Essay: Bridging Data and Humanity: Enhancing Long-Term Forecasting for a Sustainable Future
Announcing: The ‘Essays on Longtermism’ Competition
When I first opened “Announcing: The ‘Essays on Longtermism’ Competition : Present Action for the Distant Future” and read David Rhys Bernard and Eva Vivalt’s chapter, “What Are the Prospects of Forecasting the Far Future?” I felt a mix of skepticism and curiosity. As someone who has wrestled with data models to predict solar energy adoption in rural Nigeria, I know the frustration of grappling with uncertainty. Bernard and Vivalt argue that forecasting events decades or centuries ahead—whether climate tipping points, technological leaps, or societal shifts—is riddled with challenges: nonlinear dynamics, incomplete data, and the unpredictability of human behavior. Their caution resonates deeply, especially when I think about the stakes of longtermism, which urges us to prioritize the well-being of future generations, potentially numbering in the trillions. If we can’t predict the far future with confidence, how can we act to secure a sustainable world for them? Yet, their analysis left me wondering: could emerging tools like machine learning, combined with community-driven insights, transform our ability to forecast long-term trends and guide ethical interventions? In this essay, I argue that while Bernard and Vivalt rightly highlight the limits of far-future forecasting, integrating advanced data analytics—particularly machine learning—with participatory methods and interdisciplinary perspectives can improve predictive accuracy and align with longtermist goals, especially for climate resilience and equitable energy access.
Bernard and Vivalt’s chapter lays out a sobering case. They argue that long-term forecasting is undermined by several factors: historical data loses relevance over extended time horizons, feedback loops (like carbon cycles) introduce nonlinearity, and unprecedented events (like AI breakthroughs or pandemics) defy prediction. They cite examples like climate modeling, where uncertainties in policy adoption or technological innovation make century-long projections shaky. This hit home for me. In a project I led to expand solar microgrids in northern Nigeria, our team used predictive models to estimate energy demand through 2035. Even over a decade, small errors in assumptions—like government subsidy changes or community uptake—could skew results dramatically. Bernard and Vivalt’s point about the fragility of long-term predictions feels all too real when you’ve seen a model’s confidence intervals widen into near-meaninglessness.
Yet, their skepticism, while well-founded, doesn’t fully explore how emerging technologies can address these challenges. Machine learning, for instance, is uniquely suited to handle complex, nonlinear systems. Unlike traditional econometric models, which often assume linear relationships, algorithms like neural networks or ensemble methods can detect subtle patterns in vast datasets. In my work, I built a recurrent neural network to forecast solar adoption in Kano State, incorporating variables like income levels, weather patterns, and cultural attitudes toward technology. The model outperformed older methods by adapting to nonlinear trends, such as sudden spikes in demand after policy incentives. Bernard and Vivalt acknowledge the potential of computational tools but focus more on their limitations, like overfitting or data scarcity. I believe they undervalue how iterative learning—where models refine predictions as new data emerges—can narrow uncertainty over time. For climate forecasting, machine learning could integrate global datasets (e.g., satellite imagery, economic trends, and emissions records) to model scenarios centuries ahead, even if imperfectly.
Still, Bernard and Vivalt’s caution about overreliance on quantitative methods is a fair warning. Numbers alone can’t capture the full messiness of human systems. In Nigeria, I learned this the hard way when a community rejected a solar project, not because of cost but due to distrust in external organizations—a factor no dataset could quantify. This aligns with their point that qualitative factors, like cultural values or political will, often elude models. To bridge this gap, I propose a hybrid approach: combining machine learning with participatory forecasting. Involving local stakeholders—through workshops, surveys, or futures assemblies—can ground predictions in real-world context. For example, during a tree-planting initiative I helped organize in Kano (long before hearing EFFECTIVE ALTRUISM), we held community forums to understand local environmental priorities. Farmers shared concerns about drought patterns, which we incorporated into our predictive models for reforestation impacts. This human-centered input made our forecasts more robust and ethically aligned, ensuring we addressed immediate community needs while planning for long-term ecological benefits.
This hybrid approach could extend to global longtermist challenges. Take Sustainable Development Goal 7 (Affordable and Clean Energy), a priority for me given Africa’s energy poverty, where over 500 million people lack reliable electricity. Forecasting energy access over centuries requires modeling complex variables: technological innovation, population growth, and climate impacts. Machine learning can handle this complexity by synthesizing data from diverse sources—say, renewable energy costs, grid infrastructure trends, and carbon feedback loops. But as Bernard and Vivalt note, such models risk missing social barriers, like corruption or inequitable distribution. By engaging communities in forecasting—through methods like participatory scenario planning—we can identify these barriers early. In a pilot project I worked on, we used community surveys to adjust our energy models, revealing that local governance issues were a bigger bottleneck than technology costs. This insight shifted our focus to advocacy for transparent energy policies, a strategy that could scale to long-term planning.
Bernard and Vivalt also highlight the risk of “model myopia,” where forecasters overfit to historical data, missing disruptive events like existential risks. This is particularly relevant to longtermism’s focus on catastrophes, such as AI misalignment or ecological collapse. My interest in AI safety leads me to see potential here. Recent advances in reinforcement learning allow models to simulate “black swan” events by exploring edge cases. For instance, researchers at the Center for AI Safety use adversarial testing to predict how AI systems might behave under extreme conditions, like power-seeking scenarios (a topic Joe Carlsmith explores in Chapter 22 of the collection). Applying similar techniques to climate forecasting could help anticipate low-probability, high-impact events, like sudden permafrost melt or geoengineering failures. While Bernard and Vivalt argue that such events are nearly impossible to predict, stress-testing models against outliers could improve our preparedness, aligning with longtermism’s emphasis on safeguarding humanity’s trajectory.
That said, no model—however sophisticated—can fully predict human agency or moral evolution. Bernard and Vivalt note that shifts in values, like the global push for net-zero emissions, depend on collective will, which defies quantification. This limitation suggests a shift in focus: rather than chasing perfect forecasts, we should prioritize robust interventions that remain valuable across scenarios. Longtermism advocates for actions like reducing existential risks or promoting sustainable development, which hold up under uncertainty. In my work, I have seen how flexible solutions—like modular solar grids that adapt to changing needs—can hedge against unpredictable futures. For example, a community energy project I supported in Katsina State in Nigeria used scalable solar units that could adjust to population growth or policy shifts. Such resilience aligns with Toby Ord’s argument in Chapter 13 about shaping humanity’s long-term trajectory through adaptable strategies.
To make this concrete, consider a longtermist approach to energy access. By 2100, Africa’s population could exceed 4 billion, amplifying energy demands. Machine learning can forecast these needs by modeling urbanization, climate impacts, and renewable technology costs. But without community input, these models risk irrelevance. In a recent project, I collaborated with local leaders to predict energy needs in a rural district. Their insights on cultural preferences—like favoring solar lanterns over grid connections—refined our models and ensured community buy-in. Scaling this approach globally could inform policies that balance immediate needs with long-term sustainability, addressing Bernard and Vivalt’s concern about forecasting’s practical limits.
However, I must acknowledge a counterpoint: overinvesting in forecasting could divert resources from urgent action. As Amanda Askell and Sven Neth argue in Chapter 17, “Longtermist Myopia,” an obsession with future scenarios can blind us to present-day suffering. In Nigeria, I have seen communities struggling with immediate energy poverty; waiting for perfect predictions could delay critical interventions. My response is that forecasting and action doesn’t need to be mutually exclusive. By using iterative models and community feedback, we can act now while refining long-term strategies. This balance reflects Hilary Greaves and Christian Tarsney’s distinction in Chapter 18 between minimal and expansive longtermism, where we prioritize robust, near-term actions with long-term benefits.
In conclusion, Bernard and Vivalt’s chapter offers a vital caution about the limits of far-future forecasting, but it underestimates the potential of integrating machine learning with participatory methods. My experiences in sustainability and data analytics convince me that this hybrid approach—blending computational power with human insights—can enhance predictions and guide ethical interventions. For challenges like climate change and energy access, such forecasts can inform resilient strategies that align with longtermism’s goal of securing a thriving future.
I hope this essay of mine will sparks discussion on how we can combine technology and humanity to act wisely for generations to come.