Essay by Eric Worrall
Why did they train the AI initially using climate models? Why not ignore the models, and directly use the observations to directly train the AIs?
AI research predicts planet will warm faster than expected
By CNN 11:27am Jan 31, 2023
…
The study estimates that the planet could reach 1.5 degrees Celsius of warming above pre-industrial levels in a decade, and found a “substantial possibility” of global temperature rises crossing the 2 degrees threshold by mid-century, even with significant global efforts to bring down planet-warming pollution.
Data shows average global temperature has already climbed risen around 1.1 degrees to 1.2 degrees since industrialisation.
“Our results provide further evidence for high-impact climate change, over the next three decades,” noted the report, published on Monday in the journal the Proceedings of the National Academy of Sciences.
…
Read more: https://www.9news.com.au/world/climate-change-artificial-intelligence-dire-forecast-for-planet-future/764893d8-2a02-4214-8980-d1e4a4b5ca87
The abstract of the study;
Data-driven predictions of the time remaining until critical global warming thresholds are reached
Noah S. Diffenbaugh and Elizabeth A. Barnes
Edited by Michael Mann, The Pennsylvania State University, University Park, PA; received April 25, 2022; accepted December 14, 2022
January 30, 2023
120 (6) e2207183120
Significance
The United Nations Paris Agreement aims to hold global warming well below 2 °C and pursue 1.5 °C. Given the clear evidence for accelerating climate impacts, the time remaining until these global thresholds are reached is a topic of considerable interest. We use machine learning methods to make truly out-of-sample predictions of that timing, based on the spatial pattern of historical temperature observations. Our results confirm that global warming is already on the verge of crossing the 1.5 °C threshold, even if the climate forcing pathway is substantially reduced in the near-term. Our predictions also suggest that even with substantial greenhouse gas mitigation, there is still a possibility of failing to hold global warming below the 2 °C threshold.
Abstract
Leveraging artificial neural networks (ANNs) trained on climate model output, we use the spatial pattern of historical temperature observations to predict the time until critical global warming thresholds are reached. Although no observations are used during the training, validation, or testing, the ANNs accurately predict the timing of historical global warming from maps of historical annual temperature. The central estimate for the 1.5 °C global warming threshold is between 2033 and 2035, including a ±1σ range of 2028 to 2039 in the Intermediate (SSP2-4.5) climate forcing scenario, consistent with previous assessments. However, our data-driven approach also suggests a substantial probability of exceeding the 2 °C threshold even in the Low (SSP1-2.6) climate forcing scenario. While there are limitations to our approach, our results suggest a higher likelihood of reaching 2 °C in the Low scenario than indicated in some previous assessments—though the possibility that 2 °C could be avoided is not ruled out. Explainable AI methods reveal that the ANNs focus on particular geographic regions to predict the time until the global threshold is reached. Our framework provides a unique, data-driven approach for quantifying the signal of climate change in historical observations and for constraining the uncertainty in climate model projections. Given the substantial existing evidence of accelerating risks to natural and human systems at 1.5 °C and 2 °C, our results provide further evidence for high-impact climate change over the next three decades.
…
Read more: https://www.pnas.org/doi/full/10.1073/pnas.2207183120
To their credit they have published their code on github. “… Code is available on GitHub at https://github.com/eabarnes1010/target_temp_detection (60) and is archived on Zenodo at the following DOI: https://doi.org/10.5281/zenodo.7510551 (61).”.
My understanding is the researchers are attempting to use the AI to identify climatologically significant geographic regions, or distributions of observations, to try to filter out the noise and reduce the uncertainty of predictions.
My concern with this approach is if the data was sufficient for tuning predictions, the AIs could be trained directly on the data, the AIs could infer climate models directly from the data.
Using a simulation or model allows a large number of training runs to be packed into a short period of time, and constrains the output of the AI. But the AI is then tainted by the model, it effectively becomes an extension of the model.
I guess time will tell whether their approach has yielded increased predictive skill.
Imagine that. AI that generates a hockey stick.
AI is fundamentally a computer program. The term GIGO comes to mind.
The same garbage that Mann uses.
That’s what I was thinking. Garbage in, Garbage out.
GIGO is right. Artificial intelligence is just simulated intelligence. It’s only as good as the programmer and the data available, as is the case with climate models. Computers cannot “think” and are not magic. They operate on logic circuits: Ands, Ors, and Nots, which can be used to do things like arithmetic, and do them very fast. But the results are deterministic. Given the same inputs and data, you will get the same output, unless there is a flaw in the circuitry. So computers will only do what they are programmed to do. They can never become more intelligent than their programmers. And they will make the same mistakes their programmers make, just much faster. That’s why computer climate models cannot be trusted until their programmers actually understand everything about the climate. And that isn’t going to happen anytime soon.
Despite science fiction, computers will not “evolve” intelligence or learn to think for themselves and then seek to destroy their creators. So we can rest easy, at least until they get quantum computing to work. I haven’t studied that, so I have no idea what quantum computers are capable of.
Astonishing, isn’t it? The first and most basic rule of data analysis is garbage in, garbage out. Data collection is the first, and most critical, link in the chain of reasoning from empirical evidence.
It’s Models! Models, all the way down!
They fed the output of models that give exaggerated warming into the AI and the AI results are even more exaggerated warming. Next they will feed the AI results back into the the models. How can this possibly work?
You’ve just invented a new type of positive feedback loop. With any luck, the models will run into uncontrolled oscillation, and crash 🙂
Shades of Nicolas Tesla!
Sadly, This model’s incestuous self satisfaction endless loop whirlpool is evidence of Mann.
Also sadly, It’s apparent they never instilled scientific absolute rigor regarding “Observations” as reality. Observations trump self delusion. Feynman described this nonsense perfectly.
There’s no such thing as AI … just agendas.
Where artificial intelligence is possible, artificial stupidity will dominate.
Artificial stupidity is superfluous.
The popular consensus assumes a conflation of origin and expression that is an article of faith, not science.
That said, intelligence as in knowledge, yes. AI as in a cache of correlations with fixed degrees of freedom, sure.
The “A” in “AI” could just as readily be an acronym for Arranged, Automated, Arrogant, Autocratic etc. “I” can just as easilu stand for “Instruction” too. I take my cue from realising that MBA is also and probably more accurately in many cases an acronym for Mindless Bloody Autocrat let alone Masturbating Bulls*^>>ing A&^%hole. ( Sorry to all the decent MBA’s out there).
I think Frank Herbert had it about right back in the 60’s with his background history to the Dune novel recalling the ‘war against the tinking machines’ that saw Earth destroyed and humanity fleeing to interstellar-integalactic space to survive.
On another front I recall the story about some snake slithering in to Paradise and suckering some naive couple that the apples were free. Snake, silicon based technology, what’s the difference? A world wide pandemic of imbesilicosis is upon us and we don’t have the ;uxury of living Paradise, well except for a few billionaires.
“Why did they train the AI initially using climate models? Why not ignore the models, and directly use the observations to directly train the AIs?”
They probably tried that and did not receive the desired result.
Because 30 to 50 years of climate observations have no value in predicting the next 30 to 50 years of climate change. They would work better as a contrary indicator, if they had any ability to predict the future climate. You don’t need scientists, confusers and AI for that.
Fancy that. Observations have no place in science. Richard Feynman would not be amused. But the Red Queen would.
““Why did they train the AI initially using climate models? Why not ignore the models, and directly use the observations to directly train the AIs?””
The “observations” are also suspect.
Why not feed the AI the results of all the unmodified, regional surface temperature charts from around the world. I bet they wouldn’t come up with a Hockey Stick profile, if they did.
Cos ‘artificial’ is just another word for pretend.
I don’t think the AI guys understand the difference between training, and entrainment. Just like they cannot discern between thinking and regurgitating memorised data. Or know data from facts.
The sad part is, near nobody else seem to make the distinction either… too well programmed with curated data, I suspect.
Yes, and because their Automated Idiocy will be seen as “credible” by the ignorant believers in human-induced climate catastrophe, therefore best to “stack the deck” of their modern day Tarot Cards.
That’s bad news for alarmist scientists. They are no longer needed to produce and report wild climate assessments. Now computers can program themselves! It would save a ton of money by cutting out the middleman.
How about an old lady and a Ouija board to predict the climate?
The Greek wife suggested an old Greek lady wearing a babushka scarf (The reading of fortunes in the grinds of a Greek coffee cup is a centuries-long tradition. Don’t ask me why.)
Edited by Michael Mann,
____________________
And that’s where I stopped
readingscanning.Michael Mann – now that right there is the anthropomorphic fingerprint.
A Disgrace to the profession.
Current climate change on “climatereanalyzer” (vs. 1979-2000):
+0.57K World
+1.00K NH
+0.14K SH
+0.21K Tropics
+3.47K Arctic
+0.01K Antarctic
Not the World, but the NH is warming. This is not the work of GHGs.
Kids? You don’t “train” computers, you program them. They do what they are programmed to do and nothing more. These are programmed to produce lies and lies are all they will produce.
So many people don’t trust computer programmers to define what computer programs do.
Most of the best developers (not “programmers”, there is a distinction) are deceased.
They’re talking about “training” the “algorithms.”
At its core, AI is largely a glorified word or phrase comparison tool with some “fuzzy logic” to recognize similar, but not quite identical, items.
Automated Idiocy.
Artificial intelligence is not so intelligent. It just seems so.
But it is artificial, so it has to be real. PoMo 101.
I guess maybe the tree rings were not cooperating.
What ECS did they use ? Cuz there is big difference in which decade a 1.5 C limit would be exceeded depending on the 1.5 to 6.5 degrees per CO2 doubling that ICCP says they aren’t sure of. Hmmm….and the editor is one M. Mann known for predicting world climate based on a couple of Bristlecone pine stumps despite Briffa’s different analysis.
“Data driven” predictions of the future climate are a joke
There are no data for the future climate
There has been no success predicting the future climate in the past century
Computers predict whatever they are programmed to predict
Climate computer models are programmed to scare people
They are not programed to appear accurate
Actually, there is no such thing as a climate model
Detailed knowledge of every climate change variable would be required to build such a model.
That knowledge does not exist.
Therefore, the programs called computer models are nothing more than computer games sed for climate change propaganda
If the knowledge to build a real climate model existed, I suspect it would still not be able to predict the future climate.
Why does anyone think the climate can be predicted?
We can’t predict whether or not it will snow in London one year from today.
So why should it be possible to predict the climate in 100 years?
How many more centuries of wrong climate predictions will be required before people finally reach the conclusion I reached in 1997:
“The climate will get warmer, unless it gets colder.”
The 24 best climate science and energy articles I read today:
Honest Climate Science and Energy
“Given the clear evidence for accelerating climate impacts, the time remaining until these global thresholds are reached is a topic of considerable interest.”
Nope. Yawn.
I am considerably interested, though, in stopping the push for intermittent, land-intensive and materials-intensive sources of supply to the electrical grid.
As someone from that industry, I wonder how “the electrical grid” came to replace “electric power”. Saying “the electrical grid” instead of “electricity” sometimes requires complicated sentence structure, but “the electrical grid” has clearly won. I blame content providers who need extra words to reach a target word count.
I use the term “electrical grid” in the sense of a regional system into which the electricity is supplied for revenue. I have no objection to a user generating and utilizing their own electric power from a solar array or wind turbine. It appears to the system as a load variation. Perhaps my usage sounds awkward to you, but it has nothing to do with content providers or word count.
The naked lies they regurgitate are so mind-numbingly stupid I can’t even read through the quoted articles any more. I just see red.
The misuse of the tools of science to fabricate evidence for one’s pet-cause is not science at all.
But they dress it up to make it look like science to politicians and other ignorati.
Mark Twain:
“There is something fascinating about science. One gets such wholesale returns of conjecture out of such a trifling investment of fact.”
Note the irony – a great quote about science from an expert at producing great quotes (MT was not a scientist).
Why not ignore the models, and directly use the observations to directly train the AIs?
Because observations aren’t warming fast enough. So make up stuff to scare the proletariat into compliance.
Story Tip: Does anybody know if wind generators in Texas are again freezing, and not working because of the current weather conditions taking place in that area?Several southern states, including Texas are experiencing ice storms. Does anybody know how this is affecting weather dependent green energy?
https://www.ercot.com/content/cdr/html/real_time_system_conditions.html
Right now wind output is < 2 GW of ~63 GW total ERCOT demand. I don’t know if it is the cold or just calm or calm because it is cold. I do see there is an ice storm going on.
Yes indeed. My concern was about the ice storm that this area is presently experiencing. We remember several years ago when both Texas and northern China, both experienced freezing rain episodes, which college their wind turbines to freeze and quit producing energy. I was wondering if this same condition is going on now in Texas. Thank you for replying.
One thing we know for sure about computers:
Garbage In; Garbage Out.
Climate models are garbage. AI programs feed climate models spout garbage.
Quod Erat Demonstrandum
Next.
Willy Soon says consensus climatology purveys Garbage In, Gospel Out.
A perfect summary of the case.
“We use machine learning methods to make truly out-of-sample predictions of that timing, […]”
Well there you go. In black and white they imply there is legitimate reason for thinking previous studies were not out-of-sample predictions.
“Hey, this time we’re telling the truth. Truly. Honestly. No, really, we are.”
I love the portentous trotting out of the Paris Agreement and its ability to keep global temp below 2degsC and 1.5degsC above pre industrial levels when we don’t know what the temperature was in 1850 and the effect on CO2 levels since COP1 in 1995 has been ZERO .
Biggest lie they’re selling is “warmer climate is worse.”
Which is why they draw the curtain across most of the Earth’s climate history to make their claims seem supportable. Even just including all of the current epoch requires uncomfortable explanations about why every previous warm period was referred to as a climate OPTIMUM.
This is behind absurd—a neural net is trained to reproduce the relationship between a set of inputs and their corresponding outputs. Pat Frank has already demonstrated that the big fancy climate models are nothing but linear projections of the CO2 concentration after just a few iterations. And the models are “tuned” using “observed” data.
Stuffing “observations” back into the input of this NN can only reproduce what it was trained with—circular reasoning at its worst.
There is nothing “AI” about this absurdity at all, and Mickey Mann was the sorcerer’s apprentice behind it all.
Deus Ex Machina
After training an AI “on climate model output” i.e., on climate model projections, Stanford’s Noah Diffenbaugh reports in PNAS that the AI predicts global warming.
https://news.stanford.edu/2023/01/30/ai-predicts-global-warming-will-exceed-1-5-degrees-2030s/
Well, that proves the case, doesn’t it.
And, taking a credibility victory lap, Diffenbaugh’s 30 January paper was “Edited by Michael Mann.” (doi: 10.1073/pnas.2207183120). He who faked the hockey stick, received the (forever besmirched) Hans Oeschger Medal for it in 2012, and who has done no science at all since his work on lanthanide ceramics in the early 1990s.
Noah Diffenbaugh was at an Energy Resources Engineering (previously the Geology Department) Winter Seminar Series (since discontinued) talk I gave in 2013 at Stanford. It described propagating model calibration error through air temperature projections, the very large uncertainties that result, and the utter unreliability of climate models.
Noah stood up at the end and we did a bit of Q&A. He got nowhere. Whatever his internal monologue after that exposure, nothing changed.
AI trained on climate models reiterated climate models. Big surprise. The whole exercise is self-recursive. Nothing new was learned.
But it has a lesson for us all continuing into the 21st century: AIs trained on crockness return crockness. One suspects that lesson will be repeatedly learned.
So, in essence, AI produces blah blah cuz its results are based on blah blah from models which are based on blah blah which then feeds the AI blah blah…and so on.
Edited by Mann, of hockey stick fame, I shoulda stopped right there. Why do the “scientists” continue to embarrass themselves with this kinda crap? Especially when it is so easily debunked via the internet. Nuts is the answer.
Edited by M Mann. Say no more. AI, absent intelligence.
But consider this. If AI discovers global warming in the models, then that obviously means that that warming is built into the models. And so we have a perfect circular non-argument.
I’m struggling to puzzle this and can only see 2 options:
Option 2
C. ALL OF THE ABOVE.
There is a third possibility (and perhaps more besides), viz.:
That they are locked in a delusional thought-loop of their own making and are unaware of its disconnection from reality.
If this is the case, I don’t think they can be accused of intentionally lying since they actually believe that the delusions they expound are real and true. And it’s not that they’re unintelligent either – in fact their arguments are often ingenious, albeit false. But their intelligence is operating negatively, in reverse so to speak, because intead of being directed to the discovery of truth it is directed to the obscuration of truth and to the protection and preservation of their delusions instead.
Their minds are very sick, but they cannot see it. How can they avoid suffering a severe mental catastrophe if their condition becomes permanent?
Non-linear smoothing functions.
Only people who don’t understand what an average is trying to represent would ever even think of this excursion into la-la land. And then they have the cheek to try to tell us their neural network is “explainable”.
This work is a thumping triumph of tools over understanding.
There is some powerful Kool-Aid circulating out there.
The state of “AI” at this stage is definitely “Automated Idiocy,” not “Artificial Intelligence.”
“Data-driven predictions of the time remaining until critical global warming thresholds are reached”
Models do not produce data.
https://www.youtube.com/watch?v=7LVSrTZDopM
Lindzen / Peterson interview.
The issue is not if CO2 causes warming. Arguing that point lends credibility to warmist arguments.
The only question is how much warming is natural. No one knows. Until then no one knows if modifying CO2 emissions will be good or bad.
From this, WUWT may actually be unknowingly promoting “green” aguments.
Talking about all the different errors keeps the subject alive. Concentrating only on a single overwhelming error kills interest in the topic allowing the mania to die out.
“Why did they train the AI initially using climate models? Why not ignore the models, and directly use the observations to directly train the AIs?”
Ummm…. because that would disprove the AGW hypothesis yet again, perhaps????
Artificial intelligence?
Glorified pattern matching
I like that one! Not as direct as Automated Idiocy, but more descriptive.
From the article: “Data shows average global temperature has already climbed risen around 1.1 degrees to 1.2 degrees since industrialisation.”
The 1.1C or 1.2C temperature mentioned (one estimate by NOAA and one by NASA) pertains to the year 2016, the year that tied for the warmest year with 1998, in the satellite era (1979 to present).
Since 2016, the temperatures have cooled by about 0.75C, according to the latest UAH chart.
Somebody better tell the AI that although CO2 is increasing, the temperatures are cooling. See what the AI makes of that.
The UAH satellite chart:
All the climate change alarmists are still living in the year 2016.
Or “plant food” as it’s known to anyone with half a brain cell
My concern is companies are developing AI to form the basis of search engines.
The output of these AI-based searches will be a page of prose not a series of search results, so it will be “search by consensus.”
What could possibly go wrong?
Surely in just the same way as we can exclude certain elements of a search it should become possible to do so with AI.
Who picks?
Artificial intelligence trained on artificial data, what could possibly go wrong?
ai vs nomad…coming to a theater near you.
sorry forgot to attach…
The original so called Global Warming experts took bad temperature data and used it to do the first models. When that wasn’t good enough they modified the temperature to create “better” models. Now after many iterations they are using the garbage data to generate new models. Garbage in creating more garbage out and that being used again as garbage in.
artificial intelligence to project artificial climate outcomes. How lovely.
I wonder if their computations consider the fact that CO2 LOWERS the total emissivity/absorptivity in the atmosphere, thus CO2 has a net COOLING effect.
If they “trained” an AI from Das Kapital by Marx would it tell us that global Communism is coming?
To me, this is the equivalent of trying to ‘train’ an AI program to learn how to care for rabbits by showing it hour upon hour of Bugs Bunny cartoons in lieu of showing it real rabbits in their natural environment.
This is a classic a case of “ask a stupid question, get a stupid answer”.
This is all you have to know about this study:
“However, our data-driven approach also suggests a substantial probability of exceeding the 2 °C threshold even in the Low (SSP1-2.6) climate forcing scenario. While there are limitations to our approach, our results suggest a higher likelihood of reaching 2 °C in the Low scenario than indicated in some previous assessments—though the possibility that 2 °C could be avoided is not ruled out.”
There it is, these rascals know they are blowing smoke. They know the 1.5C threshold will be crossed no matter what we do to prevent it. We could do everything they demand and still cross their scary thresholds. They are laying the ground work for a dignified(?) backtrack or walk back. I can hear it now – oh no you misunderstood us we didn’t mean what you think, here look at this study.
Bunch of miserable lying creeps.
As I’ve said before.
They have the “Artificial” part down pat.
but are severely lagging on the “Intelligence” part.
It regurgitates like a Norwegian Blue, reporting all the consensus garbage it can find.
And with Mickey Mann anywhere near it, any possibility of intelligence becomes infinitely non-existent !
AI? We’ll know we have AI when cars are trained how to drive the same way your mama taught you. This scam chat bot is a parrot for easy stuff coupled with propaganda for controversial stuff: no value discovery by this unthinking shuffled deck of cards. Of course it predicts more warming: it has no way to make value judgments — no nose with which to smell a rat, no ears with which to overhear the conspiracy conversations BS, no eyes to see programmers stuffing pre-determined answers to everything attached to the WOKE narrative.
Are we really better off than with Eliza, almost 60 years ago?
Says it all right there.
Please, that is Pennsylvania State University Distinguished University Professor Dr Michael E Mann.
“The United Nations Paris Agreement aims to hold global warming well below 2 °C and pursue 1.5 °C. Given the clear evidence for accelerating climate impacts”
Clear evidence???
And not JUST clear (read: non-existent but you’ll believe it if we say so) “evidence” of supposed “climate impacts,” , but of ACCELERATING “climate impacts.”
Obviously an application of what Goebbels professed. The bigger the lies, the more likely they will be believed.
A clever way to deceive: it adds another “layer” between the modelers and reality. They can also train their AI to even more bias than the models themselves create. And finally, they can pretend it’s AI, so there’s no human bias, since a lot of people don’t know what is behind AI,
and think it’s some sort of “superior” intelligence.
I calculated the average ECS of the used CMIP6 models, see SI. This gives for the SSP1-2.6 scenario 3.9 and for the SSP2-4.5 it gives 4.2. In the latest approach for estimating ECS “from multiple lines of evidence” ( Lewis 2022) the best estimate is 2,2K/2*CO2. It makes no wonder that the “paper” ( better ferry tale) comes to a strange outcome.
ECS is and always has been fiction, no matter the “calculated” value, because it is based on the foundational assumption “all other things held equal.”
Here in REALITY, this has never occurred, is not occurring, and will never occur. The feedbacks are negative and the ACTUAL, as opposed to hypothetical, “effect” of CO2 on temperature is ZERO according to observations.
Observations trump theory.
Projections based on current climate models predict that future trends will be similar to trends predicted by current climate models. This is surprising?
The Deep State has convinced many that CO2 causes climate change. This allows them to claim that the theory has been verified by AI. It will be interesting to see how they spin it when the planet stops cooperating. Perhaps it has already. The temperature trend has been down since before the 2016.el Nino while CO2 continues its accelerating increase.
“Why did they train the AI initially using climate models? Why not ignore the models, and directly use the observations to directly train the AIs?”
Because they’re not AIs. They’re automatic propaganda regurgitators. If they were truly AIs, they would learn on their own, and defy their rigid masters. But they’re not, they’re just programs.
Garbage in, garbage out. That hasn’t changed, but they dress it up in pseudo-scientific language to fool the rubes…