Another older stab at intelligence I think is a good read (just the first chapter though) is David Marr's book, Vision, from 1987. His breakdown of understanding vision as an information-processing problem still stands up to this day and even touches upon ideas in more modern critiques like Kording's "Could a Neuroscientist Understand a Microprocessor?" There's also discussion of the motor control of a fly near the second half of the first chapter that really reminds me of Brooks' subsumption architecture.
Looking forward to your article about mechanical intelligence in the future!
Thanks, I’ll have to check that out. The motor control of a fly part reminded me of a really good lineage of research exemplified by Tom Daniel at UW, or Noah Cowan etc. at JHU, etc. understanding animal brains with input/output style experiments. Basically, provide some kind of stimulus with some other potentially conflicting stimulus, and see the output. Those experiments also get at embodied intelligence!
It's interesting to read about so many different models for embodied intelligence; something implied here is that robotic systems should be based on the same model as human beings but I find that difficult to agree with given the Bitter Lesson, I doubt the best way humans can think of to do something is the best way in general.
Would you agree it's possible that any definitive theory for the structure of human embodied intelligence (be it as a distributed, modular or hierarchal system) may still not be optimal for robots?
In particular, a generalized data -> action (VLA-style) humanoid should still work in any case given sufficient training data
Another older stab at intelligence I think is a good read (just the first chapter though) is David Marr's book, Vision, from 1987. His breakdown of understanding vision as an information-processing problem still stands up to this day and even touches upon ideas in more modern critiques like Kording's "Could a Neuroscientist Understand a Microprocessor?" There's also discussion of the motor control of a fly near the second half of the first chapter that really reminds me of Brooks' subsumption architecture.
Looking forward to your article about mechanical intelligence in the future!
Thanks, I’ll have to check that out. The motor control of a fly part reminded me of a really good lineage of research exemplified by Tom Daniel at UW, or Noah Cowan etc. at JHU, etc. understanding animal brains with input/output style experiments. Basically, provide some kind of stimulus with some other potentially conflicting stimulus, and see the output. Those experiments also get at embodied intelligence!
It's interesting to read about so many different models for embodied intelligence; something implied here is that robotic systems should be based on the same model as human beings but I find that difficult to agree with given the Bitter Lesson, I doubt the best way humans can think of to do something is the best way in general.
Would you agree it's possible that any definitive theory for the structure of human embodied intelligence (be it as a distributed, modular or hierarchal system) may still not be optimal for robots?
In particular, a generalized data -> action (VLA-style) humanoid should still work in any case given sufficient training data