We need human learning more than machine learning – Michigan Today

Man vs. machineArtificial intelligence and machine learning are very fashionable right now. We tend to stereotype scientists and engineers as belonging to the more dispassionate and objective elements of society. But in my experience, scientists are often like what the Kinks called “a dedicated follower of fashion.” An idea or a technology emerges. It promises groundbreaking, game-changing potential. It commands attention, focus, and money from federal programs and, in some cases, industry. Once there is money, it garners even more attention.
For the record, I am not an expert in artificial intelligence, more commonly known as AI. My first encounter with AI and machine learning (ML) was at NASA in the 1990s. Much like today (at least at NASA), both were extremely fashionable. Engineers couched the technology in the term “neural networks.”  MIT News provides a good explanation of neural networks and the evolution of the language that comprises artificial intelligence and deep learning.
In most cases the fashionable idea is important; it leads to new techniques and, in some cases, important new understanding. However, it is not magic.
Weather, climate, and AI
My field is abuzz with AI and ML. NOAA has developed an Artificial Intelligence Strategic Plan. There is reason to expect significant contributions that AI will make to weather and climate science; for example, in the representation of physical processes, such as the formation of clouds, and in the statistical calibration of model simulations to tailor that data to actual weather forecasts.
The lead image above may look like a pretty sunset. But it actually is a view of the Marshall Fire in Boulder County, Colo., in December 2021. Author Ricky Rood took this photo from the porch of his home in Colorado. People interested in the topic should read this recent and perspicacious article on machine learning and climate modeling from Royal Society Publishing. This piece provides an excellent history of climate modeling and its relationship to computational capabilities. The author makes an important distinction with this footnote: “AI, or artificial intelligence, is a term we shall generally avoid here in favor of terms like ML, which emphasize the statistical aspect, without implying insight.”
This article argues that ML techniques might become essential because there are proven limits on our ability to include more and more accurate physical, chemical, and biological processes in climate models.
I understand that we get excited about ideas that are new to us. (That’s what drives the fashion industry, after all!) We imagine all the ways the new idea might influence our world. Plus, we like to think about game-changing ideas. We oversell things. As I tell my students, it’s built into the system, much like political manipulation of scientific uncertainty. Expect it and manage it.
Game-changing ideas like AI, though thrilling, can also disrupt society in painful, even generational ways. Deep disruption displaces people from work and transfers wealth. These large economic and societal disruptions make it more difficult to address problems such as climate change. This is outside the scope of this article, so I table it.
My less fashionable, more skeptical, take on AI is based on two principles. The first is the concept of the wicked versus the tamed problem. Computers are able to play chess and other games. And though chess is complex and strategic, it is tamed. It has rules. It is played in a confined place. The computer has a chance in a tamed problem.
There is far less chance of the computer solving a wicked problem. The wicked problem is, perhaps, at best, weakly rule-bound, and its behavior is not strongly bound. We expect behavior that has not been previously experienced.
In short, AI’s potential to “solve” problems is exaggerated by fashion. These computational and statistically based techniques have inherent limitations, and while there are “tamed problems” in weather and climate that artificial intelligence and machine learning will address, I fear we rely too much on technology to manage or fix our growing climate disruptions.
Human intelligence
In my last Climate Blue column I wrote, “The weather is training us to think differently about the climate – human learning, perhaps.” Since that time, I have watched 991 houses burn about three miles from our back porch in Colorado.
The weather is shouting at us: “Come, learn.”
The Dec. 30, 2021, fire in Boulder County started during a windstorm. Such a windstorm is a normal occurrence in this part of Colorado. But the fire started after months of scant precipitation and record-high temperatures.
In short, AI’s potential to “solve” problems is exaggerated by fashion. These computational and statistically based techniques have inherent limitations, and while there are “tamed problems” in weather and climate that artificial intelligence and machine learning will address, I fear we rely too much on technology to manage or fix our growing climate disruptions.There was open space with dry, high grass and scrub. Next to this open space were houses. In some places old, isolated homes, some of them old miner homes – coal mines most likely. In other places, suburban subdevelopment homes on well-kept lots.
The windstorm, a normal weather feature, occurred at a time when it had been hot and dry. The hotness and the dryness provide the background of the event; they represent the climate. Accumulating heat and rising temperatures are part of the warming climate.
The windstorm, the tornado, the winter storm, and the hurricane do not have to be any different than they were 50 years ago for us to experience the influence of climate change. But this formula of a weather event, perhaps an extreme weather event, taking place on a changing background is exactly how climate change will punch us: A firestorm in an unusual season. A tornado in winter. A winter storm or hurricane on the Atlantic Coast at a record-high tide.
Even if we were to stop all emissions from fossil fuels today, we could not reverse the persistent accumulation of heat and rising temperatures in Earth’s climate. We need to understand we are no longer experiencing once-in-a-lifetime events. Things are changing and that change is accelerating.
Building back better is not enough. We need to take on zoning where we build. We need to change engineering standards in building codes. We need to develop resource management and land-use practices on public space and in residential settings to fight fires and manage floods more strategically. We need to think about systems at the neighborhood, community, city, county, and regional scales. We must address education, emergency management, and communication.
Adapting to climate change requires more than AI and ML. This problem demands human learning followed by informed human behavior. I am confident that some humans are learning. Those early adopters need to lead and take on the challenging tasks of organizing to promote collective behavior.  Lead image: A view of the Marshall Fire in Boulder County, Colo., in December 2021. Author Ricky Rood took this photo from his porch in Boulder County. (Image credit: Ricky Rood.)
Source: https://michigantoday.umich.edu/2022/02/11/we-need-human-learning-more-than-machine-learning/