How can AI be used to more reliably analyze and plan around the tie between climate and emissions?

Question

Note to the Duplicate Police

This question is not a duplicate of the Q&A thread referenced in the close request. The only text even remotely related in that other thread is the brief mention of climate change in the Q and two sentences in the sole answer: "Identify deforestation and the rate at which it's happening using computer vision and help in fighting back based on how critical the rate is. The World Resources Institute had entered into a partnership with Orbital Insight on this."

If you look at the four bullet items below, you will find that this question asks a very specific thing about the relationship between climate and emissions. Neither that question nor that answer overlaps with the content of this question in any meaningful way. For instance, it is well known that CO² is NOT causing deforestation. The additional carbon dioxide in the atmosphere causes faster regrowth. This is because plants need CO² to grow. Hydroponic containers deliberately boost it to improve growth rates. Plants manufacture their own oxygen from the CO² via chlorophyll.

If you recall from fifth grade biology, that's why they are plants.

---

Now Back to the Question

Several climate models have been proposed and used to model the relationship between human carbon emissions, added to the natural carbon emissions of life forms on earth, and features of climate that could damage the biosphere.

Population growth and industrialization have many impacts on the biosphere, including loss of terrain and pollution. Negative oceanic effects, including unpredictable changes in plankton and cyanobacteria are under study. Carbon emissions from combustion has received attention in recent decades just as sulfur emissions were central to concerns a century or more ago.

Predicting weather and climate is certainly difficult because it is complex and chaotic, as typical inaccuracies in forecasts clearly demonstrate, but that is looking forward. Looking backward, analyses of data already collected have shown a high probability that ocean and surface temperature rises followed increases in industrial and transportation related combustion of fuels.

How might AI be used to produce some of the key models humans need to protect the biosphere from severe damage.

• A more reliable analysis of what has already occurred, since there is some legitimacy to the differing views as to how gross the effect of carbon emissions has been on extinctions of species in the biosphere and on arctic and antarctic melting

• A better understanding as to whether the climate of the biosphere behaves as a buffer of climate, always tending to re-balance after a volcanic eruption, meteor stroke, or other event, or whether the runaway scenario described by some climatologist, where there is a point of no return, is realistic

• A better model to use in trying out scenarios so that solutions can be applied in the order that makes sense from both environmental and economic perspectives

• Automation of climate planning so that the harmful effects of the irresponsibility of one geopolitical entity wishing to industrialize without constraint on other geopolitical entities can be mitigated

Can pattern recognition, feature extraction, the learned functionality of deep networks, or generative techniques be used to accomplish these things? Can rules of climate be learned? Are there discrete or graph based tools that should be used?

Answer 1

Can AI provide a more reliable analysis of the gross effects of carbon emissions on extinctions of species ice-cap melting, and other effects?

Yes. The work of Judea Pearl and others over the last 20 years began out of a desire to address uncertainty within AI. Eventually, this led Pearl to become fascinated by the need to quantifiably determine when one event has caused another, the problem at the root of "correlation is not causation". He substantively succeeded with the combination of causal modeling and the do-calculus. The do-calculus allows you to formulate queries of the form "To what degree did X cause Y?", and to automatically determine what measurements are needed to determine the answer in a statistically meaningful way. The algorithms used under the hood are closely related to those used in AI and robotics systems to reason about uncertainty (i.e. Bayesian Inference). Causal modeling is still relatively new, and is not yet as widely used as it could be. An open problem is how to specify what the model of the world looks like from data (rather than just reasoning over a model that is given). If this problem is solved, we could see major improvements in the ability to provide analyses like those you ask about.

Can AI provide a better understanding about whether the runaway scenario described by some climatologists is realistic?

Probably not. Predicting future events without any observable precedents is not something that anyone can be sure about. The questions raised in those kinds of scenarios aren't about calculation, but about modeling. For example, if you believe that there is a large amount of methane trapped in the arctic permafrost, and you believe that temperatures above a certain range release this much faster than in the past, and you believe that methane warms the climate rapidly, then you would tend to believe that the runaway scenarios are plausible. AI can't tell us whether the methane is there (we have to go measure it, though maybe AI can make the measurement more accurate), and can't tell us how temperature will affect the release rate (again, we have models of this, but they rely on different assumptions, and we have to go measure to find out which are right).

Can AI provide better simulations of the impacts of interventions on both climate and the economy, to inform decision making?

Maybe. Agent-based modeling can help with this to some degree, and is arguably part of AI. In fact, it already has been to some degree, by Beeger & Troost in J. Agriculture Economics, in 2014 and again in 2017, though interest in this kind of modeling looks like a pretty new development in this area. Although ABM can give us reasonable models and help simulate the impact of interventions, ultimately they are just one modeling tool among many. Their potency may improve if more realistic agent models are used, but it is not clear that AI is going to provide advances in this area in the near future.

Automation of climate planning so that the harmful effects of the irresponsibility of one geopolitical entity wishing to industrialize without constraint on other geopolitical entities can be mitigated

Probably not. Although AI techniques have made some kinds of economic planning problems a lot easier, the main barrier to the effects you describe is a social/political one: countries are sovereign, and the world operates as a de facto anarchy (i.e. the UN is impotent). Your AI model can tell, say, India not to industrialize, but Indians want to enjoy the same kind of lifestyle improvements that Americans do, and would rather enjoy them sooner than later. India would collectively rather that Americans put an enormous tax on their carbon emissions, drive less, eat much, much less beef, and stop flying everywhere, than that Indians continue living on the equivalent of $7,000 each per year. In contrast, Americans would rather that Indians just wait a few decades while the developed world decarbonizes, and only industrialize once we have enough solar panels to replace all our current needs within current industrial economies.

Basically this is a resource allocation problem within an anarchy: we can only burn X carbon within Y time, everyone wants to burn some, and the only way to enforce contracts is with the threat of massive violence (or massive economic sanctions, which in turn, are imposed through the threat of violence against other actors if they trade with the sanctioned country). AI can help us answer questions like "How much should the USA pay India in exchange for India not industrializing this year?", see, e.g. work in Auction Theory, but AI can't actually make nations do those things if they can't reach a diplomatic compromise.