The Coming Singularity
IEEE Spectrum had a full issue on the Singularity this month, raising the level interest in that topic quite a bit.
I had a couple of problems with some of the discusion. First, some of the discussion claimed that in order to build a human level AI, we must first understand how conciousness occurs. However, we have an existance proof showing this is a fallacy: evolution did not understand conciousness before it created conciousness.
Second, some of the discussion claims that human level AI will not be possible and that nanotech manufactre is nigh near impossible. However, I find it absurd to believe that humans cannot reverse engineer evolution and figure out ways to tweak the process and make it better.
Vinge's article at the end set me to thinking:
The AI, IA, and internet scenarios are all effectively the same. There is a fundamental equivalence between serial and parallel processing. So it isn't really interesting to ask about when we will get a certain amount of processing power available in a cubic foot box.
The entire processing power of the world -- the human+internet scenario -- is the interesting scenario. Improvements in transportation and the invention of writing were major contributors to helping multiple human minds work in a more integrated parallel fashion. Today, the hardware basically provides improved communication so that the wetware portion of world processing power is more effective.
Today, we can obtain rapid increases in total world intelligence by doing a better job of educating children throughout the world. Better health care and longevity also increase total world intelligence by small amounts.
An interesting inflection point occurs when the cost to manfacture a human equivalent of processing power costs less than growing and training a human. Another interesting inflection point occurs when the total manufactured processing power contributed to the world each year exceeds the increase in the total organic processing power.
Man and computer are already interlinked in a global web. Civilization is self aware. Civilization is concious. Civilization does currently display interesting emergent behaviors -- landing a man on the moon and exploring the solar system with robots is not the work of an individual.
In twenty to forty years, we will start to double the total processing power of human civilization every year or two. The challenge until then is to grow our human processing power at its potential. Today we can manufacture on the order of 1,000 human equivalent processors per year. But we can train and grow on the order of 10,000,000 humans simply by doing a better job of educating the world's children.
I had a couple of problems with some of the discusion. First, some of the discussion claimed that in order to build a human level AI, we must first understand how conciousness occurs. However, we have an existance proof showing this is a fallacy: evolution did not understand conciousness before it created conciousness.
Second, some of the discussion claims that human level AI will not be possible and that nanotech manufactre is nigh near impossible. However, I find it absurd to believe that humans cannot reverse engineer evolution and figure out ways to tweak the process and make it better.
Vinge's article at the end set me to thinking:
The AI, IA, and internet scenarios are all effectively the same. There is a fundamental equivalence between serial and parallel processing. So it isn't really interesting to ask about when we will get a certain amount of processing power available in a cubic foot box.
The entire processing power of the world -- the human+internet scenario -- is the interesting scenario. Improvements in transportation and the invention of writing were major contributors to helping multiple human minds work in a more integrated parallel fashion. Today, the hardware basically provides improved communication so that the wetware portion of world processing power is more effective.
Today, we can obtain rapid increases in total world intelligence by doing a better job of educating children throughout the world. Better health care and longevity also increase total world intelligence by small amounts.
An interesting inflection point occurs when the cost to manfacture a human equivalent of processing power costs less than growing and training a human. Another interesting inflection point occurs when the total manufactured processing power contributed to the world each year exceeds the increase in the total organic processing power.
Man and computer are already interlinked in a global web. Civilization is self aware. Civilization is concious. Civilization does currently display interesting emergent behaviors -- landing a man on the moon and exploring the solar system with robots is not the work of an individual.
In twenty to forty years, we will start to double the total processing power of human civilization every year or two. The challenge until then is to grow our human processing power at its potential. Today we can manufacture on the order of 1,000 human equivalent processors per year. But we can train and grow on the order of 10,000,000 humans simply by doing a better job of educating the world's children.
posted by Cesium at 10:29
0 Comments:
Post a Comment
<< Home