My tutorial on, and assessment of, the current status of AI is available at:
Cheers,
Earl
My tutorial on, and assessment of, the current status of AI is available at:
Cheers,
Earl
Very interesting stuff.
Thanks. Tell your friends
Cheers,
Earl
Excellent explanation
Should be a compulsory read for science students, particularly those who may be attempting to model complex systems and the inherent risks in the choice and use of training material.
Referencing Turing, I used to tell my students that a model is a approximation of the real system and it’s usefulness for an application is dependent on the level of repeatable accuracy required for that application.
Can You confirm please it also applies to climate models.??
Very interesting article about a complex subject, covers a lot of ground.
A good example of the extreme difficulty of modeling complex systems.
And
How accurate does the model have to be, to be useful ?
Conops.
A long but very interesting, and educational, read.
It reminded me of a small sidebar on a page in FATHOM, a publication of the US Naval Safety Center, back in the 1970s. It was titled “Rules of Thumb and Ballpark Estimates” and contained these two bits that I have never forgotten.
“It is better to solve a problem with a crude approximation and know the truth, plus or minus ten percent, than to demand an exact solution and never know the truth at all”.
“An easily understood, workable falsehood is more useful than a complex, incomprehensible truth”.
More useful or will get you into trouble quicker, seems to me. Depending on the degree of falsity.
Dead Reckoning useful
Just not for interplanetary travel, I think.
Thanks for this very thoughtful essay!
Seems that a great number of “…easily understood, workable falsehoods” of a rather high degree of falsity have proven very useful in many political and social areas in recent times. JMHO
Celestial navigation is a complex workable falsehood. It assumes that the Earth is the center of the universe and that the universe revolves around the Earth.
So is terrestrial navigation if you get in the weeds of how accurate charts and publications really are (they arent).
Updated:
Cheers,
Earl
This absolutely reeks of common sense. Also in a way reminds me of Conway’s Law. The product reflects the org structure that created it. No concern for “safety” during development = unsafe product?
OT: Conway was quite a character. He wrote an assembler for the Burroughs 220 called “SAVE.” The name didn’t mean anything, you just lost fewer card decks and listings because they all started with the word SAVE. I still have my copy of the manual.
Cheers,
Earl
It depends on what you are trying to do.
No climate model in existence will tell me if it will be a nice day to take my wife sailing 1 month from now and there probably never will be one that can forecast accurate weather for one spot 30 days out, the system is too chaotic.
OTOH I used to walk across the river in January and now we sail 12 months most years
I have an AI project going now to read 4 GPS inputs and throw out the outlier and average the closest 3. Mainly just for fun and the AI is not ever correct on the first pass, but it saves me a lot of grunt work. I wouldn’t have spare time for this otherwise.