Hide table of contents

Algorithmic forecasting is the process of making predictions about future events using computational algorithms. This term is analogous to "Algorithmic Trading" in the context of trading, where automated systems make decisions based on predefined rules and data inputs. Algorithmic forecasting can be implemented through various approaches, such as algorithms placing bets on binary prediction markets or submitting predictions directly to forecasting platforms.

There is no universally agreed-upon formal definition of algorithmic forecasting. In broader contexts outside of effective altruism, it generally refers to models that leverage large datasets and statistical methods to predict future trends or events. These models often incorporate techniques from machine learning to improve their accuracy over time by learning from new data. Within effective altruism, the scope of algorithmic forecasting can also include algorithms based on subjective beliefs rather than solely on empirical data.

Further Reading:

AI-Powered Forecasting

One specific approach to algorithmic forecasting is to have AI systems directly generate forecasts. There is a growing body of work using AI systems to generate forecasts on various questions, similar to how other judgmental forecasting platforms operate.

"AI-Powered Forecasting" is also not an agreed-upon term, though there is a mutual understanding of the concept. It should not be confused with "AI Forecasting," which refers to forecasting about AI, not by AI.

Notable Companies:

  • FutureSearch, a startup that generates domain-general forecasts using AI

Notable Studies:

  • A study by Danny Halawi et al. explores the potential of language models to approach human-level forecasting, demonstrating that these models can provide accurate predictions at scale and inform decision-making (Halawi et al., 2022).
  • Research by Philipp Schoenegger et al. shows that AI-augmented predictions with GPT-4-Turbo assistants can enhance human forecasting accuracy by 23%, indicating significant benefits even from biased AI assistance (Schoenegger et al., 2023).