Advanced Chips and Robotics?

Hello,

This year Intel will produce a chip with a billion transistors. I remember back around a decade ago when chips reached the "astronomical" level of 80 million transistors. There were predictions of intelligent robots, advanced AI applications, & more technological wonders.

The closest we have to a robot is the Honda robot - which still cannot handle many basic functions. AI has gradually increased, but nothing spectacular. Why has there not been more progress?

Joel

Reply to
Joel
Loading thread data ...

I do not want to be overly critical of the Honda robot, at least it is a start!

Reply to
Joel

Easy money.

Reply to
Jeff Fox

Because you haven't come up with a brilliant solution yet, of course. And neither has anyone else. The problem turns out to be harder than we expected it to be.

Reply to
Joseph Hansen

I could believe that 'doing' AI was merely a matter of the computational resource thrown at it, if somebody could show me a very very slow intelligent system ...

Dave

Reply to
Dave Garnett

Well, except for little things like Google. ("What, you mean I can just ask a computer pretty much anything, and it will give me a wide range of answers from context?")

Apart from that, we're mechanically limited. (Says a man with a vested interest :-)

cheers, Rich.

Reply to
Rich Walker

--------------- What do you MEAN by "intelligent"?? There's lots of it around now that didn't exist even ten years ago!

How slow? You see, these things jump powers of ten quickly when you get to wanting things like "self-awareness" or such. We first have to get near the sub-tasks required to even WRITE an awareness function, and then it may be more obviously in sight.

If an awareness ran at a thousand times slower, would you know it was working correctly or even know it WAS working?

-Steve

Reply to
R. Steve Walz

It's a no brainer (:-)) pardon the pun (:-)). It takes 20 years to raise a human being.

Tom

Joel ( snipped-for-privacy@cox.net) wrote: : Hello,

: This year Intel will produce a chip with a billion transistors. I remember : back around a decade ago when chips reached the "astronomical" level of 80 : million transistors. There were predictions of intelligent robots, advanced : AI applications, & more technological wonders.

: The closest we have to a robot is the Honda robot - which still cannot : handle many basic functions. AI has gradually increased, but nothing : spectacular. Why has there not been more progress?

: Joel

--

Reply to
Thomas D. Ireland

And somewhere around the time of 05/28/2004 19:14, the world stopped and listened as R. Steve Walz contributed the following to humanity:

Myself and a co-worker had a discussion about awareness awhile back. One point that he made was that a device as simple as a smoke detector could be considered self-aware of it's environment. If there's smoke, then there's fire, and I need to make a sound to wake someone up.

Any program can determine that there is a problem and then decide on what actions to take to correct it. The problem with this is that every possible problem and it's solutions must be pre-programmed before hand. The goal of machine intelligence is to deduce there is a problem, and come up with creative solutions to fix it, which may not have even been thought of by the programmer.

Basically, the final goal is to build a Lt. Commander Data using current technology, which is impossible at this point in time.

Reply to
Daniel Rudy

The Honda Robot is not even remotely close to closest to human like intelligence, it may well be one of the better ones at modeling human walking, but it is still miles away from being remotely close to human physical performance for reasons I will get to later after addressing the first question.

A billion transistors IC is not remotely close to the complexity of a human brain. A human brain has upwards of 1 trillion neurons, each with upwards of 1000 weighted, timed connections to other neurons. In addition, there are apparently a fairly large number of specialized structures in the brain which likely greatly increase our brains performance over a bunch of raw neurons. To get structure that complex in a digital computer I think you would need a lot more than 1 trillion transistors. I read recently that all the computers combined in the world were roughly equal to a single human brain in complexity. This was using the older more simplestic models of human Neurons, which some experiments (not duplicated to my knowledge) indicate could be under rating the human neurons by several orders of magnitude of comparative complexity. In additon, there is the problem that no one has a good model of how humans do relatively low level thinking. So even if you had your computer with enough complexity to duplicate human thought, it might take decades to come up with all the algorithms needed make it work.

To get a good example of how far we may have to go just look at the problems we are having in knowledge repersentation. As far as I know, there is still no working method of repersenting general knowledge of facts, much less emotions and the like.

The other problem with human like robots in general is that we don't have the mechnical ability to duplicate human's bodies yet, and aren't that close. There are hundreds of muscles involved in walking even on flat an level ground, and humans don't have a working method to buld the muscles or to control them to date. This could certainly change, but its still a problem. Its important to note that a Human Brain can't control its own muslces in detail either, there are effectively

100's, if not thousands of little neural sub-processors in the human body that do local control. So while the Honda Robot is impressive, its not intended to be more than an early first step at best towards human like mobility.
Reply to
Robert Posey

I completely agree with this point. To summarize: "deduction" "creativity" is one of the main missing puzzles in the game -AI/Robotics. Yet, I don't think that the actual technology is what is stopping us - a human brain might have trillions of neural cells but an ant has far less. Yet our machines/programs aren't yet capable to be at least as creative/aware as an ant is. I know some might argue around this - try to observe a small colony of ants, play with them by placing obstacles or food around, separate one from the group and see how they react and then think of any machine out there capable to "behave" such us?

I dare to say that ai today doesn't even reach as far as some unicellular life forms - is not technology what we lack here but understanding of some principles - a much as a bigger hammer wont fix your car . right? :)

As for comparing billions of transistors with trillions or neural cells - well, I wont look at numbers as much as I will care about the fact that our machine hardware now is mostly sequential - you might have a billion transistors but only capable to run a very limited execution paths where a every single neural cell has 's own. If you try to program/emulate a neural net as such as the real one you will see how much of these billion transistors you have to "burn" to achieve the necessary level of parallel processing(time sync) for a much more limited number of units(neurons). We need progress in parallel processing as for the moment is still very much in it's infancy.

Now, another point to look at is what do people want these machines to be/do? As far as I can see they just look to copy from nature/life and that being the case I think AI is a must for a real intelligent robot. And that will be here whenever humans understand what life is about. Another way to say it - you need to know what you are in order to be able to consciously create "life".

Larry

Reply to
Larry O.

The problem I think with asking questions like this on usenet is that they don't say 'Why have we not seen ... in the popular press.' What people believe is generally determined by what they have seen or what they have read from sources that they tend to believe. So my first question to people who ask these questions is "What is your security clearance?"

I see the problem in asking about what beliefs are most popular in a forum like usenet as no different than asking they guy on the street about nuclear fission in the early fourties or asking people about controled heavier than air human flight for quite a while after the Wright Brothers had demonstrated it. Consider:

A year before the Wright brothers flew their airplane at Kitty Hawk, Rear-Admiral George Melville, chief engineer of the US Navy, declared that attempting to fly a heavier-than-air aircraft was simply "absurd." A few weeks before the airplane flew, Simon Newcomb, a distinguished professor of mathematics and astronomy at Johns Hopkins University, stated that heavier than air powered human flight was, in scientific terms, "utterly impossible." According to Newcomb, any form of powered flight would require the discovery of an entirely new force. With such eminence behind these statements, the mainstream media of the day meekly followed the lead of the authorities, and sneered at the ridiculous notion of powered flight.

To add injury to insult, more than two years after the Wright brothers had first flown their aircraft, and in spite of the fact that dozens of eyewitnesses had actually seen them fly, the popular Scientific American magazine continued to ridicule the "alleged" flights. An editorial in the magazine explained why:

If such sensational and tremendously important experiments are being conducted ... on a subject in which almost everybody feels the most profound interest, is it possible to believe that the enterprising American reporter ... - even if he has to scale a fifteen-story skyscraper to do so - would not have ascertained all about them and published them long ago?

Many years later, when the editor of the Wright brothers' hometown newspaper was asked why he had refused to publish anything about their amazing accomplishment, he replied "We just didn't believe it. Of course, you remember that the Wrights at that time were terribly secretive." The interviewer responded incredulously, "You mean they were secretive about the fact that they were flying over an open field?"

The editor considered the question and replied sheepishly, "I guess the truth is we were just plain dumb."

At about the same time as the Wright brothers were flying their impossible machine, Einstein, Bohr, Heisenberg and others had begun to revolutionize physics with the quantum theory. Einstein's theories were vigorously attacked on the basis that their acceptance would throw back science to the Dark Ages. Robert Millikan, recipient of the 1923 Nobel Prize for Physics, unequivocally asserted,

"There is no likelihood man can ever tap the power of the atom. Nature has introduced a few foolproof devices into the great majority of elements that constitute the bulk of the world, and they have no energy to give up in the process of disintegration."

Today there are people who make their living claiming that AI is simply impossible for a variety of reasons. Read comp.ai.philosophy for such endless debates. In a tread there earlier this year the original poster in the thread concluded that even if such machines did already exist that they didn't count unless he could buy them at Sears.

Today product marketing is carefully staged. They don't release new stuff while the old stuff is making easy money. But eventually competition will force old things that have been held back to appear in the marketplace. This is basic business. Humans are still cheap, cheaper than ever.

Best Wishes

Reply to
Jeff Fox

And, as the world should know . . . 10 Months *BEFORE* the Wright Bros, The worlds first manned flight was done by Richard Pearce in New Zealand using a homebuilt lashup of bamboo with a homebuilt engine.

On or about 31 March 1903 Richard William Pearse of Waitohi New Zealand, became airborne in a high-wing monoplane he designed and built himself. This aircraft, of prophetic design, was powered by an ingenious petrol engine which he also designed and constructured. It was not until 17 December 1903 that the Wright Brothers' Flier I took to the air at the Kill Devil Hills in North Carolina.

Though Pearse himself later conceded that the Americans deserved the honour of being the first to make a controlled and sustained flight, it is almost certain that he got into the air under power before they did." Gordon Ogilvie, "The Riddle of Richard Pearse", Reed Books, Auckland, New Zealand, 1994, p.xiv.

Check out this WebSite for more information.

formatting link

Reply to
Imagineering Unlimited

Yes. The Wright Brothers never claimed to have done the first manned heavier than air or power flight. They were the first to work out controled flight, in particular the technique of wing warping that let them stay in the air with a reasonable degree of control.

My point was merely that the media was still claiming for years that it was simply impossible and that claims that it could be done were ludicrous. According to the experts of the man on the street it was simply and impossibility.

People accepted what they heard in those horse and buggy days just as they do today. Most people today are not impressed by indusrial robots or robot toys. They see robots in fiction movies and assume that anything beyond toys must be fiction. And some people make a living insisting that intelligent machines can never happen anyway.

Reply to
Jeff Fox

Could it be that most people are trying desperatly to make it more difficult for everyone instead of making it easier? The war on terrorism is certainly not helping things any! It is likely worth gazillions in economy as well and for that reason the technology may sit on the self until it is no longer recognizable as the technology it was originally. Look at what happened to laser disc technology. It was available in the video industry for 30 years before someone reinvented it as the CD ROM. Technology in the marketplace sometimes will go unnoticed for that long until other technologies can be developed that will compliment it. It is a simbiotic relationship I guess. Anyway if someone wants to make it easy someday it will happen. In the meantime Osama is still number one on the wanted list. Very deep ethical issues too.

Tom

Joel ( snipped-for-privacy@cox.net) wrote: : Hello,

: This year Intel will produce a chip with a billion transistors. I remember : back around a decade ago when chips reached the "astronomical" level of 80 : million transistors. There were predictions of intelligent robots, advanced : AI applications, & more technological wonders.

: The closest we have to a robot is the Honda robot - which still cannot : handle many basic functions. AI has gradually increased, but nothing : spectacular. Why has there not been more progress?

: Joel

--

Reply to
Thomas D. Ireland

Why most artificial intelligence programs aren't - my summary of necessary functionality for general-purpose AI:

Perception takes raw sensory data and divides into discrete objects, attributes, relations

Induction detects patterns in data, and converts into a concise form for memory, communication, reasoning, and extrapolation

Memory efficient indexing (including non-exact matches), appropriately deleting old data, context sensitivity needed

Deduction extracts specific parts of previously detected patterns for use in current situation (induction+deduction effectively form a lossy compression system that can extrapolate from detected patterns in data as well as it can reconstitute a compressed pattern. this is usually a good thing, tho it can lead to false memories in some cases)

Decision given multiple scenarios, and actions that lead to one rather than another, commits to action with highest expected utility

Action translate high level action descriptions into motor command sequences

Tolerance of uncertainty essential, thus inference must be probablistic, except parts of the system that can be isolated from uncertainty or overriden by the rest of the system.

Due to uncertainty and combinatorial complexity, there must be multiple methods for performing most mental tasks - any one method will fail in many cases. therefore, the perceive-reason-action cycle must be applied not just to choosing appropriate actions in the world but to choosing appropriate mental actions (self-awareness). Also, the AI 'operating system' must have equivalents of threading and memory protection so any failed mental process doesn't crash the entire system.

The human brain isn't rigidly modularized - perceptual functions are used to build mental models and to reconstitute stored memories, perception and action are directly linked e.g. when walking, perception and action are somehow used internally for self awareness/control, there is learning in perception and action, reasoning by analogy combines induction and deduction...

One method for performing any of these functions is a difficult task in itself, can occupy entire fields of AI, and where most AI programs stop. Multiple systems for all required components usefully interfaced, given they are tightly coupled is very hard...

For a useful humanoid robot, throw in lots of high-quality actuators, much better battery technology than presently common (e.g., fuel cells), high-resolution tactile sensors ('robot skin'), and twenty years for the thing to learn common sense.

Reply to
sam kayley

While I grant you that its possible some defense contractor or investment resesarch firm has found a golden bullet algorithm that allows efficient use of the vastly superior math processing capability of modern computers to produce human like results, I am positive that is the ONLY way that someone could produce a artifical human level mind in the near future. Outside of the NSA, the level of processing capability is fairly well known to the community at large. Inside the NSA, I grant its fairly likely that they have computers that at

10-100x the speed and memory of any publically known machines, but that still isn't close. The one exception would be quatum computers, which its remotely possible the NSA has prefected far in advance of the publically disclosed work.

Other than the NSA option(which seems unlikely, since computers are already far better than any human at code breaking which is the NSA's main focus), we are left with the miricle algorithm concept, which is of course possible, but I think unlikely. The models of the human brain that seems most likely to be vaguely right indicate that modern computers will have to get much faster and larger to approach human level intelligence in general tasks. Note,there are already many known examples where computers can out perform humans in limited areas, and I would expect the number of areas to increase very rapidly.

I would predict that next great advance would be a good computer based translateion program that can truly translate human geneated text in a limited area into English(or whatever language). That would be a great advance for the world, as it would increse the number of books available to non-English readeres by 10 to 100 fold over night.

Reply to
Robert Posey

Note those saying heavier than air flight was impossible were clearly wrong. Birds weight far more than the air they displace, and those knowledgable person of the Wright's bothers day were well aware of this, so the comparison is not valid. No one that is knowledgable is saying computers will never out think humans, they are just saying it will be a while.

Reply to
Robert Posey

Or that it will be a while before humans will admit it even after it has happened. Most human's define intelligence as 'I' and are thus never really very convinced that any other intelligence is really provable.

Reply to
Jeff Fox

PolyTech Forum website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.