Memory-Based Architectures for AI
Thesis: Advanced intelligences harnessing pattern recognition will be memory-based technologies.
An post on the J Curve (Venture Capital dynamo Steve Jurvetson's blog) makes an interesting poitn with regard to artificial intelligence. Steve talks about a book he read that argues that artificial intelligence systems, like the brain, are memory based rather than processor based. His analysis echoes the theme of pattern recognition discussed in some of my recent posts. A system based on pattern recognition is using data it has collected through experience (or been hardwired with) to recognize objects, events, situations, solutions, etc... through comparison. The brain is essentially a large memory device (with ~8 terabyte capacity according to Jurvetson's post) that relies on data itself to interact globally in a parallel manner in order to produce intelligent behavior. Today's computer systems run a program (which only requires a fraction of the total memory of the machine and exists in RAM) which can only process one instruction in a given instant resulting in an archetectual bottleneck. The brain, on the other hand, has a cache of life experiences organized in an intelligent network, and when brain activity occurs, all of the data communicates and interacts simultaneously,resulting in intelligent behavior. True, non-human AI systems will likely employ a similar memory-based architecture.
An post on the J Curve (Venture Capital dynamo Steve Jurvetson's blog) makes an interesting poitn with regard to artificial intelligence. Steve talks about a book he read that argues that artificial intelligence systems, like the brain, are memory based rather than processor based. His analysis echoes the theme of pattern recognition discussed in some of my recent posts. A system based on pattern recognition is using data it has collected through experience (or been hardwired with) to recognize objects, events, situations, solutions, etc... through comparison. The brain is essentially a large memory device (with ~8 terabyte capacity according to Jurvetson's post) that relies on data itself to interact globally in a parallel manner in order to produce intelligent behavior. Today's computer systems run a program (which only requires a fraction of the total memory of the machine and exists in RAM) which can only process one instruction in a given instant resulting in an archetectual bottleneck. The brain, on the other hand, has a cache of life experiences organized in an intelligent network, and when brain activity occurs, all of the data communicates and interacts simultaneously,resulting in intelligent behavior. True, non-human AI systems will likely employ a similar memory-based architecture.
posted by Clayton at 9:36 PM
0 Comments:
Post a Comment
<< Home