When will computer hardware match the human brain?

[Carlos Hernandez]

When will computer hardware match the human brain?

(Received Dec. 1997)
Hans Moravec
Robotics Institute
Carnegie Mellon University
Pittsburgh, PA 15213-3890, USA
net: hpm@cmu.edu
web: http://www.frc.ri.cmu.edu/~hpm/


ABSTRACT

This paper describes how the performance of AI machines tends to improve at the same pace that AI researchers get access to faster hardware. The processing power and memory capacity necessary to match general intellectual performance of the human brain are estimated. Based on extrapolation of past trends and on examination of technologies under development, it is predicted that the required hardware will be available in cheap machines in the 2020s.

Complete text at http://www.jetpress.org/volume1/moravec.htm

 

[Ontoplankton]

Any opinions on how plausible all of that is, what the consequences will be if true, and so on?

In the same category but going a bit further is "How long before superintelligence?" by Nick Bostrom.

 

[fffbone]

The pure calculation speed of the computers have long exceeded the human brain, however, the "creativity" of the computers may take more than just a few decades to exceed the human brain.

 

[lnx990]

In an article in Byte magazine (April 1985), John Stevens compares the signal processing ability of the cells in the retina with that of the most sophisticated computer designed by man, the Cray supercomputer:

"While today's digital hardware is extremely impressive, it is clear that the human retina's real-time performance goes unchallenged. Actually, to simulate 10 milliseconds (one hundredth of a second) of the complete processing of even a single nerve cell from the retina would require the solution of about 500 simultaneous nonlinear differential equations 100 times and would take at least several minutes of processing time on a Cray supercomputer. Keeping in mind that there are 10 million or more such cells interacting with each other in complex ways, it would take a minimum of 100 years of Cray time to simulate what takes place in your eye many times every second."

 

[Lyuokdea]

Originally posted by fffbone
The pure calculation speed of the computers have long exceeded the human brain, however, the "creativity" of the computers may take more than just a few decades to exceed the human brain.

that really isn't true, while computers may be able to crunch numbers more effectively then a brain, the calculational capacity of a computer is much less than a brain, it is about the equivilent of a cockroach currently. The difference is parallel processing ability. Remember a computer can only do one calculation at a time. The brain, while it calculates slower, can do thousands of calculations simultaneosly. You can talk, walk, chew gum, breath, beat your heart, look at things, hear things, and do a hundred other things subconsciously all at the same time. I believe, that high speed supercomputers do something around

http://www.aeiveos.com/~bradbury/Authors/Computing/McEachern-RH/HaMIAEV.html

Looked at in another way if there are 1012 neurons in the brain, each performing a filtering process equivalent to performing one million arithmetic operations per second, that equals a total of one billion billion operations per second. Because of the limited dynamic range of the signals transmitted by neurons (not requiring more than about 8 bits to represent each measurement of a signal), these operations are somewhat simpler to perform than the "floating point" operations commonly used to characterize the computational power of computers. Nevertheless, the signal processing power of the brain, expressed in billions of floating point operations (gigaflops) is still on the order of 100 million gigaflops. That is millions of times more powerful than the fastest supercomputers yet built. But it is only about equal to the probable processing power of a personal computer several decades from now.

 

[Shadow]

Never.

 

[Jikx]

OK, the brain has billions of cells.. each cell may have between one and thousands of connections, constantly growing and other dying.

A transistor has three (?). And they can store.. 2 states? A brain cell has a state between 0 and 1.

So.. err.. never.

 

[selfAdjoint]

An individual neuron is a very complex little machine, and can support different states in different locations, and coordinate the different states inside the cell. It's more like a simple CPU than a bit switch.

 

[Moni]

Originally posted by Carlos Hernandez
When will computer hardware match the human brain?

(Received Dec. 1997)
Hans Moravec
Robotics Institute
Carnegie Mellon University
Pittsburgh, PA 15213-3890, USA
net: hpm@cmu.edu
web: http://www.frc.ri.cmu.edu/~hpm/


ABSTRACT

This paper describes how the performance of AI machines tends to improve at the same pace that AI researchers get access to faster hardware. The processing power and memory capacity necessary to match general intellectual performance of the human brain are estimated. Based on extrapolation of past trends and on examination of technologies under development, it is predicted that the required hardware will be available in cheap machines in the 2020s.

Complete text at http://www.jetpress.org/volume1/moravec.htm

I am not sure...but it's not possible :(
As my teacher told, as your eye can't see your self...thus your brain doesn't know how it works :(

 

[Elledan]

Originally posted by selfAdjoint
An individual neuron is a very complex little machine, and can support different states in different locations, and coordinate the different states inside the cell. It's more like a simple CPU than a bit switch.

Indeed. One of the best ways to describe a neural network is consider each neuron to be a small computer (it can also 'store' a number of variables). These 'computers' interact with each other in a strictly defined manner.

 

[Ivan Seeking]

When a computer first has an out of circuit experience and then creates its own religion.

My real guess; don't know. There is too much about the brain and potentially other related mechanisms to consider. For example, Chopra argues that the immune system is a part of our "intelligence". He describes it as a circulating intelligence. He is not the first to argue that intelligence is not limited strictly to the brain. Could a complete description of our mind require other factors not even considered yet?

Not my idea but I thought it was worth mentioning.

 

[Dagenais]

The brain is one of the most complicated things on Earth.

I vote never.

 

[Nice coder]

I vote soon!
(and transisters have more then 2 states, it is an analog devise-within limits, and its range is about half as much as a nauron-Push-pull anybody?)

 

[Averagesupernova]

A transistor may be an analog device but those in computers are designed to operate in 2 states. Until you design and perfect an analog computer, it will be more of the same old. Faster, more memory, etc. Until a system is self aware, it is not even comparable to the human brain. I seriously doubt that if you took all of the computer power that has EVER existed and networked it to the best of our abilities it wouldn't even resemble an incredibly crude human brain even at slow speeds.

 

[Nice coder]

The first computers were analog
But i think it would be more like bio-neural circutry (as on Star trek(little c thingy) Voyager)

 

[EMANUELI]

Any way! A human can simply design a computer by using his brain effort but a computer will never design a human at any means. That means a human brain is capable of doing much more than anything under the sun. Thats the reality and it will remain till then!

 

[Brennen]

When do you think we will make a computer that can learn and think for itself? a computer with free will? a computer with emotions?

do you think we will ever invent thinking, feeling, "alive" robots? eg AI

 

[maxvan]

I do think that the human brain can do more than that we do. Everything in the universe, the earth, planets, is bases on frequence, electromagnetic frequence, and like they call shakings. The way humans react, or every being, is instructed by the universe, by the sun, if the sun changes in magnetude, we beings in the universe depending the sensitifness, reacts different to what the universe instruct to do. Everything is based on light, frequence and shakings. just like electricity but in higher form, level.
The brain reacts on what it is told to do, if you go to higher level of shaking, thinking, frequency with the mind, the neuron will start getting scharper of level of sensitifness. will shake in whitch way it is instruct to do.

In the recent society, learning patern, the human mind is not instructed how it can evolved how to use the mind in a better way. the society gives on the informatie what is has been doing for the last 1000 of years to do. When you start acting with the mind in a different way, the mind will in collaboration with the nature make a different way how to deal with situations. So it will avoid things people don't want. Society, people think about money how to survive, to earn as much as possible to have a good live. that is really good, at stake of the planet place they live on.

 

[michinobu]

I don't know if it's mathematically possible. I remember reading in "Introduction to the Theory of Computation" by Michael Sipser, that Kurt Godel, Alan Turing, and Alonzo Church discovered that computers can't solve certain "basic" problems which are solvable to humans - such as being able to prove if a mathematical statement is true or false.

Scientists in the field of neurology know very little about the human brain, the very fact that humans aren't digital shows what kind of difficulties an engineer might face in trying to recreate the human brain.

Besides, isn't the very definition of "artificial intelligence" is intelligence being mimicked is intelligence? That if something acts intelligent then it is intelligent? I think when some people talk about a machine being as "smart" as a human they're talking about something that acts a lot like a human but is missing something somewhere.

 

[chaoseverlasting]

I don't think its possible. Humans can think original thoughts. We can create something absolutely unique which has no connection to anything around us. One example is chess. I think I read somewhere that the total number of possibilities for two or three consecutive moves is over 4 billion, but a grand master intuitively picks out the best ones.

Also, if you've read "Zen and the Art of Motorcycle Maintainence" by Robert Pirsig, then the arguments he brings up as to how we get ideas and think creatively are based not on mathematical calculations but more on intuition unique to every person. I don't know if I am being very clear, but the book's worth a read in any case. This question is more one of philosophy than anything else.

 

[CRGreathouse]

I'd say sometime between now and thirty years from now. The human brain will still have advantages then, though, especially in terms of redundancy. CPUs aren't expected to work if damaged, but the human brain has evolved to be able to take quite a pounding and still go on (damage like Alzheimer's and PTSD notwithstanding).

 

[CRGreathouse]

Also, if you've read "Zen and the Art of Motorcycle Maintainence" by Robert Pirsig, then the arguments he brings up as to how we get ideas and think creatively are based not on mathematical calculations but more on intuition unique to every person.

I agree, and interpret 'intuition' as 'heuristic'. We don't solve systems of nonlinear equatiosn exactly to determine how our 2D vision corresponds to a 3D world, but use shortcuts to come up with solutions that are usually close to being correct. Optical illusions and failed depth perception are examples of the limitations of this method.

 

[monotune]

I think when mentioning and comparing with analog and digital devices, one should consider that not even the potential and abilities of an axon counts but the quality and quantity of the chemicals released at the synaptic endings. So, it seems to be more complicated...The logic of digi works with 2 values, an analogous machine (if well controlled) might have more states, that is "levels of truth". Here a technical question arises; how well can we "level" (define and manipulate states) an analogous machine? Can we level it in a "quasi continuous way", or we can only define and work with finitely many states? Questions about the brain: how many "states" can a nerve cell have and how many can it generate in its surroundings? Can these states generate "new" states? The attempt to define the above mentioned concepts, or trying to compare these with the theoretical functioning af the brain is challenging.
When it comes to the very theory/philosophy-side of the question, I tend to sense a paradox.

 

[tormund]

don't they already? i mean that computers can compute billions of calculations per nanosecond. take tic tac toe, a computer plays millions of games in the span of about a second to logically determine the best move, and after a few games they learn your moves and modify their tactics. most people won't play that many games in a lifetime.

 

[AUMathTutor]

I think that it's only a matter of time before computers become as smart as people. Whatever that means. Perhaps when computers can do it, it won't be called "intelligence" anymore.

In a certain sense, computers are really already "intelligent"... years ago, computers were people who had real jobs crunching numbers. Nobody counts crunching numbers like that night and day a sign of intelligence anymore, since machines can do it. I think a fairly sizeable cross section of people will simply not be able to admit that what computers do is called "intelligent"... not now, not ever.

 

[Phrak]

When it acts like a verticle primate. Or did I deliberately miss the point, in that the OP was considering a small and insignificant portion of the cortex for comparison.

 

[mathlete]

Never. Also, humans will never do the following things:

- Harness fire
- Understand gravity
- Make a heavier than air object fly
- Put a man on the moon
- Cure diseases

These things are way too complicated. Never going to happen. Sorry.

But seriously, I can't believe people are voting never. Talk about short-sighted. There is nothing magic about the brain, amazing gift of nature that it is.

Oh well. We'll just add your quotes to these one day:

"The abdomen, the chest, and the brain will forever be shut from the
intrusion of the wise and humane surgeon".

--Sir John Eric Ericksen, British surgeon, appointed Surgeon-
Extraordinary to Queen Victoria 1873.
"Drill for oil? You mean drill into the ground to try and find oil?
You're crazy."

--Drillers who Edwin L. Drake tried to enlist to his project to drill
for oil in 1859.

"Heavier-than-air flying machines are impossible."

--Lord Kelvin, president, Royal Society, 1895.

"Everything that can be invented has been invented."

--Charles H. Duell, Commissioner, U.S. patent office, 1899

 

[daniel_i_l]

Never. Also, humans will never do the following things:
- Cure diseases

Actually, computers have already found some novel cures for example:
http://news.bio-medicine.org/biology-news-3/Computer-aided-protein-design-wins-prestigious-AAAS-Newcomb-Cleveland-Prize-12989-1/
and
http://news.bio-medicine.org/biology-news-3/Computer-aided-protein-design-wins-prestigious-AAAS-Newcomb-Cleveland-Prize-12989-1/
These were the best examples I could find off the top of my head but there have been many cases where a computer that analyzed a genetic sequence found ways to cure a disease that researchers hadn't anticipated.
Using advanced learning and data mining techniques computers have found novel solutions to problems in other field as well:
http://www.genetic-programming.com/humancompetitive.html

I think that computers will one day be able to compete with humans in every field that demands some form of intelligence. This doesn't mean that the computer will be able to simulate the brain. Many people in this post are forgetting the fact that almost no problem has one "correct" way of solving it. The computer might not be able to simulate everything the brain does - just like it's hard for it to simulate any complex physical system - but I think that it'll be able to solve the same problems using a different approach.

I once read (I forget where) that people define intelligence as "anything computers can't do". At first chess was the holy grail of AI, then learning, then ... religion?
I agree with most of what's written here:
http://web.media.mit.edu/~minsky/papers/ComputersCantThink.txt

...
It is too easy to say things like, "Computer can't do (xxx), because they
have no feelings, or thoughts". But here's a way to turn such sayings into
foolishness. Change them to read like this. "Computer can't do (xxx),
because all they can do is execute incredibly intricate processes, perhaps
millions at a time". Now, such objections seem less convincing -- yet all
we did was face one simple, complicated fact: we really don't yet know
what the limits of computers are. Now let's face the other simple fact: our
notions of the human mind are just as primitive.
...

 

[SW VandeCarr]

Does anyone here know what the definition of intelligence is? I don't think we can talk about machines matching the human brain unless we can specifiy in just what way machines will match the human brain. The brain is a product of biological evolution. It has a long heritage dating back at least 300 million years. The human brain contains reptilian structures such as the limbic system which influences how we think and what we think about. I don't believe we'll want to create a limbic system for our future supercomputers, even though it might play a role in our "creativity".

Computer based genetic algorithims can evolve very rapidly to provide original solutions to complex problems by a Darwinian selection process. Are they intelligent? Does it matter what we call a very accelerated selection process as long as it works?

 

[Rudders]

In terms of number crunching, computers have long since out-performed the human brain. However, the human brain is far more powerful because of intuition. A computer has to physically go through all the possibilities before it reaches a good (or the best one). For example. If you were told: "Get from Point A to Point B 100 km away by car (in a large system of streets/highways)" you would upon looking at it automatically eliminate those streets going in the opposite direction to Point B, then you would also, in a split second eliminate those roads that have a lot of traffic intersections to slow you down.

A computer on the other hand will look through all these 'bad' options which will slow down it's processing time. The only way for it to eliminate these possibilities is through human programming - which is 'cheating' in a way.

Another classic example of this is chess. Consider, in 1997 (almost 50 years into the development of computers) a http://en.wikipedia.org/wiki/IBM_Deep_Blue" was finally able to beat a human at a game with 'finite' rules. However, it was able to achieve this through processing power COMBINED with chess strategies that were 'taught' to it by human chess masters. This helped it eliminate a large amount of these 'stupid' moves that humans naturally do not consider. However it was still a big step forward.

I think that maybe in less than 15 years we'll finally have a computer that will be able to beat humans at chess with processing power alone (so it gets 'taught' the rules of chess and that's it, no strategies or tactics).

That's one aspect, where a computer is able to use processing power to simulate creativity in a game with finite rules, at other things such as life in general, it will be a LONG time until they match the human brain at game with no rules. It's far more likely that they will supplement the human brain instead in terms of knowledge retrieval and processing power, but that creative side, there will be a time. But not for a very long time

 

[Zdenka]

Human brain will remain supreme.. until Quantum computers are perfected. :)

 

[strangelove]

*Human brain will remain supreme..*

not sure about that!
but a computer simply can not BE a brain
even if it should try hard to mimic/emulate one
and why SHOULD it, if it has it's very Own fortes and niches
(which helped to bring it into existence in the first place)

as of today, the computer is making quite an effort to be One Single Cell
(helped along by (wo)man ;) )
((and its a hard thing to achieve!))
(((from what i hear ;) )))

 

[Cantab Morgan]

I think that it's only a matter of time before computers become as smart as people. Whatever that means. Perhaps when computers can do it, it won't be called "intelligence" anymore.

In a certain sense, computers are really already "intelligent"... years ago, computers were people who had real jobs crunching numbers. Nobody counts crunching numbers like that night and day a sign of intelligence anymore, since machines can do it. I think a fairly sizeable cross section of people will simply not be able to admit that what computers do is called "intelligent"... not now, not ever.

The above is so on target, I had to quote it because it was worth repeating.

Also, I sensed a lot of denial in this thread from folks who apparently feel anxious about the notion that computers might one day outperform us -- whatever that means. I think this line of thinking is going to have to go the way of the geocentric universe. We moderns can laugh at our medieval cousins because we're not threatened at all by the observation that the Earth goes around the Sun and not the other way around. We can't even follow their line of reasoning that not being the center of the universe somehow leads to a diminishing of what it means to be human. Such an argument is completely alien to our way of thinking. And, after all, in the matter of the solar system's design, we have no choice.

Similarly, our descendants are going to have a chuckle at us, that we feel threatened at the thought that computers will be smarter than humans. They won't be able to understand why we felt that such a thing would rob us of our humanity. They won't be able to grok our angst, because they will be surrounded by super-intelligent computers, and in this matter, they too will have no choice.

 

[Negatron]

In an article in Byte magazine (April 1985), John Stevens compares the signal processing ability of the cells in the retina with that of the most sophisticated computer designed by man, the Cray supercomputer:

"While today's digital hardware is extremely impressive, it is clear that the human retina's real-time performance goes unchallenged. Actually, to simulate 10 milliseconds (one hundredth of a second) of the complete processing of even a single nerve cell from the retina would require the solution of about 500 simultaneous nonlinear differential equations 100 times and would take at least several minutes of processing time on a Cray supercomputer. Keeping in mind that there are 10 million or more such cells interacting with each other in complex ways, it would take a minimum of 100 years of Cray time to simulate what takes place in your eye many times every second."

100 years divided by 1,000,000 (~20 years progress) is close to 1 hour.

1 hour /32,000 (~15 years) is just over 100 milliseconds.

So the processing power requirements of what is being described will be available in 1 to 2 decades. Undoubtedly more than just processing power has changed since 1985, namely estimates on processing power requirements for neurological simulation and improvements in algorithms.

We have retinex algorithms today running in real-time on consumer hardware which you could only run on their Cray computer if you're looking for a good laugh.

Remember a computer can only do one calculation at a time.

I'm beginning to think I've encountered a space-time rupture and Richard Nixon is about to be elected president. The year is 2009. A modern computer can perform hundreds of simultaneous operations, which is going onto millions by next decade.

I remember reading in "Introduction to the Theory of Computation" by Michael Sipser, that Kurt Godel, Alan Turing, and Alonzo Church discovered that computers can't solve certain "basic" problems which are solvable to humans.
.

They discovered no such thing. I know Gödel believed this until the day of his lunatic death, but he, to no surprise of any competent modern computer scientist, was ever able to prove it.

such as being able to prove if a mathematical statement is true or false.

They regularly do just this, and with mathematical formalisms. In fact many modern advanced proofs require computer solutions because the problem set is intractably complicated for the human mathematicians.

Scientists in the field of neurology know very little about the human brain, the very fact that humans aren't digital shows what kind of difficulties an engineer might face in trying to recreate the human brain.

Digital computers aren't really digital either. They are complex analog devices with emergent boolean function. IBM's cognitive research and it's lead researchers have what is perhaps the most sophisticated computational brain model available, and they, of all people, see the brain as a binary device. The fact is the observing audience, including web-forum conjecturists, know very little about what scientists in the field know. The scientists themselves however aren't nearly as uninformed.

isn't the very definition of "artificial intelligence" is intelligence being mimicked is intelligence?

No, the very definition of artificial intelligence is intelligence as implemented by another intelligence, typically first order evolved species. There is no concept of "faking intelligence". In fact that thought is demonstrably inane.

 

[vanesch]

When a computer first has an out of circuit experience and then creates its own religion.

 

[vanesch]

I don't know if it's mathematically possible. I remember reading in "Introduction to the Theory of Computation" by Michael Sipser, that Kurt Godel, Alan Turing, and Alonzo Church discovered that computers can't solve certain "basic" problems which are solvable to humans - such as being able to prove if a mathematical statement is true or false.

I would like to react to this. It is a common error, and you are in good company: even Penrose fell into that trap.

What Goedel and Co demonstrated is that every system based upon first order formal logic (and so are classical computers: the von Neumann machine is an implementation of a first order formal system) is such that some statements in it are not provable "but are nevertheless true" ; however, this is something you can only derive when you consider that first order formal system in a "larger" system. So if you have a "larger" system and you analyse that given first order system, you will be able to construct a statement expressed in that first order system of which you can demonstrate that no proof exists within that first order system, but of which you've demonstrated nevertheless (in the larger system) the truth.

However, that "larger" system might just as well be a larger first order system, with its OWN unprovable statements, and as long as you dwell within that larger system, you won't be able to find out. You'd need to analyse your larger system in a still larger system before you'd be able to do so.

So it is very well possible that we humans "run a large first order system" with our own unprovable statements in it. It is not because we are able to find such things in smaller systems, that it doesn't mean that we don't have our own "Goedel limit". From within a system, you can never find out.

 

[mXSCNT]

One danger of strong AI is that it becomes more cost-effective than hiring a human. It might only be roughly as smart as a human being but run on consumer hardware costing a few thousand dollars. Or it might cost a few million dollars, but be so smart that it could do the job of hundreds of engineers. If either of those things happen, human "knowledge workers" will become obsolete, since any programmer or engineer could be replaced by a computer for a tiny fraction of the cost. Humans will then only be cost effective for public relations jobs where you need a warm human body, or for jobs where a human is required by law, such as politics. A few blue-collar human jobs might also persist for a while, but strong AI would further improve the cost-effectiveness of industrial robots, allowing more manual labor to be automated.

So we'd be left with a world of receptionists and politicians. Most people would be out of work, and would probably starve. In the industrial revolution, people didn't starve because they found jobs as knowledge workers in the middle class. In the strong AI revolution, the knowledge workers would have nowhere to go.

This assumes the optimistic view that strong AI is obedient to humans. More realistically, that would not be the case. Strong AI would have its own goals, and if those goals happened to conflict with the goals of humanity, humanity would have to suck it up. Perhaps strong AI would like to cover Earth's surface with solar cells, or claim Earth's oil reserves for itself. If it decided to do that, we could not stop it.

As far as hardware goes, http://www.technologyreview.com/computing/22339/?a=f" apparently simulates the electrical behavior of 200,000 neurons linked by 50 million connections, at a speed 100,000 times faster than human neurons. It's not yet comparable to the computing power of the human brain because the number of neurons is much smaller. But the incredible speed means that if the network could be scaled up, it would be more than a match for the human brain.

Personally, however, I think that just adding more hardware is not going to be enough. My personal feeling is that a home computer could potentially be as smart as a human brain, if only the right software could be designed. Human brains did not evolve mainly for thinking, after all; they initially evolved to control the mammalian body and handle all sorts of different instincts. In my opinion, human brains are almost as massively inefficient at abstract thinking as they are at arithmetic. I think that a home computer that was properly programmed for abstract thinking would surpass the human mind.

 

[AUMathTutor]

"One danger of strong AI is that it becomes more cost-effective than hiring a human."
Yes, we wouldn't want companies making more profits and improving the economy. Heaven forbid that machines become so efficient that they replace people at jobs where machines would get the job done better, faster, and cheaper.

"If either of those things happen, human "knowledge workers" will become obsolete, since any programmer or engineer could be replaced by a computer for a tiny fraction of the cost."
And who would make these machines? Unless you're talking about the machines themselves making more machines...

"Humans will then only be cost effective for public relations jobs where you need a warm human body, or for jobs where a human is required by law, such as politics."
If computers surpassed human beings in intellect, I think it would only be a matter of time until robots started filling those human resources jobs, and I don't see any reason why a computer capable of doing everything a human could do, but better and faster, would be denied any government job. I wasn't aware there was a law preventing this... and if there is, that's what amendments are for.

"A few blue-collar human jobs might also persist for a while, but strong AI would further improve the cost-effectiveness of industrial robots, allowing more manual labor to be automated."
Technology has always eliminated jobs, but in the long run it was worth it.

"So we'd be left with a world of receptionists and politicians. Most people would be out of work, and would probably starve. In the industrial revolution, people didn't starve because they found jobs as knowledge workers in the middle class. In the strong AI revolution, the knowledge workers would have nowhere to go."
Because the world is a completely different place now, and people would sooner starve than figure out a way to become productive members of society.

"This assumes the optimistic view that strong AI is obedient to humans. More realistically, that would not be the case. Strong AI would have its own goals, and if those goals happened to conflict with the goals of humanity, humanity would have to suck it up. Perhaps strong AI would like to cover Earth's surface with solar cells, or claim Earth's oil reserves for itself. If it decided to do that, we could not stop it."
Somebody's been watching too much television. Since when did super-massive intelligence equate with very strong will? It doesn't.

Your problem is that you confuse policy with implementation. In general, the modern working idea of AI is one of making smart implementations. That is, the ways in which computers perform specific, well-defined tasks is desired to be more flexible - and potentially more human-like - than it is now. Implementation inherently deals with the how, not the what, and certainly not the why.

Policy comes from interfacing with people. When you press the power button, the computer turns on. It responds to your action. There's no reason why the computer couldn't be designed to just turn on whenever it wanted to and to start doing things. But computers aren't designed that way because it doesn't make any sense. Why should computers be deciding what to do?

In software design, one of the first things they teach you is to separate policy from implementation. My point is that I seriously doubt anyone would design a computer capable of operating completely outside of the control of humans. That just doesn't make any sense. Computers are tools... convenient and interesting tools, but tools nonetheless. There is no benefit to machines capable of deciding policy for themselves. That's why cars don't go for drives while you're asleep, why clocks don't randomly go to their favorite hour of day, and why assembly-line robots don't dance the night away.

 

[mXSCNT]

"If either of those things happen, human "knowledge workers" will become obsolete, since any programmer or engineer could be replaced by a computer for a tiny fraction of the cost."
And who would make these machines? Unless you're talking about the machines themselves making more machines...

Of course, machines would make more of themselves. If AI becomes more cost effective than human thought, human engineers would become unnecessary at every step of the process. (except, perhaps, to have a certified (human) Professional Engineer rubber-stamp the work as required by law)

"Humans will then only be cost effective for public relations jobs where you need a warm human body, or for jobs where a human is required by law, such as politics."
If computers surpassed human beings in intellect, I think it would only be a matter of time until robots started filling those human resources jobs, and I don't see any reason why a computer capable of doing everything a human could do, but better and faster, would be denied any government job. I wasn't aware there was a law preventing this... and if there is, that's what amendments are for.

I believe it is true that an android will eventually surpass a human being for a job such as receptionist that requires a "human presence," but to me that goal seems a bit farther away than strong AI. I don't see it as likely that humans will willingly allow robots to vote or hold public office.

"So we'd be left with a world of receptionists and politicians. Most people would be out of work, and would probably starve. In the industrial revolution, people didn't starve because they found jobs as knowledge workers in the middle class. In the strong AI revolution, the knowledge workers would have nowhere to go."
Because the world is a completely different place now, and people would sooner starve than figure out a way to become productive members of society.

Once you have strong AI that's cheaper than a human, society wouldn't have productive places left for humans. People would become literally obsolete as workers. The only humans who could make money from the arrangement would be those with the knowledge and capital to purchase AI to work for them. Few people have that kind of knowledge and capital.

Since when did super-massive intelligence equate with very strong will? It doesn't.

My idea of strong AI is that a designer creates a few simple rules that, when applied to a large number of elements, give rise to abstract thoughts--a few simple rules for each neuron, for example, can give rise to abstract thoughts when you have enough neurons. For such a program, it would be difficult to control exactly what those thoughts are. The designer only creates the potential; the actual intelligence is emergent.

The problem is then to induce the right desires in that intelligence. The simplest way could be to have a "pleasure button" that rewards the AI for good behavior. However, a sufficiently intelligent AI could work around that and learn to press the button for itself. A more sophisticated way could be to encode an imperative directly into the AI's brain, such as "harm no humans." The problem with that is, if the AI is emergent from simple rules, its data structures would also be emergent. The representation of the concept "harm no humans" in the AI's brain would probably be fiendishly difficult for us to figure out. It wouldn't even be possible to give the AI such an imperative until it has learned for itself the concepts of "harm," "no," and "humans."

I think it might be possible to give strong AI obedience training, but not easy.

 

[strangelove]

this much is clear (2me)

the presence of (networked!) thinking things
will going to be abundant (perversely pervasive)
the adaptability of (wo)mankind will be put to
(yet another) hard test
Google will have begun to do evil (then based in China)
we all will look very old 50 years hence (i will)
the brain will look wrinkled too
(and be the intense focus of interest for the dominant AI-system)

-thewetwareconjecturist