Re-evaluating Prof. Philip Tetlock's forecasting work on existential risk

At Founders Pledge, we’re always on the lookout for new opportunities to do immense good in the world. We also continuously re-evaluate our existing recommendations to make sure they remain high-impact funding opportunities. In this post, I briefly explain how we approach re-evaluating existing charities with a recent example: Professor Philip Tetlock’s second-generation forecasting work, research that we believe could be highly valuable in anticipating and mitigating some of the most pressing global risks.

We had last evaluated Professor Tetlock’s work in November 2020 when, after carefully evaluating all the evidence, we concluded that the proposed work could help with the prediction of early-warning indicators of some of the most important issues facing humanity in the future. We also concluded that Professor Tetlock had the track record and expertise to carry out this research and face its challenges. However, we were careful to note that while the research had the potential to have a tremendously positive impact on humanity, it was also high risk: there was a high likelihood that the project would fail to find any actionable tools for forecasting global catastrophic risks.

Much has changed since 2020. The researchers delivered a research product, they received additional funding from other sources, and the US government expressed a renewed interest in probabilistic forecasting. In short, we felt it was high time to re-evaluate this funding opportunity, to assess whether these changes might affect our rating of it.

We started by diving deep into the most recent literature on forecasting research. Most importantly, Professor Tetlock and his collaborators Dr. Ezra Karger and Dr. Pavel Atanasov had published their latest research (funded by Founders Pledge) in a paper called “Improving Judgments of Existential Risk: Better Forecasts, Questions, Explanations, Policies." New work in crowdsourced forecasting had re-affirmed the value of collective intelligence in modeling disease outbreak scenarios during the Covid-19 pandemic. Others wrote pieces critical of the “superforecasting” research agenda, and Tetlock and his collaborators Yunzi Lu and Barbara Mellers responded to the criticism. And before joining Founders Pledge late last year, I had co-authored a report and articles about the implementation of forecasting, and helped to organize a conference on the state of forecasting in global policy.

After familiarizing ourselves with the newest research, we moved to the next step in re-evaluating our funding opportunities, and asked ourselves three questions:

1. What has changed with this funding opportunity?
2. What relevant factors have changed in the world?
3. How has our knowledge about this issue changed?

To help answer the first question, our Philanthropic Services team routinely asks the organizations we fund to provide updates on their situation. We keep this as brief as possible, to minimize the time that organizations need to spend on answering our questions (and maximize the time they spend on effective interventions). Next, we asked Professor Tetlock directly for any more recent updates. We learned that they had been successful in securing additional grants from other funders. In further discussion with Professor Tetlock, we came to understand that they could productively absorb even more funding, and scale up their research in effective ways. Learning about this room for funding provided a small update that nudged us to believe that Professor Tetlock’s work could remain a high-impact funding opportunity.

The second answer to “what has changed with this funding opportunity?” was that Professor Tetlock and his team successfully completed the paper on existential risk forecasting that Founders Pledge helped to fund: “Improving Judgments of Existential Risk". The paper is highly technical in parts, but essentially achieves two things. First, it outlines ten challenges that researchers need to overcome in order to be able to make better forecasts about global catastrophic risks (for example, how do we balance the rigor of narrow precise forecasts with the relevance of vague big-picture questions?). Second, it outlines potential solutions to these challenges, by using “second-generation forecasting tournaments,” which combine the proven techniques of earlier experiments with new research on tools like Reciprocal Scoring and Bayesian question clusters. The paper opens up a large field of new challenges and opportunities, including a method for measuring the effect size of policy interventions with forecasting — how much might a specific policy option mitigate risk — which we believe could be highly valuable to mitigating large-scale risks.

What has changed in the world? In addition to the new publications and advancements in forecasting research, the biggest change relates to government interest in probabilistic forecasting. First, the UK government launched a new forecasting platform known as COSMIC BAZAAR, based on the principles of Professor Tetlock’s earlier research. Second, the US government expressed a renewed interest in the same techniques, when Deputy Director of National Intelligence for Mission Integration, Morgan Muir, announced in September 2021 that they, too, were heeding the call to incorporate probabilistic forecasting and working on a new crowdsourced forecasting platform.

What should we make of these changes? On the one hand, government interest in forecasting has waxed and waned periodically over the years, so perhaps we should not make much of it at all. On the other hand, this very recent interest suggests that basic research in forecasting wouldn’t just languish in academia. There is now a clear path from academic forecasting research to policy application; investments in the former might be multiplied many times in the latter. In short, a window of opportunity may have opened.

Still, we wanted to avoid engaging in motivated reasoning, and asked ourselves: could this development mean that forecasting research was less neglected now? After parsing the public statements, we came to the conclusion that these new government platforms were focused on “first generation” forecasting, not the “second generation” work that Professor Tetlock is pursuing. This conclusion also led us to believe that this research continues to be a good funding opportunity.

Finally, we asked ourselves what had changed in our knowledge and views about this issue. We continue to believe that forecasting and mitigating global catastrophic risks could have an extremely large positive impact on the world. However, we no longer believe that this impact necessarily relies on the ability to make long-term forecasts. Rather, there are ways that forecasting can improve decision-making about existential risk even in the near-term. For example, in “Improving Judgments for Existential Risk,” the authors outline a method for using superforecasters — highly accurate and proven predictors — to assess the hypothetical impacts of certain policies on risk mitigation. Using this technique, for example, a decision-maker in the middle of a pandemic could have a clearer sense of the expected value of their options, and make a more informed decision to reduce the human suffering wrought by the disease. None of this requires long-term foresight beyond a few months. The value of this research is therefore much more than just experimenting with long-term forecasting.

Moreover, we now believe that the value of this work reaches far beyond policy and government decision-making. For example, if these techniques prove to be workable in Professor Tetlock’s tournaments, data-driven philanthropic organizations like Founders Pledge could use them to estimate the effect of their grants on hard-to measure issues like reducing long-term risk by funding various organizations. Such an intervention could multiply the effectiveness of the philanthropic sector as a whole.

In the end, we weighed all the available evidence about the research team, their recent successes, the development of the literature, changes in the world, and changes in our own knowledge, and concluded that Professor Tetlock and his team continue to be a high-impact funding opportunity for our members, on a hits-based (high-risk, high-reward) model of giving. We will continue to re-evaluate this and all our recommended funding opportunities to ensure we give the most up-to-date advice on how to do the most good in the world.