__________
#8 | Posted by sitzkrieg at 2026-10-08 07:59 AM
here's how accuracy, precision, and recall in machine learning is really measured. It's not binary.
Definitely not binary... nor a decision-tree - it's fuzzy logic, with help from Bayesian Networks and Confusion matrix.
en.wikipedia.org - Fuzzy Logic
|------- Many of the early successful applications of fuzzy logic were implemented in Japan. A first notable application was on the Sendai Subway 1000 series, in which fuzzy logic was able to improve the economy, comfort, and precision of the ride. It has also been used for handwriting recognition in Sony pocket computers, helicopter flight aids, subway system controls, improving automobile fuel efficiency, single-button washing machine controls, automatic power controls in vacuum cleaners, and early recognition of earthquakes through the Institute of Seismology Bureau of Meteorology, Japan. ...
Neural networks based artificial intelligence and fuzzy logic are, when analyzed, the same thing - the underlying logic of neural networks is fuzzy. ...
In the 1980s, researchers were divided about the most effective approach to machine learning: decision tree learning or neural networks. The former approach uses binary logic, matching the hardware on which it runs, but despite great efforts it did not result in intelligent systems. Neural networks, by contrast, did result in accurate models of complex situations...
They can also now be implemented directly on analog microchips, as opposed to the previous pseudo-analog implementations on digital chips. ...
-------|
Some of the better books on the subject were Neural Networks and Fuzzy Systems: A Dynamical Systems Approach to Machine Intelligence (1991), Fuzzy Thinking: The New Science of Fuzzy Logic (1993) and Heaven in a Chip: Fuzzy Visions of Society and Science in the Digital Age (2000) by Bart Kosko (en.wikipedia.org).
#13 | Posted by LampLighter at 2026-10-09 07:41 PM
Oh please, AI has only been around for a decade or so.
Where did you get that idea? N2 and ML and have been around and used in complex systems (like forecasting weather) for decades, but they've been extremely expensive, limited by today's standards, unfamiliar and unavailable for wide commercialization (and recognition) until few years ago... same as "computers" were before PCs became widely available and affordable for mass personal use in late 1970s - early 1980s.
#16 | Posted by Dbt2 at 2026-10-09 07:52 PM
So how much reliable input does AI have to go on here? And in climate, the future is not going to be acting like in the past, because there's no recorded precedent.
"The past" has very little to do with most of the inputs, unless you're talking about some well-documented phenomena like El Nino or La Nina (ENSO), Gulf Stream, AMOC, etc., which have some known and predictable patterns that could be used as part of the input data.
__________
__________
#8 | Posted by sitzkrieg at 2026-10-08 07:59 AM
here's how accuracy, precision, and recall in machine learning is really measured. It's not binary.
Definitely not binary... nor a decision-tree - it's fuzzy logic, with help from Bayesian Networks and Confusion matrix.
en.wikipedia.org - Fuzzy Logic
|------- Many of the early successful applications of fuzzy logic were implemented in Japan. A first notable application was on the Sendai Subway 1000 series, in which fuzzy logic was able to improve the economy, comfort, and precision of the ride. It has also been used for handwriting recognition in Sony pocket computers, helicopter flight aids, subway system controls, improving automobile fuel efficiency, single-button washing machine controls, automatic power controls in vacuum cleaners, and early recognition of earthquakes through the Institute of Seismology Bureau of Meteorology, Japan. ...
Neural networks based artificial intelligence and fuzzy logic are, when analyzed, the same thing - the underlying logic of neural networks is fuzzy. ...
In the 1980s, researchers were divided about the most effective approach to machine learning: decision tree learning or neural networks. The former approach uses binary logic, matching the hardware on which it runs, but despite great efforts it did not result in intelligent systems. Neural networks, by contrast, did result in accurate models of complex situations...
They can also now be implemented directly on analog microchips, as opposed to the previous pseudo-analog implementations on digital chips. ...
-------|
Some of the better books on the subject were Neural Networks and Fuzzy Systems: A Dynamical Systems Approach to Machine Intelligence (1991), Fuzzy Thinking: The New Science of Fuzzy Logic (1993) and Heaven in a Chip: Fuzzy Visions of Society and Science in the Digital Age (2000) by Bart Kosko (en.wikipedia.org).
#13 | Posted by LampLighter at 2026-10-09 07:41 PM
Oh please, AI has only been around for a decade or so.
Where did you get that idea? N2 and ML and have been around and used in complex systems (like forecasting weather) for decades, but they've been extremely expensive, limited by today's standards, unfamiliar and unavailable for wide commercialization (and recognition) until few years ago... same as "computers" were before PCs became widely available and affordable for mass personal use in late 1970s - early 1980s.
#16 | Posted by Dbt2 at 2026-10-09 07:52 PM
So how much reliable input does AI have to go on here? And in climate, the future is not going to be acting like in the past, because there's no recorded precedent.
"The past" has very little to do with most of the inputs, unless you're talking about some well-documented phenomena like El Nino or La Nina (ENSO), Gulf Stream, AMOC, etc., which have some known and predictable patterns that could be used as part of the input data.
__________