This site is not maintained. Click here for the new website of Richard Dawkins.

← Rethinking artificial intelligence

Rethinking artificial intelligence - Comments

Friend Giskard's Avatar Comment 1 by Friend Giskard

They are floundering because they do not have positronic circuits.

Mon, 07 Dec 2009 20:55:00 UTC | #421136

TheDawgLives's Avatar Comment 2 by TheDawgLives

Interesting. As a computer scientist, I've thought for a while that AI research was focused in the wrong area. I think we should be looking at the hardware of AI instead of the software. For example, we should be building hardware neural networks instead of modeling them in software. While software is easier to build up and tear down, I think the massive parallelism required of AI demands a hardware approach.

Mon, 07 Dec 2009 21:00:00 UTC | #421137

Cook@Tahiti's Avatar Comment 3 by Cook@Tahiti

Whatever. Call me when I can buy a robot that can do the chores.

Mon, 07 Dec 2009 21:17:00 UTC | #421143

Friend Giskard's Avatar Comment 4 by Friend Giskard

there are supercomputers for a few thousand dollars now that can do a trillion operations per second


For comparison, my good friend Mr Data's (quite primitive, first generation) positronic brain operates at 60 trillion operations per second. Artificial stupidity might be a more realistic goal, for now.

Mon, 07 Dec 2009 21:17:00 UTC | #421144

God fearing Atheist's Avatar Comment 5 by God fearing Atheist

2. Comment #439626 by TheDawgLives
I think we should be looking at the hardware of AI instead of the software. For example, we should be building hardware neural networks instead of modeling them in software.


A prototype neuro-chip is under early tests at the moment. Its currently only one processor, but the architecture is designed to allow expansion to 1 milion processors.

There are two big problem:

1) Silicon computational systems take about 105 to 106 the power of biological computational systems (neurons);

2) Massively parallel neural network type systems will require going back to the drawing board with software. Hence the type of project in the article.

EDIT:- It is the SpiNNaker project.

Mon, 07 Dec 2009 21:30:00 UTC | #421150

NewEnglandBob's Avatar Comment 6 by NewEnglandBob


...the very nature of the human mind, evolved over billions of years as a complex mix of different functions and systems...

No, the human mind did NOT evolve over billions of years. All eukaryote life evolved over 500 million years or so. Human life over 250,000 years or so.

Mon, 07 Dec 2009 21:45:00 UTC | #421159

God fearing Atheist's Avatar Comment 7 by God fearing Atheist

4. Comment #439633
Mr Data's (quite primitive, first generation) positronic brain operates at 60 trillion operations per second.


A human brain has about 1014 synapses. At only one operation per second per synapse Mr Data's brain is only performing at 60% of the human brain.

Mon, 07 Dec 2009 21:55:00 UTC | #421161

glenister_m's Avatar Comment 8 by glenister_m

Couldn't one write a self-evolving program to emulate some example of intelligence? I understand that some robotics systems have gone this way, so the processor, through trial and error, learns what to do to move around a room.

In essence, why go to all the trouble of writing code. If you can build a simple neural net system in hardware, then let software evolution do the rest.

...or would that lead to Skynet...

Mon, 07 Dec 2009 22:02:00 UTC | #421166

Art Vandelay's Avatar Comment 9 by Art Vandelay

I'm tempted to think that any functioning brain needs some sort of body to work with. The human brain learns from sensory input even before birth: it soaks up sights and sounds and smells and it tells the limbs to move and explore and and manipulate.

Then it would need a theory of mind to interact with others and gather more information by proxy by comparing incoming information to its own experience. I think that would be the tricky bit.

And a fourth law of robotics: "Don't believe everything people tell you." We don't want to end up with electric priests.

Mon, 07 Dec 2009 22:35:00 UTC | #421180

andersemil's Avatar Comment 10 by andersemil


One of the projects being developed by the group is a form of assistive technology they call a brain co-processor. This system, also referred to as a cognitive assistive system, would initially be aimed at people suffering from cognitive disorders such as Alzheimer’s disease.


Alzheimer's is probably one of the most dreadful diseases we have and I sincerely hope that we will be able to, if not cure it, then at least help patients in this way. It would be truly remarkable if you could in any way relieve the symptoms.

Mon, 07 Dec 2009 22:54:00 UTC | #421194

God fearing Atheist's Avatar Comment 11 by God fearing Atheist

8. Comment #439655 by glenister_m
Couldn't one write a self-evolving program to emulate some example of intelligence? I understand that some robotics systems have gone this way, ...


9. Comment #439669 by Art Vandelay
I'm tempted to think that any functioning brain needs some sort of body to work with.


Evolving, embodied brain, Darwin VII:-

http://vesicle.nsi.edu/nomad/darwinvii.html

Mon, 07 Dec 2009 22:57:00 UTC | #421197

andersemil's Avatar Comment 12 by andersemil

8. Comment #439655 by glenister_m

That is pretty much how neural networks function. Given a set of input data and expected outputs, they build a more or less complex function which can then be used to process new input according to experience. Currently, however, the hardware is just too limited for using them to their full potential. As you say, the exciting bit would be to get hardware which was considerably more efficient at doing this than existing CPUs, which were never built for this kind of operation. I've heard that IBMs next generation of CPUs will consist of 48 parallel cores running at much lower frequency than today's 4 or 8 core CPUs. Parallelization is definitely the way things are going, but it's a far scream from what is needed in true AI development.

Mon, 07 Dec 2009 23:05:00 UTC | #421200

AshtonBlack's Avatar Comment 13 by AshtonBlack

The 48 core CPU is an Intel one, designed for cloud environments. Very cool tech.

Mon, 07 Dec 2009 23:14:00 UTC | #421205

blitz442's Avatar Comment 14 by blitz442

. Comment #439655 by glenister_m

...or would that lead to Skynet...


No, Cylons

Mon, 07 Dec 2009 23:17:00 UTC | #421209

TreenonPoet's Avatar Comment 15 by TreenonPoet

Comment #439669 by Art Vandelay:

We don't want to end up with electric priests.

I think we could end up with robots of whatever 'profession' we chose. What is to stop some groups choosing strategic planners to give them the edge over rivals? Could we ever get one step ahead of the first group to successfully produce robots with superhuman strategic planning ability? Should we try to avoid such a scenario (perhaps using religion to suppress science), or accept the evolution of robots?

Mon, 07 Dec 2009 23:46:00 UTC | #421219

Art Vandelay's Avatar Comment 16 by Art Vandelay

Thanks, GfA, that's fascinating. I note especially the " importance of self-generated movement in the aquisition of perceptual capability".

Mon, 07 Dec 2009 23:50:00 UTC | #421221

dumbcountryhick's Avatar Comment 17 by dumbcountryhick

Why not use the real thing? With advances in such areas as genetic engineering and stem cell research a future involving computers that are partially organic seems much more promising than trying to emulate (either in software or hardware) the most advanced nanotechnology on the planet.

Tue, 08 Dec 2009 03:20:00 UTC | #421291

anthonzi's Avatar Comment 18 by anthonzi

I don't see the point in this project. We are all using intelligent machines to read this article.

Tue, 08 Dec 2009 05:12:00 UTC | #421333

mbannonb's Avatar Comment 19 by mbannonb

When are they going to get to the really important work, like a real, functioning light saber.

Tue, 08 Dec 2009 05:31:00 UTC | #421338

Bala's Avatar Comment 20 by Bala

I work on AI for the gaming industry. Even the brightest of programs makes 3d game characters do dumb things like constantly walk into walls.

Tue, 08 Dec 2009 05:37:00 UTC | #421339

root2squared's Avatar Comment 21 by root2squared

Heres a test for a true AI. It will not corect this second mistake, but it should correct the first one.

Tue, 08 Dec 2009 06:33:00 UTC | #421344

m19d1a5's Avatar Comment 22 by m19d1a5

What about quantum computers. Give information it's it's substance...

Tue, 08 Dec 2009 06:40:00 UTC | #421347

Fall's Profit's Avatar Comment 23 by Fall's Profit

I work on AI for the gaming industry. Even the brightest of programs makes 3d game characters do dumb things like constantly walk into walls.


So you are the reason Ellis won't get into the god damn safe room in Left 4 Dead 2!

Tue, 08 Dec 2009 07:53:00 UTC | #421357

InfuriatedSciTeacher's Avatar Comment 24 by InfuriatedSciTeacher

A human brain has about 1014 synapses. At only one operation per second per synapse Mr Data's brain is only performing at 60% of the human brain.

You're assuming one synapse per operation, when operations require far more than one neuron to run. Does your figure for the number of synapses also include those involved in automatic functions, as opposed to conscious operation or working memory£ I've a feeling the (probably not all that well thought out) Star Trek example was referring only to working memory...

@ Fall's Profit> My copy of Syphon Filter still has a glitch where (can't remember his name) won't walk into the lab he's supposed to enter while you follow him, so I feel your pain. That should also give an idea of the last time I bought a game system.

Tue, 08 Dec 2009 16:54:00 UTC | #421472

rod-the-farmer's Avatar Comment 25 by rod-the-farmer


Electric priests

I like it. Then we can unplug them. No, a better idea...... Over-power them. They might then just sit there, on, but not doing anything, so people can see what a dim idea they are in the first place.

Ohm my god. Up the resistance !

Tue, 08 Dec 2009 19:05:00 UTC | #421534

SilentMike's Avatar Comment 26 by SilentMike

I've thought for some years now that it might be a better project to try and augment the existing human intelligence before we go about creating the whole thing from scratch. Not only is it more useful to us (fix us if we're damaged, improve us if we're not), it's also easier. True AI requires a complete system that integrates all different facets if intelligence, if we get one wrong, or if we don't know how to make it play nice with others, then it won't work.

In a way we have been augmenting our intelligence since we invented writing. A book allows for better recollection, a computer is more flexible than a book, a laptop is more handy and portable than the traditional desktop, etc. The internet is another improvement, as are hand-held devices. Hell, so was the calculator when we invented it. As our knowledge improves, why not continue to get closer to the source? The end-result of an AI from scratch may ultimately depend on the existence of minds much more capable than ours.

Wed, 09 Dec 2009 02:00:00 UTC | #421648

SaintStephen's Avatar Comment 27 by SaintStephen

Very interesting article. Skeptics of Artificial Intelligence would do well to remember these famously lucid predictions:

"To place a man in a multi-stage rocket and project him into the controlling gravitational field of the moon where the passengers can make scientific observations, perhaps land alive, and then return to earth--all that constitutes a wild dream worthy of Jules Verne. I am bold enough to say that such a man-made voyage will never occur regardless of all future advances."

Lee deForest, inventor of the vacuum tube, 1957
"Space Travel is utter bilge."

Sir Richard Wooley, astronomer, 1956.
"There is practically no chance communications space satellites will be used to provide better telephone, telegraph, television, or radio service inside the United States."

Tunis Augustus Macdonough Craven, FCC Commissioner, 1961

(The first commercial communications satellite went into service in 1965)
"Transmission of documents via telephone wires is possible in principle, but the apparatus required is so expensive that it will never become a practical proposition."

Dennis Gabor, British physicist and author of Inventing the Future, 1962
* * * * *

On a different note, AI Pioneer Marvin Minsky:
"I would like to be able to download the ability to juggle. There's nothing more boring than learning to juggle."
Now look below the article. Even Artificial Intelligence has its stupid peanut gallery:



Minsky? A closed-minded simpleton? Incredible.

Wed, 09 Dec 2009 10:27:00 UTC | #421709

Reckless Monkey's Avatar Comment 28 by Reckless Monkey

Trouble with AI is you're likely to invent a fallible computer. Anything else is just running a set of pre-defined steps. Net result will be a very expensive computer that might well end up believing in some form of Deity. We already have several billion of those. Want AI? Have sex.

Wed, 09 Dec 2009 10:55:00 UTC | #421712

SaintStephen's Avatar Comment 29 by SaintStephen

28. Comment #440229 by Reckless Monkey on December 9, 2009 at 10:55 am

Want AI? Have sex.
I know you were kidding. But think about it. You go ahead and have your kid, and I'll develop AI technology over the next 18 years. You spend 18 years raising your child to a young adult. I get 18 years of Moore's Law, which says computer power doubles approximately every 18 months -- so I get 12 doublings. This means the computer I run my AI program on, after 18 years, will be 2^12 times as fast as IBM's Blue Gene L is right now, for instance.

2^12 = 4096. In 18 years I will have a computer that is over four thousand times faster than today's best, in addition to any software/algorithm/architecture advances that arise in AI research over those same 18 years, or more than 3 iterations on the Mind Machine Project duration described in this article.

Fast and powerful indeed, but maybe your 18-year-old still wins at poker. But what about when your 18-year-old becomes 27 years old? My computer will then be 2^20 = 1,048,576... more than a MILLION times as fast. As your child lives through his thirties, with essentially the same brain power as when he was 18, my computer accelerates in power as follows

Year 33: 4,194,304 times as fast
Year 36: 16,777,216 times as fast
Year 39: 67,108,864 times as fast...

Getting worried yet?

Wed, 09 Dec 2009 12:00:00 UTC | #421717

SilentMike's Avatar Comment 30 by SilentMike

27. Comment #440226 by SaintStephen

On the other hand, consider this quote:

Before man reaches the moon your mail will be delivered within hours from New York to California, to England, to India or to Australia by guided missiles.... We stand on the threshold of rocket mail.

Arthur Summerfield, 54th United States Postmaster General, 1959

Wed, 09 Dec 2009 12:05:00 UTC | #421719