Guest essay by Eric Worrall
According to the big computer we are doomed to suffer ever more damaging weather extremes. But researchers can’t tell us exactly why, because their black box neural net won’t explain its prediction.
Human activity influencing global rainfall, study finds
Anthropogenic warming of climate has been a factor in extreme precipitation events globally, researchers say
Wed 7 Jul 2021 15.00 AEST
While there are regional differences, and some places are becoming drier, Met Office data shows that overall, intense rainfall is increasing globally, meaning the rainiest days of the year are getting wetter. Changes to rainfall extremes – the number of very heavy rainfall days – are also a problem. These short, intense periods of rainfall can lead to flash flooding, with devastating impacts on infrastructure and the environment.
“We are already observing a 1.2C warming compared to pre-industrial levels,” pointed out Dr Sihan Li, a senior research associate at the University of Oxford, who was not involved in the study. She said: “If warming continues to increase, we will get more intense episodes of extreme precipitation, but also extreme drought events as well.”
Li said that while the machine-learning method used in the study was cutting edge, it currently did not allow for the attribution of individual factors that can influence precipitation extremes, such as anthropogenic aerosols, land-use change, or volcanic eruptions.
The method of machine learning used in the study learned from data alone. Madakumbura pointed out that in the future, “we can aid this learning by imposing climate physics in the algorithm, so it will not only learn whether the extreme precipitation has changed, but also the mechanisms, why it has changed”. “That’s the next step,” he said.
…Read more: https://www.theguardian.com/environment/2021/jul/07/human-activity-influencing-global-rainfall-study-finds
The abstract of the study;
Anthropogenic influence on extreme precipitation over global land areas seen in multiple observational datasets
The intensification of extreme precipitation under anthropogenic forcing is robustly projected by global climate models, but highly challenging to detect in the observational record. Large internal variability distorts this anthropogenic signal. Models produce diverse magnitudes of precipitation response to anthropogenic forcing, largely due to differing schemes for parameterizing subgrid-scale processes. Meanwhile, multiple global observational datasets of daily precipitation exist, developed using varying techniques and inhomogeneously sampled data in space and time. Previous attempts to detect human influence on extreme precipitation have not incorporated model uncertainty, and have been limited to specific regions and observational datasets. Using machine learning methods that can account for these uncertainties and capable of identifying the time evolution of the spatial patterns, we find a physically interpretable anthropogenic signal that is detectable in all global observational datasets. Machine learning efficiently generates multiple lines of evidence supporting detection of an anthropogenic signal in global extreme precipitation.Read more: https://www.nature.com/articles/s41467-021-24262-x
As an IT expert who has built commercial AI systems, I find it incredible that the researchers seem so naive as to think their AI machine output has value, without corroborating evidence. They admit they are going to try to understand how their AI works – but in my opinion they have jumped the gun, making big claims on the basis of a black box result.
Consider the following;
Amazon ditched AI recruiting tool that favored men for technical jobs
Specialists had been building computer programs since 2014 to review résumés in an effort to automate the search process
Amazon’s machine-learning specialists uncovered a big problem: their new recruiting engine did not like women.
But by 2015, the company realized its new system was not rating candidates for software developer jobs and other technical posts in a gender-neutral way.
That is because Amazon’s computer models were trained to vet applicants by observing patterns in résumés submitted to the company over a 10-year period. Most came from men, a reflection of male dominance across the tech industry.
In effect, Amazon’s system taught itself that male candidates were preferable. It penalized résumés that included the word “women’s”, as in “women’s chess club captain”. And it downgraded graduates of two all-women’s colleges, according to people familiar with the matter.
Amazon edited the programs to make them neutral to these particular terms. But that was no guarantee that the machines would not devise other ways of sorting candidates that could prove discriminatory, the people said.
…Read more: https://www.theguardian.com/technology/2018/oct/10/amazon-hiring-ai-gender-bias-recruiting-engine
In hindsight it is obvious what happened. The Amazon AI was told to try to select the most suitable candidates, and it noticed more male candidates were being accepted for technical jobs, likely because there were more male candidates applying. So it concluded men are more suitable for technical jobs.
It is important to note this male bias in technical jobs is purely a Western cultural issue. When I visited a software development shop in Taipei, there were just as many women as men developing software. The women I have met, in Western IT shops and in that IT shop in Taipei, were just as smart and technically capable as any man. Somehow we are persuading our women not to pursue technical careers.
My point is, when scientists unleash a black box AI on a set of data, they have no way of knowing whether the output of that AI is what they think it is, until they painstakingly rip the AI apart to work out exactly how it formed its conclusions.
The climate scientists think they have discovered a significant camouflaged anthropogenic influence. Or they may have discovered a large hidden bias in their data or models. To be fair they admit there might be problems with their training data, and the climate models they use to hindcast what conditions would have been without anthropogenic influence. “… In addition, the training GCMs might be undersampling the low-frequency natural variability such as Atlantic Multidecadal variability and Pacific Decadal Oscillation. …“. This admission should have been their headline.
Until they break their black box system down, work out exactly how their AI is reaching its conclusion, and present the real method for review, the method which is currently hidden inside their AI, it seems remarkably premature to go for a big announcement, just because they like the look of their result.