September 11th, 2006
|mathemajician||12:19 am - 1.6 Peta FLOPS and other news...|
According to this article, IBM's next supercomputer called Roadrunner will start to be constructed soon. When completed in 2008 it should perform 1.6 Peta FLOPS. I plotted all this out on a graph:
It looks like we could reach 1016 FLOPS by about 2010, a few years earlier than Kurzweil predicted. You can also see that the curve is bending upwards. As the y-axis is log scaled, this shows that the growth is slightly super-exponential as Kurzweil likes to point out.
Ok, so maybe the next thing is not really singularity-ish, but these driver-less robotic cars that are going to appear at Heathrow look pretty sci-fi if you ask me:
The singularity is going to make really smart machines right? But what exactly is intelligence? Recently I put together a list of definitions of intelligence and posted them to my work page here. Amazingly the page has received 5,000 hits over the last week and somebody even added it to Wikipedia... I feel positively famous ;-)
I don't post more technical singularity type stuff here, it's more main stream. However if you're into the more hard core side of things then you might want to check out my research blog: www.vetta.org. This just got started last week. It will contain interesting research papers that I find as well as bits of my own research (I research mathematical models of machine super intelligence and am also working on my own AI project on the side).
Interesting. You might get more response by posting to the Singularity Now
community. It has a bit few more eyes (even though both have been quiet latley).
If you want to find a sort of definition of intelligence would be a being that can recognize patterns and predict outcomes or at least know what does what and cause and effect.
This may or may not require free will or concioussness since one can argue that only the observer in the universe is the only real sentient being and all other humans are just souless automotons... But since we aren't that bitter about existence we agree with everyone that we exist even though we can't prove it.
That said... If an intelligent machine argues that it exists and has free will and makes an argument well enough that we decide to agree with it then by default it is so.
A while back I did a guess of what it takes to brute force Strong AI. According to Kurzweil, you need about 100 Billion MIPS (his estimate) in order to simulate the brain. So if an AMD-FX 60 can do that today (assuming we can just buy them in parallel) you would be looking at $1,011.98 or so odd per chip (that price might have changed latley) and would be around $3,734,243,542 (about $4 billion if you want to include labor and shipping), but if Moore's law held true it would only cost $60 million by 2012 and then $2,000 some by 2018. Which would mean the average PC would have the power of the human mind... Then by 2024 it would be litterly less than a penny. Of course it might take something more than that and we might the barrier to Moore's Law between now and then... And I doubt we'll be using massive amounts of AMD FX-60 chips in 2018 to do this... But is something to think about it...
I posted here because I'm the moderator... ;-)
If I work out the numbers then I get the machine costing $5 million in 2020. I used the idea that computer power doubles every 18 months for the same cost. To get a curve like yours I need to assume that computer power goes up by a factor of 4 for the same price every year! That's way too extreme.
Also, I think you might be a bit off mark due to the new cell processors coming out from IBM. They cost around $250 each (indeed you will shortly be able to buy a Playstation 3 with one for about $399) and have a peak performance of about 250,000 MIPS. That's how they are going to build the Roadrunner system that I mentioned in the post. Now if you work the cost out, you get 100 billion MIPS for just $100 million next year. If prices then half every 18 months, in 10 years time it will cost about $1 million. 10 years isn't all that far into the future.