5 ms·

Has the potential for this approach been stripped, or does it just have a different name now? There are matrices and feedback loops. What was the key insight th

by ipnon 7y ago

Has the potential for this approach been stripped, or does it just have a different name now? There are matrices and feedback loops. What was the key insight that kept cybernetics from following machine learning towards neural nets?

api 7y ago
It's been renamed several times as things like complexity science, forked into spinoffs like artificial life and dynamical systems theory, and it crosses over into game theory quite a bit. The whole field is fascinating, but it's got a bad case of neologismania. That's a term I just made up to describe the tendency in some scientific fields to constantly make up new terms even if they're redundant.
- afc 7y ago
  Thank you for writing your comment. It sounds like you're familiar with this field. Is there some book you'd recommend (for learning about this topic, by whatever names) above this one? Thanks in advance!
  - patcon 7y ago
    Re: Complexity science. I highly recommend you always perk your ears up when someone mentions the Santa Fe Institute! Cormac McCarthy hangs out with that crowd: https://www.santafe.edu/news-center/news/cormac-mccarthy-publishes-his-first-science-nonfiction-nautilus https://www.santafe.edu/news-center/news/cormac-mccarthy-pub... The have a great podcast called Complexity: https://www.santafe.edu/culture/podcast https://www.santafe.edu/culture/podcast
  - stevenbedrick 7y ago
    Melanie Mitchell’s “Complexity: A Guided Tour” is a superb introduction to this area. And while I’m plugging her books, her latest one is one of the best-written lay-audience books about machine learning and AI that I’ve seen. I strongly recommended it for anybody curious about the field, and also for people working in the area- her explanations are crystal-clear but not dumbed down in the slightest. It’s called “Artificial Intelligence: A Guide for the Thinking Person.”
zozbot234 7y ago
Cybernetics is basically a quaint, archaic name for what we now call control theory. And control theory is basically everywhere these days. Though neural nets are probably way too lacking in robustness, explainability, etc. to be of real interest to control theory folks, or in control-related applications more generally. (Yes, I know they're being used for computer vision, but I don't think anyone is really satisfied with this state of things.)
- cma 7y ago
  In the book Cybernetics Weiner examines work on random circuit theory with tunable weights, mentioned as a potential mechanism for replicating black box behavior and even potentially being how life replicates (it was written pre-Watson and Crick). Link to description of that chapter: https://en.wikipedia.org/wiki/Cybernetics:_Or_Control_and_Communication_in_the_Animal_and_the_Machine#On_Learning_and_Self-Reproducing_Machines https://en.wikipedia.org/wiki/Cybernetics:_Or_Control_and_Co... (I think it is the Dennis Gabor part)
lidHanteyk 7y ago
Second-order cybernetics happened, and as a result there was a split. Like neighboring comments say, there is control theory and game theory for first-order cybernetics, but second-order cybernetics is extremely philosophical and failed to make an impact aside from broad ideas like the emergence of complex behavior from simple designs. The Constructivist movement has a journal [0] and is still alive, but I'm not sure what impact they are having. [0] https://constructivist.info/ https://constructivist.info/
- di4na 7y ago
  I would like to point out that second order cybernetics is alive and well in System Thinking, which you will find in Safety-II, Safety-differently, Resilience Engineering, etc etc
- metroholografix 7y ago
  One could say that second-order cybernetics has never been more relevant. Sometimes you see so far ahead that it takes time for your vision and its implications to fully become apparent. Second-order cybernetics defines not only today's technological substrate (especially obvious in cybersecurity and risk modeling) but also contemporary culture.
- someguyorother 7y ago
  It feels like elements of that split was always there. You have, on the one hand, very practical engineering cybernetics, what's standard control theory now (PID, optimal linear control, impulse responses and transfer functions, and so on). On the other hand, you have the more general (philosophical, "woolly") part that's filled with comparisons to biology, that deals with society and variety, general patterns, etc. I once gave a friend of mine Medina's "Cybernetic Revolutionaries", and I think he found it hard to get through because of the latter kind of cybernetics, even though he was very familiar with the former kind.
  - zozbot234 7y ago
    > On the other hand, you have the more general (philosophical, "woolly") part that's filled with comparisons to biology But see, that "other" part is also really a generalized/philosophical/speculative etc. take on control theory. "Systems theory" in the cybernetics-related sense is basically a name for 'this control theory stuff is so cool/impressive, let's see if we can generalize it beyond its standard engineering domain and draw any uselful conclusions from it.' Of course this kind of interdisciplinarity would be seen as standard nowadays, you wouldn't need to call attention to it by trying to name a new field.

by ipnon 7y ago

api 7y ago

It's been renamed several times as things like complexity science, forked into spinoffs like artificial life and dynamical systems theory, and it crosses over into game theory quite a bit. The whole field is fascinating, but it's got a bad case of neologismania. That's a term I just made up to describe the tendency in some scientific fields to constantly make up new terms even if they're redundant.

afc 7y ago
Thank you for writing your comment. It sounds like you're familiar with this field. Is there some book you'd recommend (for learning about this topic, by whatever names) above this one? Thanks in advance!
- patcon 7y ago
  Re: Complexity science. I highly recommend you always perk your ears up when someone mentions the Santa Fe Institute! Cormac McCarthy hangs out with that crowd: https://www.santafe.edu/news-center/news/cormac-mccarthy-publishes-his-first-science-nonfiction-nautilus https://www.santafe.edu/news-center/news/cormac-mccarthy-pub... The have a great podcast called Complexity: https://www.santafe.edu/culture/podcast https://www.santafe.edu/culture/podcast
- stevenbedrick 7y ago
  Melanie Mitchell’s “Complexity: A Guided Tour” is a superb introduction to this area. And while I’m plugging her books, her latest one is one of the best-written lay-audience books about machine learning and AI that I’ve seen. I strongly recommended it for anybody curious about the field, and also for people working in the area- her explanations are crystal-clear but not dumbed down in the slightest. It’s called “Artificial Intelligence: A Guide for the Thinking Person.”

zozbot234 7y ago

Cybernetics is basically a quaint, archaic name for what we now call control theory. And control theory is basically everywhere these days. Though neural nets are probably way too lacking in robustness, explainability, etc. to be of real interest to control theory folks, or in control-related applications more generally. (Yes, I know they're being used for computer vision, but I don't think anyone is really satisfied with this state of things.)

cma 7y ago
In the book Cybernetics Weiner examines work on random circuit theory with tunable weights, mentioned as a potential mechanism for replicating black box behavior and even potentially being how life replicates (it was written pre-Watson and Crick). Link to description of that chapter: https://en.wikipedia.org/wiki/Cybernetics:_Or_Control_and_Communication_in_the_Animal_and_the_Machine#On_Learning_and_Self-Reproducing_Machines https://en.wikipedia.org/wiki/Cybernetics:_Or_Control_and_Co... (I think it is the Dennis Gabor part)

lidHanteyk 7y ago

Second-order cybernetics happened, and as a result there was a split. Like neighboring comments say, there is control theory and game theory for first-order cybernetics, but second-order cybernetics is extremely philosophical and failed to make an impact aside from broad ideas like the emergence of complex behavior from simple designs. The Constructivist movement has a journal [0] and is still alive, but I'm not sure what impact they are having. [0] https://constructivist.info/ https://constructivist.info/

di4na 7y ago
I would like to point out that second order cybernetics is alive and well in System Thinking, which you will find in Safety-II, Safety-differently, Resilience Engineering, etc etc
metroholografix 7y ago
One could say that second-order cybernetics has never been more relevant. Sometimes you see so far ahead that it takes time for your vision and its implications to fully become apparent. Second-order cybernetics defines not only today's technological substrate (especially obvious in cybersecurity and risk modeling) but also contemporary culture.
someguyorother 7y ago
It feels like elements of that split was always there. You have, on the one hand, very practical engineering cybernetics, what's standard control theory now (PID, optimal linear control, impulse responses and transfer functions, and so on). On the other hand, you have the more general (philosophical, "woolly") part that's filled with comparisons to biology, that deals with society and variety, general patterns, etc. I once gave a friend of mine Medina's "Cybernetic Revolutionaries", and I think he found it hard to get through because of the latter kind of cybernetics, even though he was very familiar with the former kind.
- zozbot234 7y ago
  > On the other hand, you have the more general (philosophical, "woolly") part that's filled with comparisons to biology But see, that "other" part is also really a generalized/philosophical/speculative etc. take on control theory. "Systems theory" in the cybernetics-related sense is basically a name for 'this control theory stuff is so cool/impressive, let's see if we can generalize it beyond its standard engineering domain and draw any uselful conclusions from it.' Of course this kind of interdisciplinarity would be seen as standard nowadays, you wouldn't need to call attention to it by trying to name a new field.