Multiple Studies Now Suggest That AI Will Make Us Morons

Marantz Cinema Series AV Receivers
Meanwhile, AI, sitting in the corner, grinning an evil grin, munching on some tasty bits, reading this thread.
 
Analogous said:

AI Has Improved Monsoon Forecast Accuracy in India by 20 Percent​

Click to expand...
How come? What has changed suddenly?
Weather forecasting always relied on statistical analysis, since the weather models follow chaotic patterns.
AI is a subset of statistical analysis, and most of the AI algorithms have been available for decades, just that we didn't have enough cheap compute power. But IMD always had access to super computers.
 
alpha1 said:
How come? What has changed suddenly?
Weather forecasting always relied on statistical analysis, since the weather models follow chaotic patterns.
AI is a subset of statistical analysis, and most of the AI algorithms have been available for decades, just that we didn't have enough cheap compute power. But IMD always had access to super computers.
Click to expand...
I don’t have a clue.
The article does mention improvements in forecasts by 20%.
Maybe someone with more knowledge on the subject could explain.
Interestingly, the link to the article seems to be not linked anymore
 
alpha1 said:
How come? What has changed suddenly?
Weather forecasting always relied on statistical analysis, since the weather models follow chaotic patterns.
AI is a subset of statistical analysis, and most of the AI algorithms have been available for decades, just that we didn't have enough cheap compute power. But IMD always had access to super computers.
Click to expand...

Traditional weather modeling is based on Physics. Modeling the atmosphere with partial differentials and fluid dynamics. Statistics was used to normalise the output of these models. Supercomputers were huge banks of CPUs capable of running these complex equations. The problem here is that even the best supercomputers were only as good as the "model" the scientist feed it.

Typical AI based weather prediction is based on neural networks. You "train" the network with high quality labled data about observed weather over multiple decades. The network "learns" how a set of inputs affects outcomes. Traditional weather modeling required scientist to understand how weather works from a physics perceptive and model it. Nerual networks on the other hand will "learn" brute force with billions of "examples". It's knowledge is in the form of weights in a complex neural network. But this is a black box. No one knows how it learnt nor will it be able to give explaination as to how it came to a conclusion. But this works. If we have high quality, billions of data points to feed them, they perform remarkable well. Till around 2010, there was no hardware available to create massive deep neural networks. But since the advent of GPUs and their massive number crunching capability, this has become a reality. Apart from military use, most decision makers believe that a black box is fine. It really does not matter for the network to give us how it came to the conclusion as long as its output is useful and correct.
 
arunvenkats said:
Traditional weather modeling is based on Physics. Modeling the atmosphere with partial differentials and fluid dynamics. Statistics was used to normalise the output of these models. Supercomputers were huge banks of CPUs capable of running these complex equations. The problem here is that even the best supercomputers were only as good as the "model" the scientist feed it.

Typical AI based weather prediction is based on neural networks. You "train" the network with high quality labled data about observed weather over multiple decades. The network "learns" how a set of inputs affects outcomes. Traditional weather modeling required scientist to understand how weather works from a physics perceptive and model it. Nerual networks on the other hand will "learn" brute force with billions of "examples". It's knowledge is in the form of weights in a complex neural network. But this is a black box. No one knows how it learnt nor will it be able to give explaination as to how it came to a conclusion. But this works. If we have high quality, billions of data points to feed them, they perform remarkable well. Till around 2010, there was no hardware available to create massive deep neural networks. But since the advent of GPUs and their massive number crunching capability, this has become a reality. Apart from military use, most decision makers believe that a black box is fine. It really does not matter for the network to give us how it came to the conclusion as long as its output is useful and correct.
Click to expand...
As much as I would love to believe it, I find it difficult to believe.
The CFD requirement for a simple small, isolated body is so intense (e.g. modeling the flow on F1 car) that it requires hours to generate results on multi core, multi GPU machines of today. Also let's not forget the intense effort required to create the finite element meshing and other processing work in order to make the CFD converge to a solution.

Doing this entire process for atmospheric conditions (and geographic boundary layers like sea, hills etc) appear madness to me!
 
alpha1 said:
As much as I would love to believe it, I find it difficult to believe.
The CFD requirement for a simple small, isolated body is so intense (e.g. modeling the flow on F1 car) that it requires hours to generate results on multi core, multi GPU machines of today. Also let's not forget the intense effort required to create the finite element meshing and other processing work in order to make the CFD converge to a solution.

Doing this entire process for atmospheric conditions (and geographic boundary layers like sea, hills etc) appear madness to me!
Click to expand...

It is this madness which has been helping us be safe from the vagaries of weather!

Refer to this:
www.encyclopedie-environnement.org

Weather forecasting models - Encyclopedia of the Environment

Weather forecasting is also done through numerical modelling of the atmosphere. What are the basic principles of this modeling?
www.encyclopedie-environnement.org www.encyclopedie-environnement.org
In the case of the atmosphere, the laws of fluid mechanics are formulated to predict atmospheric parameters such as wind, temperature, pressure and humidity. These are very complex equations because there are many cross interactions between these parameters. These equations also involve processes at very different scales, from the scale of the planet to that of the raindrop, as well as interactions with the underlying surface (land, sea, vegetation cover, see Biosphere, hydrosphere and cryosphere models) and space. Their extreme complexity, due in particular to their non-linearity, precludes their analytical solution. The only way out is therefore to use approximate numerical techniques to calculate the evolution of an initial state. This initial state, called analysis, is itself manufactured using highly sophisticated mathematical and numerical methods for assimilating atmospheric observations (see Assimilation of Meteorological Data).
Click to expand...

And more info about current AI models:
deepmind.google

GraphCast: AI model for faster and more accurate global weather forecasting

Our state-of-the-art model delivers 10-day weather predictions at unprecedented accuracy in under one minute
deepmind.google deepmind.google

Regards,
Arun
 
Rega Turntables & Amplifiers
Order your Rega Turntables & Amplifiers from HiFiMART.com - India's reputed online dealer.

Similar threads

Hari Iyer
Q & A - Multiple choice after post #10
Replies
0
Views
4K
Hari Iyer
Hari Iyer
Analogous
Insect Welfare
Replies
2
Views
4K
mbhangui
mbhangui
A
Audio system for a beginner under 1lakh
Replies
8
Views
3K
DB1989
D
Analogous
Music and productivity? Read this before you make your playlist
Replies
0
Views
350
Analogous
Analogous
Analogous
The beats that make us want to move
Replies
0
Views
184
Analogous
Analogous
Back
Top