The Danger of Leaving Weather Prediction To AI (wired.com) 46
When it comes to forecasting the elements, many seem ready to welcome the machine. But humans still outperform the algorithms -- especially in bad conditions. From a report: [...] Similarly, research published by NOAA Weather Prediction Service director David Novak and his colleagues show that while human forecasters may not be able to "beat" the models on your typical sunny, fair-weather day, they still produce more accurate predictions than the algorithm-crunchers in bad weather. Over the two decades of information Novak's team studied, humans were 20 to 40 percent more accurate at forecasting near-future precipitation than the Global Forecast System (GFS) and the North American Mesoscale Forecast System (NAM), the most commonly used national models. Humans also made statistically significant improvements to temperature forecasting over both model's guidance. "Oftentimes, we find that in the bigger events is when the forecasters can make some value-added improvements to the automated guidance," says Novak. Particularly in adverse conditions, great improvements to the model's forecast were usually due to human augmentation, he adds. This is even more true for local, severe events like thunderstorms and tornadoes, which rely on split-second decision-making in order to save lives. As forecasters become more familiar with a particular model, they begin to notice its biases and failings, Novak adds. Just like the model learns from us, we learn from the model.
Not "ai" (Score:5, Informative)
I know everything gets labeled "ai" now. But GFS is definitely not "ai". It is a model, and is essentially deterministic (using differential equations to describe weather behavior).
Re: (Score:2)
Re: (Score:2)
Why wouldn't be AI?
Is it artificial?
Obviously, yes. It's man-made.
Does it produce output or do what corresponds with what an intelligent entity would have (or could have) produced or done?
Yes?
Then it's AI. Full stop.
How it gets there is irrelevant. How do you even know that other human beings are intelligent, for example?
Re: Not "ai" (Score:5, Insightful)
In that case presumably you think a pocket calculator is AI too? If you define any program that can execute mathematical operations as AI the term becomes meaningless.
Re: (Score:3)
That depends whether the relevant software is executing said mathematical operation through information it has autonomously learned, as memory is a key component of intelligence.
If the calculator learned how to do those operations by observing them being done intelligently and then applied those observations to be able to perform operations on new data in ways that correspond with what an intelligent entity would have otherwise done, then yes. .
But I don't think your typical pocket calculator would fit
Re: Not "ai" (Score:2)
Neither does a weather model. It doesnt learn anything, it gets spoonfed data from satellites, ships etc.
Re: (Score:2)
AI does not mean learning.
(Artificial) Neural Networks and machine learning, aka deep learning, is just a tiny subset of AI.
Re: Not "ai" (Score:1)
No, that's not AI. That's an expert system.
To qualify as AI, it must be indistinguishable from an intelligent person, not just perform a specific task of one.
It's the difference between two functions being the same versus intersecting at a point.
Re: (Score:3)
Thats not AI , thats the Turing test. Which most researchers do not consider a valid test for AI, as it only tests for the abilitty to emulate humans, which is specialized AI (AI that only solves one problem), not general AI (Which can solve any).
Re: (Score:2)
There is no formal definition of intelligence, other than Turing's, which is basically a mathematical one. If f(x)=g(x) for all x, then f=g.
Re: (Score:2)
But if intelligence exists, there is no reason that AI cannot.
Re: (Score:2)
That's absolutely true.
Re: (Score:2)
The articles was written by people with little or no understand of what AI is. I'm almost entirely sure they were in general thinking the technology they were comparing themselves to and their limited understanding of the technology was AI rather than machine learning... which would still be wrong.
As was mentioned by someone earlier, we tend to use a whole lot of partial differential equations to predict whether.
I'm not entirely sure how to accurately write i
Re: (Score:2)
It's not intelligent.
Re: (Score:2)
It's just a clickbait headline. The article doesn't say the models are AI - it only strongly implies it to get the click. The only place AI appears is in the headline.
Re: (Score:1)
Al Gore?
Re: (Score:2)
A meteorologist does the same thing (or used to, before computers). Why would you consider the work of a meteorologist to be the product of intelligence but the work of a machine to not be?
Or would you say that meteorologists are not intelligent either?
Hybrid Decision Making (Score:3)
Ignoring the whole AI vs algorithm debate, isn't it usually the case that a combination of algorithm plus human judgement yields the best results? Whenever I am working on projects which could be considered decision management systems, one of the most important aspects is identifying ways for human operators to enhance the results of automated systems. Rarely is human judgement or automated results better than a combination of the two.
Re: (Score:3)
Re: Hybrid Decision Making (Score:4, Interesting)
With weather, you're absolutely correct. You use a monte carlo simulation of a few billion possible starting conditions and external variables. These produce a range of possible outcomes.
Humans are very good at identifying clusters of very very improbable outcomes and can eliminate them. Expert systems are very poor at this, at this time and it's this noise that will throw current systems.
If the weather is unstable, you'll get a lot more of this noise because, weather being a chaotic system, many small errors get amplified quickly. Therefore your initial randomised assumptions have to magically be near perfect. This means you need a lot more tries, most of which will be badly wrong, so the signal is weak.
Whereas, for calm conditions, the amplification is minimal so, over the period on question, you can afford to have your initial guesses be a lot more inaccurate and still get good results. So most of your guesses will be good enough and the right answer will be glaringly obvious.
Monte Carlo?! Its a huge finite element model (Score:2)
What makes you think that has anything to do with it.
My understanding is that it solves a huge differential equation. The atmosphere is modeled as roughly kilometer cubed chucks of air, and each cube interacts with all the others in a fairly unstable way.
I am sure there is a lot more to it. But Monte Carlo?
As to human in the loop, what is certainly true is that human + computer easily beats computer playing chess.
Re: (Score:2)
Because that's how it's done. The computer gets only approximate values and only for some locations and performs billions of simulations to see what happens under different initial conditions. This is what the Met Office describes.
Re: (Score:1)
Wrong comparison? (Score:5, Interesting)
On the other hand, it's well known [forbes.com] (many other citations available) that the "most commonly used national models" perform worse than the European and British models, so how do they stack up against the human predictions.
Re: Wrong comparison? (Score:5, Funny)
All the british models have to do is print "Chance of rain" as their output and theyd usually be right.
Re: (Score:2)
Only when there's no fog.
Re: Wrong comparison? (Score:2)
The British use high-end supercomputers. Back in the 1980s, these would be a generation or two more advanced than the American systems. And Britain is an awful lot smaller, with a relatively simple topography.
Computers plus humans are still the best way to go, though. Even so, as fans of Michael Fish will have to admit, spectacular blunders are made.
Re: (Score:3)
I should have been more explicit. The European and British models do better than the US ones at predicting weather in the USA. Rather embarrassing for the American meteorologists.
USA is more used (Score:2)
But most weather sites use the USA model? Why? Because their is no crown copyright on them.
Re: (Score:2)
Most US weather forecasters get the results of two or more models, and then make their own human predictions based on those results as inputs. I often here the local weather forecaster show a map of predicted weather and say something like "the European model tends to overestimate the amount of snow in situations like this" or "don't get hung up on exactly where the big rainfalls are showing, the exact locations of the heaviest rains is impossible to predict".
Re: (Score:2)
Re: (Score:2)
Well,
during recent decades of climate research we dropped weather metering stations everywhere.
But USA thought it is enough to have a few thermometers which are hand - oh eye - read on a random point in time each day, and the results get phoned to someone who is recording them and putting them into a database.
Here we have real time metering of the core data in a very small grid, supported by radar and, satellites.
Weather reports can be kind of accurate one or even two weeks ahead.
Well, AI does have its limitations (Score:2)
Re: (Score:2)
The Danger of Clickbait (Score:2)
Re: (Score:2)
Probably the best way to train an AI is to keep the models around. Just train it with the data set of what the different models predicted along with what actually happened.
Forecasting has gone stupidly specific in bad ways (Score:2)
Thursday night they predict another 2.5 inches, and I really believe that we might get anything given their recent failings.
Re: (Score:2)
this has been a trend for years to the point where i dont bother looking at the weather forecast anymore, just look outside and make my own decisions. i am about as accurate, maybe a little more, than the weather apps these days.
Easy fix, (Score:4, Funny)
just a hire a stable genius with a Sharpie.
I have a weather rock (Score:2)
I became disillusioned with incorrect weather information and stupid weather presenters so I invested in a weather rock. [wikipedia.org] If it's wet, it's raining. Moving? It's windy. It does other things and has completely revolutionized how I think of weather.
Best of all it requires no batteries or software to work.
Re: (Score:2)
"1.34 inches of snow from 12:02-1:12" - gets 0" of snow.
in fact yesterdays "evening forecast" showed Snow.. when looking closer they predicted a 40% chance of .02 inches of snow at 7PM. did we get any snow? no, it was nearly 40F... wtf?
Re: (Score:2)
My weather rock is blowing and dry but cold, it never fails me.
So what about climate predictions? (Score:1)
Aren't those done with computer models? How is their veracity now defensible, especially given their abysmal track record thus far?
The danger of leaving ANYTHING to 'AI' (Score:2)