The end of Moores Law..

[orbiter]

A pretty interesting program I thought you guys might like called 'The end of Moores law'. I heard it on radio 4 last night. Topics cover future nano-architectures, memristors etc.

Just download from my mediafire account...

http://www.mediafire.com/?52074o5dtg4mbyu

[Markybhoy]

Thanks for the heads up.

[DJPhil]

Thanks for the heads up!

Mediafire freaked out a bit on my machine, though it may have been a problem on my end.

Here's another link to the program in case anyone else has trouble.

http://www.bbc.co.uk/programmes/b00w7ccr

[orbiter]

Just checked my link DJ and it seems to be working ok now. I think mediafire download speeds are very slow at certain times so perhaps that was the problem you had.

I just captured this stream & uploaded it to M'Fire in case it disappeared from the BBC i-player, and my EEVblog buddies missed it.

[quantumfall]

Although not a real physics law its done very well as a rule of thumb.

Interesting to see that there is still a long way to go, and other avenues to keep the pace of processing advancing and the way we use data / storage.

I would like to see an always on (Low power use) non volatile solid state PC (Without compromising performance).

I think you can always find ways to use more processing power.  Its software that is harder to improve is it not ?

[Mechatrommer]

Although not a real physics law its done very well as a rule of thumb.

more like a paradigm or postulate to me, not even deserved to be called a "theory", let alone "law".

[the_raptor]

Although not a real physics law its done very well as a rule of thumb.

It is a self-fulfilling prophecy. The big silicon companies could do better but as long as they hit Moore's law management is happy and the media don't run around like Chicken Little.

[Zero999]

It's a prediction.

I believe many improvements in technology have followed an exponential trend long before Moore came along.

Of course Moore's law can't carry on forever, we'll reach a limit sooner or later.

What does amaze me though is the slow progress with AI: computers still don't think intelligently and tend to solve problems with brute force rather than cunning. Some predict that by 2025 computers will be powerful enough to emulate the human mind, which may be true but it'll never happen without an accurate model which I'm doubtful will exist by then. Put it this way, even though today's computers are powerful enough to emulate the 1 million neurons in a bee's tiny brain, it still beats them hands down in many areas, such as the travelling salesman problem.
http://www.cbsnews.com/8301-501465_162-20020587-501465.html

[the_raptor]

Put it this way, even though today's computers are powerful enough to emulate the 1 million neurons in a bee's tiny brain, it still beats them hands down in many areas, such as the travelling salesman problem.

The trick is that a neuron is only part of the hardware. The reason we have no "real" AI is because until the last ten years or so we had a wrong and over-simplified understanding of basic neuro-anatomy which led us to believe that it was just neurons.

I am currently studying physiology and psychology (career change from IT) and honestly we know more about exo-planets then we do about how the brain actually functions. We are going to need a huge leap forward in sensing technology before we can even begin to hope to reverse engineer wetware (stuff like f-MRI is equivalent to licking your finger and seeing how much of a shock a circuit gives you, there is very little real science coming out of it).

Really the quickest way at this point in time to get "AI" is to wire real wetware up to hardware like the. But that is too macabre for the dripping wet sorts who control science funding.

P.S. I think AI is largely pointless anyway (unless you want to be a technocrat and get rid of most of the workers). The benefit of computers is that they aren't sloppy and error prone like wetware, and it is fairly obvious that the error potential in wetware is what allows it to function how it does (by building rough patterns that are used to generate approximate solutions). The real benefit will be integrating hardware and wetware so that the power of wetware can be backed by the things hardware is good at.

[Mechatrommer]

AI will be just an AI. it will not become an I (real true intelligent), its just one step ahead in robot app (machine to help human task with lesser human interference). There is one secret "soft" recipe/algorithm embedded in the real intelligence, not just a simple neurons or whatever silicons they want to try to emulate it with. my 2cnts speculation.

[the_raptor]

AI will be just an AI. it will not become an I (real true intelligent)

AI is just intelligence of an artificial nature*. The problem is that "intelligence" is loosely defined and up to this point artificial systems have been pushed as "AI" when they give comparable performance in one area of what we call "intelligence" (and we tend to redefine "intelligence" when computers get good at one area). This is IMO because AI research has generally tried to work backwards from what intelligent organisms can do to generate a comparable system, instead of looking at how intelligent organisms do what they do at a base level and building a self-constructing and self-learning system (which is what a brain is).

* Any "AI" that doesn't have fully comparable "intelligence" to a living organism isn't actually AI. An AI that could do everything a rat brain could do would be a rat-level AI.

[tyblu]

interesting sound clip. I remember giving a speech in my 2nd year engineering, in 2005, about this, as an introduction to Quantum Computing; all about how it has to be the future of computing, since, by 2020, we will be using just a few atoms to guide electrons. Also, Moore's Law assumes a linear relationship for log(# transistors)/cm, but if you do 2nd order regression on the same data, getting A+B*log(#trans)+C*(log(#trans))^2, you'll find the 'Law' peters out at about 2019.

[Hypernova]

Put it this way, even though today's computers are powerful enough to emulate the 1 million neurons in a bee's tiny brain, it still beats them hands down in many areas, such as the travelling salesman problem.

The trick is that a neuron is only part of the hardware. The reason we have no "real" AI is because until the last ten years or so we had a wrong and over-simplified understanding of basic neuro-anatomy which led us to believe that it was just neurons.

I am currently studying physiology and psychology (career change from IT) and honestly we know more about exo-planets then we do about how the brain actually functions. We are going to need a huge leap forward in sensing technology before we can even begin to hope to reverse engineer wetware (stuff like f-MRI is equivalent to licking your finger and seeing how much of a shock a circuit gives you, there is very little real science coming out of it).

Really the quickest way at this point in time to get "AI" is to wire real wetware up to hardware like the. But that is too macabre for the dripping wet sorts who control science funding.

P.S. I think AI is largely pointless anyway (unless you want to be a technocrat and get rid of most of the workers). The benefit of computers is that they aren't sloppy and error prone like wetware, and it is fairly obvious that the error potential in wetware is what allows it to function how it does (by building rough patterns that are used to generate approximate solutions). The real benefit will be integrating hardware and wetware so that the power of wetware can be backed by the things hardware is good at.

An actual brain-in-a-jar?! I want to see the jar!!!