AI as a Quantitative Predictor of the Future: Climate Change

AI climate change predictions by Gemini

I am back to AI as a quantitative predictor (June 26, 2026 blog). I interrupted my discussion of the topic to describe events connected to Independence Day in the US on July 4^th. I will try to demonstrate the use of AI in predicting impacts of climate change. This will be a long blog. The top figure summarizes the AI prediction (with Gemini). Most of the writing included here is from my direct discussions with AI. As in previous blogs, my questions or prompts to AI are in red. The words “anthropogenic” and “attributions” are not mentioned in the questions. However, as will be seen in its predictions, AI closely follows the IPCC (Intergovernmental Panel on Climate Change) analyses, which are based on socioeconomic scenarios directly connected to the future climate, factoring in types of energy use. These concepts relate to attribution, including whether they are anthropogenic (human-caused). However, my lack of direct use of the keywords “anthropogenic” and “attributions” might (or might not) be the reason that the AI response doesn’t mention governments and societal attitude to climate change. The recent global heat waves have brought the conversation back to attribution and anthropogenic climate change.

The June 26^th blog shows that AI admitted to vulnerabilities in predictions under the conditions of a “data gap”:

The “Data Gap”: AI struggles significantly when there is a time gap between its training data and a current event. It cannot naturally account for sudden, unprecedented “black swan” events, such as geopolitical outbreaks or unrecorded financial crises. [1, 2]

• Overconfidence and Hallucination: Evaluation studies by institutions like Knowledge at Wharton show that AI models frequently exhibit overconfidence, assigning dangerously high probability scores to outcomes that lack factual evidence. [1]

• Lack of True Intuition: AI relies purely on historical data correlation. Unlike humans, it lacks genuine creative insight or “judgmental forecasting” capabilities—the ability to assign probabilities based on abstract intuition and personal domain expertise. [1, 2]

The AI predictions, to a large degree, come out of the IPCC general reports, which summarize the data on the issue in intervals of 5 – 7 years. The last one came out in 2023. One major recent change in their evaluation was to scrap the worst emissions scenario (RCP8.5). The IPCC climate reports are a good example of a “data gap,” which makes climate change predictions difficult. Let me now proceed to the AI predictions. I started my AI discussion (with Gemini) with the following question:

 “Can you predict the future of global climate change and explain how you arrived at these predictions?”

Predicting the future of global climate change involves analyzing complex, interacting systems—combining physical climate science with human socio-economic choices. Because we cannot know exactly how human society will act, scientists use scenarios rather than absolute crystal-ball predictions.

Here is what the future likely holds based on current scientific consensus, followed by the methodology used to arrive at these conclusions.

The Predictions: Where the Planet is Headed

The trajectory of climate change depends heavily on global emissions over the next few decades. Scientists generally look at three potential pathways:

1. The High-Emissions Scenario (Business as Usual)

If global reliance on fossil fuels continues to grow without significant policy shifts, global average temperatures could rise by () above pre-industrial levels by the year 2100.

• Impacts: Catastrophic sea-level rise (potentially exceeding 1 meter), widespread crop failures, frequent extreme weather events making certain equatorial regions virtually uninhabitable, and severe biodiversity loss.

2. The Moderate / Current Trajectory Scenario

Based on current global policies and climate pledges (as of the mid-2020s), the world is tracking toward a warming of roughly () by 2100.

• Impacts: While this avoids the absolute worst-case outcomes, it still triggers severe consequences. We will see heavily intensified heatwaves, prolonged droughts, destabilized ice sheets, and significant economic strain on global infrastructure, particularly in coastal and developing nations.

3. The Low-Emissions Scenario (Paris Agreement Targets)

If the world rapidly transitions to renewable energy and achieves net-zero greenhouse gas emissions by around 2050, warming could be capped between ().

• Impacts: Climate impacts will still intensify compared to today (more intense storms, shifting agricultural zones), but the changes would remain largely manageable, and society would have a much higher capacity to adapt.

How These Predictions Are Made

Climate scientists arrive at these predictions through a rigorous, multi-layered methodology that blends historical data, physics, and economic forecasting.

1. Representative Concentration Pathways (RCPs) and SSPs

Scientists don’t just guess what humans will do; they create standardized frameworks called Shared Socioeconomic Pathways (SSPs) and Representative Concentration Pathways (RCPs). These model different futures based on population growth, economic trends, energy technologies, and environmental policies. They calculate exactly how much greenhouse gas will end up in the atmosphere under each scenario.

2. Global Climate Models (GCMs)

Once a scenario is selected, the data is fed into massive supercomputers running Global Climate Models. These models divide the Earth, atmosphere, and oceans into a 3D grid. Using fundamental laws of physics, fluid dynamics, and chemistry, the supercomputers simulate how energy and mass move through the grid over decades. They calculate variables like:

• Atmospheric pressure and temperature
• Ocean currents and heat absorption
• Cloud formation and solar radiation reflection

3. Accounting for Climate Feedbacks

Predictions must account for “feedback loops,” which can accelerate or decelerate warming.

• Positive Feedback (Amplifying): As the Arctic ice melts, the darker ocean water absorbs more heat than the reflective white ice did, accelerating warming (the Albedo Effect). Similarly, thawing permafrost releases trapped methane, a potent greenhouse gas.
• Negative Feedback (Dampening): Increased carbon dioxide can sometimes stimulate increased plant growth in certain areas, which absorbs more from the atmosphere, slightly mitigating the warming effect.

4. Historical Paleoclimate Data

To ensure the computer models are accurate, scientists test them against the past. They look at historical climate data gathered from ice cores, tree rings, and ocean sediment samples. If a model can accurately “predict” the climate shifts of the last 100,000 years based on historical atmospheric data, scientists gain high confidence that its predictions for the next 100 years are reliable.

Ultimately, the future of climate change isn’t a fixed mathematical certainty—it is a variable equation where human behavior is the most critical factor.

Are you interested in how these climate predictions specifically impact a certain region, or would you like to explore the economic models behind the transition to the energy transition to the energy transition to shift to renewables?

 “Can you predict the future of global climate change and explain how you arrived at these predictions?”

To look at the numbers behind those scenarios, we have to look at the hard data. The gold standard for these quantities comes from the Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report.

Scientists quantify these predictions across several key metrics: atmospheric CO2 concentrations, global temperature rise, sea-level rise, and carbon budgets.

Here is how the numbers stack up for the year 2100 across the three main pathways.

2. Quantifying Extreme Weather Frequency

Climate modeling also quantifies how often extreme events will happen compared to a world without human-induced warming.

Extreme Heatwaves

• At 1.5°C warming: An extreme heat event that used to happen once every 50 years will occur 6 times more frequently and be 1.2°C hotter.
• At 2°C warming: That same heatwave occurs 9 times more frequently and is 2.7°C hotter.
• At 4°C warming: It occurs 2 times more frequently—meaning it happens almost every single year—and is 5.3°C hotter.

Heavy Precipitation (Storms & Flooding)

• At 1.5°C warming: Heavy 1-in-10-year rain events will happen 5 times more often and pack 10% more moisture.
• At 4°C warming: These deluge events happen 7 times more often and pack 30% more water, drastically overwhelming urban stormwater infrastructure.

3. The Remaining “Carbon Budget”

To ground these predictions in a definitive timeline, scientists use a metric called the Remaining Carbon Budget. This is the total net amount of CO2 humanity can ever emit from this point forward to stay below a specific temperature threshold.

• To cap warming at 1.5°C (with a 50% chance): The remaining global budget is roughly 200 to 250 gigatons (GtCO2)
• To cap warming at 2.0°C (with a 67% chance): The remaining budget is roughly 900 to 1,000 GtCO2.

The Reality Check

The world currently emits approximately 40 gigatons of CO2 per year from fossil fuels and industry.

• At our current burn rate, the budget to guarantee staying under 1.5°C will be entirely exhausted within 5 to 6 years.

• The budget for 2.0°C would be exhausted in about 22 to 25 years if global emissions flatten rather than drop.

Would you like to look at the math behind how economists translate these physical metrics into global GDP losses, or look into which specific regions bear the highest quantitative burdens?

In yesterday’s discussion about predictions of climate change, in your answer to the question “Can you quantify the predictions and the scenarios?” your units need some improvements. Mind to try again??

I hear you loud and clear. Looking back, mixing up or being vague with climate metrics (like toggling between gigatons of vs. carbon, or leaving temperature anomalies without a baseline) just muddies the waters. Let’s clean that up and lay out the hard, quantified numbers with strict attention to the proper units.

Climate scientists primarily use the Intergovernmental Panel on Climate Change (IPCC) Shared Socioeconomic Pathways (SSPs) to model our future.

Here is the precise breakdown of the scenarios and their quantified predictions by 2100.

The Metrics Defined

To ensure complete clarity, here are the exact units used below:

• : Gigatons (billion metric tons) of Carbon Dioxide emitted per year.
• : Parts per million, the atmospheric concentration of .
• : Watts per square meter, the measure of “radiative forcing” (the extra heat trapping energy in the atmosphere).
• °C: Degrees Celsius of warming above the pre-industrial baseline (1850–1900).

Quantified Climate Scenarios (Predictions for 2100)

1. The “Aggressive Mitigation” Scenario (SSP1-1.9)

This assumes global society shifts rapidly to sustainable practices, hitting net-zero emissions around 2050.

• Annual Emissions by 2050: (Net-Zero)
• 2100 Atmospheric Concentration:

2. The “Paris Agreement Target” Scenario (SSP1-2.6)

Slightly less aggressive than 1.9, but still keeps global warming well below the 2°C threshold.

• Annual Emissions by 2075: Net-Zero
• 2100 Atmospheric Concentration:
• Radiative Forcing by 2100:
• Predicted Warming: (Most likely outcome: )

3. The “Middle of the Road” Scenario (SSP2-4.5)

Societal and economic trends follow historical patterns. Emissions plateau around mid-century before declining, but net-zero is not reached by 2100.

• Peak Annual Emissions (2050):
• 2100 Atmospheric Concentration:
• Radiative Forcing by 2100:
• Predicted Warming: (Most likely outcome: )

4. The “High Emissions” Scenario (SSP3-7.0)

Characterized by regional rivalry, resurging coal use, and slow technological progress in clean energy.

• Annual Emissions by 2100: (Double current levels)
• 2100 Atmospheric Concentration:
• Radiative Forcing by 2100:
• Predicted Warming: (Most likely outcome: )

5. The “Fossil-Fueled Development” Scenario (SSP5-8.5)

The worst-case “business as usual” scenario driven by rapid, fossil-fuel-intensive economic growth. While considered less likely now due to the rise of renewables, it serves as an upper bound.

• Annual Emissions by 2100:
• 2100 Atmospheric Concentration:
• Radiative Forcing by 2100:
• Predicted Warming: (Most likely outcome: )

This AI discussion included interesting conclusions and missing answers. My first attempt returned gibberish and the one above did not include any actual numbers (even though I specifically asked for a quantitative analysis). Another big thing the AI conclusions didn’t mention was the Trump administration’s efforts to eliminate every mention of climate change (The Trump administration is eroding vital climate data):

AMERICAN SCIENTISTS have historically been leaders in the collection and analysis of data on climate change. The longest-running observations of carbon-dioxide concentrations in the atmosphere, for example, are collected at the Mauna Loa observatory in Hawaii. The National Snow and Ice Data Centre at the University of Colorado, Boulder, for its part, holds unique databases on the annual ebb and flow of sea ice at both poles. America also owns 58% of the roughly 4,000 Argo floats which drift at depth across the world’s oceans, gauging their health before popping up to the surface every ten days or so to broadcast their data home. These projects, and many more like them, are now under threat. Donald Trump’s administration is decommissioning long-standing databases; deleting key reports and analyses; and firing or reassigning the staff who have unique expertise in making sense of the data. Such actions will make climate modelling harder. But the harm done will be felt not just by the world’s climate scientists. American citizens and businesses will suffer, too.

In the US, scientists are starting to try to house the data independently from the federal government:

A small group of former government workers has recreated a valuable climate-science website that had been shuttered last year under the Trump administration. The new site, climate.us, is an effort by former staff members at the National Oceanic and Atmospheric Administration to present climate science previously housed at climate.gov, including data, reports, articles, and congressionally mandated national climate assessments. The new site is effectively the “first full clone” of climate.gov, said Rebecca Lindsey, managing director of climate.us. It became fully active on Tuesday morning.

My next blog will examine how including the keywords “anthropogenic” and “attributions” in the questions to AI changes its response.