by: Dissipate
I'm sorry but as a computer science major I am very skeptical.

Computers at this point are purely mechanical devices. You plug in A, B, C and D and you get out E. Even seeminlgy complex, fancy shmancy object oriented programming languages are really in the end just performing extremely simplistic operations.

I didn't know the true extent of the simplicity of computers until I took a class in assembly. It was at that point I realized how ridiculously simple each instruction the computer processes really is.

Advances in computer science now (as they have always been) are basically mathematical advances. Every algorithm that a computer runs can be defined and explored in a mathematical manner (especially since the instruction set is finite).

In other words, any true advances in software are going to be made at the purely mathematical level. That is why they call it software engineering. A software engineer is just taking algorithms that have been known about for a long time and just tweaking/customizing them and putting them in the right order to get something useful out.

If you are going to invent your own algorithm you might as well get your chalkboard out because that is an exercise in mathematics. For instance, if I am going to develop a new encryption algorithm then that is going to be all math. Hence, any AI that is going to be produced on a standard computer is going to have to be within the bounds of mathematics. Does AI exist within pure math? If it did, that would be one hell of an equation. I think the most interesting investigations along those lines would be Conway's Game of Life and other celluar automatons. But those investigations are still extremely simplistic simulations, and calling the results 'life' is really stretching it.

I'm sorry to say but you simply cannot produce strong AI with today's old fashioned computing machines. That would be like trying to write a masterpiece novel using only a couple words. Any strong AI that is going to be developed is going to have to seriously diverge from the standard computing methodology.

Well, somebody did created algorithm good enough to beat world chess champion, and that was more than 8 years ago. There are certain tasks that computer can be very good at, such as making decision on things with limited scenarios and outcome, even if there can be millions of scenarios and outcome, as long as they are certain, computer can reach the conclusion with its computing power. The question is, can someone good enough to turn all events that human make decision on into logical decision trees with scenarios and outcome that computer can logically determine. If someone/team/company can do that, it isn't impossible for computer to think like human.

 

[Dissipate]

Originally posted by: rchiu
Well, somebody did created algorithm good enough to beat world chess champion, and that was more than 8 years ago. There are certain tasks that computer can be very good at, such as making decision on things with limited scenarios and outcome, even if there can be millions of scenarios and outcome, as long as they are certain, computer can reach the conclusion with its computing power.

There was nothing 'intelligent' about how the machine beat Kasparov. It was an entirely mathematically based algorithm that merely used trees and compared certain outcomes to other outcomes using numerical weights. This process may have been very complex and it may have been tweaked by expert engineers for months, but in the end it was a process that can be described perfectly on a piece of paper (even if it would be a long piece of paper) i.e. you could print it out and look at each and every instruction the computer executed. Nothing in the realm of known mathematics was breached in that match.

On the other hand, if you look at how humans approach the game of Chess it is much more elegant. Human grandmasters actually only examine a few moves called 'candidate moves.' These moves pop out at them instantly as they glance at the board. This is in stark contrast to a computer chess engine that will examine moves and lines of play that are complete garbage. The reason why the human grandmaster can do this is because he looks at the patterns of a position. The elements of the position are probably ones that he has seen a million times before and so he knows exactly what moves might potentially be good. Then he examines only a few lines of play. A computer might look at thousands of lines for any given position.

The difference? The computer runs on brute force. The human runs on ideas or tactical patterns if you will. These patterns cannot currently be described mathematically. And certainly not in a way that a computer could process.

The question is, can someone good enough to turn all events that human make decision on into logical decision trees with scenarios and outcome that computer can logically determine.

Not today's computers. Human decision making is based on the meaning of an event. For instance: what does it mean when the car in front of me crashes into a tree? Well it could mean that the people inside are severely injured or are dead. How could a computer possibly process the meaning of severe injury or death? All it can do is manipulate 1s and 0s in a completely rudimentary fashion. So unless you can show me a mathematical formula that can perfectly describe the meaning of injury or death, no computer in the world will be able to ever do so. A computer's decision making is not based on anything except how it was programmed before the event occurred.

If someone/team/company can do that, it isn't impossible for computer to think like human.

It is my belief that it is certainly impossible for today's computers to think like a human. Although, that does not mean that a computer that was designed in a manner similar to the human mind could truly become intelligent. But until that kind of computer comes off the assembly line we are basically left with merely the tools of math.

 

[Moonbeam]

IBM's "Blue Gene" supercomputer provides 1 million billion calculations per second (i.e., one billion megaflops). This is already one twentieth of the capacity of the human brain, which I estimate at a conservatively high 20 million billion calculations per second (100 billion neurons times 1,000 connections per neuron times 200 calculations per second per connection).

Thus supercomputers will achieve one human brain capacity by 2010, and personal computers will do so by around 2020. By 2030, it will take a village of human brains (around a thousand) to match $1000 of computing. By 2050, $1000 of computing will equal the processing power of all human brains on Earth.

Your chair request will come soon enough. Already AI is deeply integrated into our modern world.

 

[Moonbeam]

Originally posted by: Dissipate

Originally posted by: rchiu
Well, somebody did created algorithm good enough to beat world chess champion, and that was more than 8 years ago. There are certain tasks that computer can be very good at, such as making decision on things with limited scenarios and outcome, even if there can be millions of scenarios and outcome, as long as they are certain, computer can reach the conclusion with its computing power.

There was nothing 'intelligent' about how the machine beat Kasparov. It was an entirely mathematically based algorithm that merely used trees and compared certain outcomes to other outcomes using numerical weights. This process may have been very complex and it may have been tweaked by expert engineers for months, but in the end it was a process that can be described perfectly on a piece of paper (even if it would be a long piece of paper) i.e. you could print it out and look at each and every instruction the computer executed. Nothing in the realm of known mathematics was breached in that match.

On the other hand, if you look at how humans approach the game of Chess it is much more elegant. Human grandmasters actually only examine a few moves called 'candidate moves.' These moves pop out at them instantly as they glance at the board. This is in stark contrast to a computer chess engine that will examine moves and lines of play that are complete garbage. The reason why the human grandmaster can do this is because he looks at the patterns of a position. The elements of the position are probably ones that he has seen a million times before and so he knows exactly what moves might potentially be good. Then he examines only a few lines of play. A computer might look at thousands of lines for any given position.

The difference? The computer runs on brute force. The human runs on ideas or tactical patterns if you will. These patterns cannot currently be described mathematically. And certainly not in a way that a computer could process.

The question is, can someone good enough to turn all events that human make decision on into logical decision trees with scenarios and outcome that computer can logically determine.

Not today's computers. Human decision making is based on the meaning of an event. For instance: what does it mean when the car in front of me crashes into a tree? Well it could mean that the people inside are severely injured or are dead. How could a computer possibly process the meaning of severe injury or death? All it can do is manipulate 1s and 0s in a completely rudimentary fashion. So unless you can show me a mathematical formula that can perfectly describe the meaning of injury or death, no computer in the world will be able to ever do so. A computer's decision making is not based on anything except how it was programmed before the event occurred.

If someone/team/company can do that, it isn't impossible for computer to think like human.

It is my belief that it is certainly impossible for today's computers to think like a human. Although, that does not mean that a computer that was designed in a manner similar to the human mind could truly become intelligent. But until that kind of computer comes off the assembly line we are basically left with merely the tools of math.

Yes, we have to remember that human intelligence is possible since a biological computer with human intelligence already exists. We are a program of some kind, but our consciousness is not of the software or the program. That runs in the background, no, even though it is doubtless mathematical in nature at some level.

 

[Dissipate]

Originally posted by: Moonbeam
IBM's "Blue Gene" supercomputer provides 1 million billion calculations per second (i.e., one billion megaflops). This is already one twentieth of the capacity of the human brain, which I estimate at a conservatively high 20 million billion calculations per second (100 billion neurons times 1,000 connections per neuron times 200 calculations per second per connection).

Thus supercomputers will achieve one human brain capacity by 2010, and personal computers will do so by around 2020. By 2030, it will take a village of human brains (around a thousand) to match $1000 of computing. By 2050, $1000 of computing will equal the processing power of all human brains on Earth.

Too bad speed is only a fraction of the equation. As I explained above, there are processes in the brain that are clearly beyond the scope of merely the manipulation of 1s and 0s.

 

[Dissipate]

Originally posted by: Moonbeam

Yes, we have to remember that human intelligence is possible since a biological computer with human intelligence already exists. We are a program of some kind, but our consciousness is not of the software or the program. That runs in the background, no, even though it is doubtless mathematical in nature at some level.

I agree except who knows if you can even call what goes on in the mind a 'program.' A program is a series of a finite number of instructions. Who knows if you could even call the processes in the mind 'instructions.'

Whatever goes on in the human mind transcends our current mathematical understanding of the world. In other words, whatever is going on up there is way beyond even the most complex computer algorithm ever devised. It's on some untapped paradigm way beyond our current grasp.

 

[Moonbeam]

Originally posted by: Dissipate

Originally posted by: Moonbeam
IBM's "Blue Gene" supercomputer provides 1 million billion calculations per second (i.e., one billion megaflops). This is already one twentieth of the capacity of the human brain, which I estimate at a conservatively high 20 million billion calculations per second (100 billion neurons times 1,000 connections per neuron times 200 calculations per second per connection).

Thus supercomputers will achieve one human brain capacity by 2010, and personal computers will do so by around 2020. By 2030, it will take a village of human brains (around a thousand) to match $1000 of computing. By 2050, $1000 of computing will equal the processing power of all human brains on Earth.

Too bad speed is only a fraction of the equation. As I explained above, there are processes in the brain that are clearly beyond the scope of merely the manipulation of 1s and 0s.

Are you saying the brain is not a digital computer? I thought a synapse either fires or does not.

 

[Dissipate]

Originally posted by: Moonbeam

Are you saying the brain is not a digital computer? I thought a synapse either fires or does not.

No, it is not a digital computer. A digital computer is a purely mechanical device that performs extremely simplistic operations using 1s and 0s in logical gates. Everything inside a computer and everything that occurs within a computer can be described mathematically so as to not leave a single shred of doubt about what the machine is doing.

Perhaps a better way of saying this is that you will never run into an algorithm in a digital computer that baffles the experts as to how the computer got the result that it did. This is not to say that you will not get surprising results initially. That happens all the time in computer science. What you will not get is perpetually baffling results.

For instance if I plug A, B, and C into algorithm X and I get some unexpected result, that unexpected result will not be unexpected for long. I can go back and trace the precise instructions the computer executed to produce that result. And each instruction the computer executed is so babyishly simple there will not be a shadow of a doubt as to exactly what happened that caused the computer to spit out what it did.

The brain's architecture is entirely different. It uses interconnected neurons and weights. The complexity of the information processing within this system is most definately boggling. These really are 'black box' perpetually baffling 'algorithms.'

If a computer were to ever reach the stage of true intelligence it would probably use a similar architecture. But like I said, whatever is going on within this 'bio-computer' is beyond the scope of 100% understood/described algorithms we use every single day on digital computers.

 

[MCsommerreid]

I belive for there to be a self aware computer there would have to be near unlimited storage, as well as a algortithem giving the system the ability to learn and write its own code to allow it to use what it's learned. Once you have that there is no limit to what can happen.

 

[piasabird]

Hmm. . . .

Dave? What are you doing Dave?

Google is trying to become HAL like on Space Odessy 2020 or whatever that movie was when the computer takes over the ship.

When a company tries to take over the search engines and marketing for the Internet, it actually seeks to shape everything from politics to freedom of speech. How is Google any better or nicer than say Bill Gates?

It is like Big Brother on the Internet.

 

[KevinF]

Regardless of whether or not you think a self-aware computer will come to exist or not, Google AI will certainly come to exist in some form.

Google was able to win a machine translation contest using the statistical approach. They fed thousands of documents already translated from Arabic to English and vice versa into a machine and it was able to make statistical conclusions and translate better than everyone else.

The idea of Google AI is a big "what if" question based on Google having access to every book ever written. If the world's best AI researchers can translate arbitrary documents with enough documents about translation, what can they do with the sum total of human knowledge?

 

[Moonbeam]

Originally posted by: Dissipate

Originally posted by: Moonbeam

Are you saying the brain is not a digital computer? I thought a synapse either fires or does not.

No, it is not a digital computer. A digital computer is a purely mechanical device that performs extremely simplistic operations using 1s and 0s in logical gates. Everything inside a computer and everything that occurs within a computer can be described mathematically so as to not leave a single shred of doubt about what the machine is doing.

Perhaps a better way of saying this is that you will never run into an algorithm in a digital computer that baffles the experts as to how the computer got the result that it did. This is not to say that you will not get surprising results initially. That happens all the time in computer science. What you will not get is perpetually baffling results.

For instance if I plug A, B, and C into algorithm X and I get some unexpected result, that unexpected result will not be unexpected for long. I can go back and trace the precise instructions the computer executed to produce that result. And each instruction the computer executed is so babyishly simple there will not be a shadow of a doubt as to exactly what happened that caused the computer to spit out what it did.

The brain's architecture is entirely different. It uses interconnected neurons and weights. The complexity of the information processing within this system is most definately boggling. These really are 'black box' perpetually baffling 'algorithms.'

If a computer were to ever reach the stage of true intelligence it would probably use a similar architecture. But like I said, whatever is going on within this 'bio-computer' is beyond the scope of 100% understood/described algorithms we use every single day on digital computers.

I don't know how many machine language instructions are used in computers, but the human brain is built from only 4, adenine guanine thymine and cytosine. I believe there are computations as in the game of Life where the results can only be known by actually running the program. They can be known in no other way. To walk backward from where you've been is different than knowing where you are going.

 

[Wiktor]

Originally posted by: JackStorm
I, for one, welcome our new google AI overlords. That is, as long as they don't tell me how to live my life.

You'd have to be blind not to notice that computers already have a huge impact on how we live our lives. Just one example: the courses for my studies are planned by a computer - it will tell me when and where I have to go each day for the next semester.

I actually think Google is a very knowlagable being, one I learn a lot from each day I am also impressed at how intelligently it can make a profit.

 

[1EZduzit]

I wonder what it would be like to have to deal with an AI that was in a bad mood?

 

[biggestmuff]

why are they wasting time with this? Haven't they heard of Dr. Sbaitso?

 

[Train]

Originally posted by: CraigOley
How long (after multi-tasking 124 processors on one chip come out) do you think it will take for people to start being able to program computers that are similar to the concious mind?

The strength of the hardware will have nothing to do with it. AI will exist in a purely virtual state.

 

[Meuge]

Originally posted by: Moonbeam

Originally posted by: Dissipate

Originally posted by: Moonbeam

Are you saying the brain is not a digital computer? I thought a synapse either fires or does not.

No, it is not a digital computer. A digital computer is a purely mechanical device that performs extremely simplistic operations using 1s and 0s in logical gates. Everything inside a computer and everything that occurs within a computer can be described mathematically so as to not leave a single shred of doubt about what the machine is doing.

Perhaps a better way of saying this is that you will never run into an algorithm in a digital computer that baffles the experts as to how the computer got the result that it did. This is not to say that you will not get surprising results initially. That happens all the time in computer science. What you will not get is perpetually baffling results.

For instance if I plug A, B, and C into algorithm X and I get some unexpected result, that unexpected result will not be unexpected for long. I can go back and trace the precise instructions the computer executed to produce that result. And each instruction the computer executed is so babyishly simple there will not be a shadow of a doubt as to exactly what happened that caused the computer to spit out what it did.

The brain's architecture is entirely different. It uses interconnected neurons and weights. The complexity of the information processing within this system is most definately boggling. These really are 'black box' perpetually baffling 'algorithms.'

If a computer were to ever reach the stage of true intelligence it would probably use a similar architecture. But like I said, whatever is going on within this 'bio-computer' is beyond the scope of 100% understood/described algorithms we use every single day on digital computers.

I don't know how many machine language instructions are used in computers, but the human brain is built from only 4, adenine guanine thymine and cytosine. I believe there are computations as in the game of Life where the results can only be known by actually running the program. They can be known in no other way. To walk backward from where you've been is different than knowing where you are going.

More total gibberish. It may sound like it's clever and insightful to you, but in essence it's utter cow poop.

 

[shira]

Originally posted by: CraigOley
Any thoughts or further information/insight?

The nice thing about an AI capability is that you'll be able to pose questions like:

"Provide an objective analysis of the tax policies of George W Bush and their likely consequences."

"Is invading Iraq warranted based on currently available information?"

"Is warrantless surveillance of phone calls having one end of the connection within the U.S. a criminal act?"

And so on.

In other words, in the age of semi-omniscient AI, it will become more and more difficult for politicians to skew the truth with Rovesque propaganda.

 

[Moonbeam]

Originally posted by: Meuge

Originally posted by: Moonbeam

Originally posted by: Dissipate

Originally posted by: Moonbeam

Are you saying the brain is not a digital computer? I thought a synapse either fires or does not.

No, it is not a digital computer. A digital computer is a purely mechanical device that performs extremely simplistic operations using 1s and 0s in logical gates. Everything inside a computer and everything that occurs within a computer can be described mathematically so as to not leave a single shred of doubt about what the machine is doing.

Perhaps a better way of saying this is that you will never run into an algorithm in a digital computer that baffles the experts as to how the computer got the result that it did. This is not to say that you will not get surprising results initially. That happens all the time in computer science. What you will not get is perpetually baffling results.

For instance if I plug A, B, and C into algorithm X and I get some unexpected result, that unexpected result will not be unexpected for long. I can go back and trace the precise instructions the computer executed to produce that result. And each instruction the computer executed is so babyishly simple there will not be a shadow of a doubt as to exactly what happened that caused the computer to spit out what it did.

The brain's architecture is entirely different. It uses interconnected neurons and weights. The complexity of the information processing within this system is most definately boggling. These really are 'black box' perpetually baffling 'algorithms.'

If a computer were to ever reach the stage of true intelligence it would probably use a similar architecture. But like I said, whatever is going on within this 'bio-computer' is beyond the scope of 100% understood/described algorithms we use every single day on digital computers.

I don't know how many machine language instructions are used in computers, but the human brain is built from only 4, adenine guanine thymine and cytosine. I believe there are computations as in the game of Life where the results can only be known by actually running the program. They can be known in no other way. To walk backward from where you've been is different than knowing where you are going.

More total gibberish. It may sound like it's clever and insightful to you, but in essence it's utter cow poop.

Hehe, I think I can see what you know about what is essence by your grasp of the essentials of intelligent rebuttal. Please argue your case or use a toilet. Your dumps only serve to fertilize my garden.

 

[Moonbeam]

Originally posted by: shira

Originally posted by: CraigOley
Any thoughts or further information/insight?

The nice thing about an AI capability is that you'll be able to pose questions like:

"Provide an objective analysis of the tax policies of George W Bush and their likely consequences."

"Is invading Iraq warranted based on currently available information?"

"Is warrantless surveillance of phone calls having one end of the connection within the U.S. a criminal act?"

And so on.

In other words, in the age of semi-omniscient AI, it will become more and more difficult for politicians to skew the truth with Rovesque propaganda.

I am not so sure about this, shira. In my opinion, not worth much I know, you are an intelligent and cogent thinker whose analysis I find pretty objective. But how many people pay any attention to you. I believe it not a lack of truth that blinds people, but an active bias against seeing. "A fool convinced against his will is of the same opinion still." This applies to all manner of people in every facet of intellectual development. The will to know real truth, in my opinion is quite rare and comes from a deep and hidden place.

An artificial intelligence will arrive at objectivity, I think, by means most won't follow. This inability to comprehend due to bias will preserve the bias in place. Objectivity is a function of vision and a willingness to see. One without the other will be useless, I think, until the costs of being wrong become a greater motivator than the bias.

But then you did say more and more difficult. I hope so.