Society of Mind
The book On Intelligence by Jeff Hawkins, was mentioned in a recent post here. Another book that fits well with the same general theory of intelligence, is Society of Mind by Marvin Minsky.
Taken together they begin to sketch out a general outline for the theoretical underpinnings of a basic machine intelligence. The early machines will be crude imitations of human intelligence. Artifice imitates nature, inevitably. As the craft matures, divergence from the model will occur naturally.
Society of Mind is a much older book, and I was surprised that Hawkins gave no credit to Minsky in his references or index. Regardless, the rest of us are under no obligation to imitate Hawkins in that regard. Society of Mind is a seminal work that should be read and re-read by anyone who is serious about breaking into the machine intelligence field. Minsky fleshes out the concept of mental hierarchy in a clear and detailed fashion.
Hawkins focuses on neocortical examples of hierarchy, pattern sequencing, pattern invariance, and pattern auto-association. These are all powerful concepts that are well illustrated by actual neocortical functioning. Minsky touches on these concepts, but muses about a great deal more than Hawkins deals with. Hawkins zeroes in on the aspects of human consciousness that he feels hold the key to designing a machine intelligence. Minsky is far more general. Hawkins talks explicitly about neocortex, how it functions, how it might be imitated. Minsky stays away from the neuroscience, for the most part. Both books are highly inspirational.
The earliest functioning machine intelligences will copy a lot from the biological proof of concept. I would expect many graduate neuroscience students to perform "experiments" using machine models of neocortex, and possibly models of thalamus and hippocampus. The nascent field of neuro-prosthetics will probably be given a boost by these early neocortical models.
It is easy to conceive of "wearable sensors" that record the experiences of their human wearers, for later transfer to a neo-cortical machine. In this manner a type of record of a person's life might be maintained in parallel with the actual person's memories. As machine/neural interfaces improve, the record could easily become more accurate. These machine "parallel personalities" should not be confused for a real person. Not for a long time.
In later posts, I'll muse on possible means of utilising these neo-cortical machine personalities to augment the intelligence of biological humans.
Taken together they begin to sketch out a general outline for the theoretical underpinnings of a basic machine intelligence. The early machines will be crude imitations of human intelligence. Artifice imitates nature, inevitably. As the craft matures, divergence from the model will occur naturally.
Society of Mind is a much older book, and I was surprised that Hawkins gave no credit to Minsky in his references or index. Regardless, the rest of us are under no obligation to imitate Hawkins in that regard. Society of Mind is a seminal work that should be read and re-read by anyone who is serious about breaking into the machine intelligence field. Minsky fleshes out the concept of mental hierarchy in a clear and detailed fashion.
Hawkins focuses on neocortical examples of hierarchy, pattern sequencing, pattern invariance, and pattern auto-association. These are all powerful concepts that are well illustrated by actual neocortical functioning. Minsky touches on these concepts, but muses about a great deal more than Hawkins deals with. Hawkins zeroes in on the aspects of human consciousness that he feels hold the key to designing a machine intelligence. Minsky is far more general. Hawkins talks explicitly about neocortex, how it functions, how it might be imitated. Minsky stays away from the neuroscience, for the most part. Both books are highly inspirational.
The earliest functioning machine intelligences will copy a lot from the biological proof of concept. I would expect many graduate neuroscience students to perform "experiments" using machine models of neocortex, and possibly models of thalamus and hippocampus. The nascent field of neuro-prosthetics will probably be given a boost by these early neocortical models.
It is easy to conceive of "wearable sensors" that record the experiences of their human wearers, for later transfer to a neo-cortical machine. In this manner a type of record of a person's life might be maintained in parallel with the actual person's memories. As machine/neural interfaces improve, the record could easily become more accurate. These machine "parallel personalities" should not be confused for a real person. Not for a long time.
In later posts, I'll muse on possible means of utilising these neo-cortical machine personalities to augment the intelligence of biological humans.
Labels: artificial intelligence, neuroscience
posted by al fin at 7:36 am
0 Comments:
Post a Comment
“During times of universal deceit, telling the truth becomes a revolutionary act” _George Orwell
<< Home