AI Superforecasting Sets New Standard for Accuracy and Future Predictions

Subscribe to our newsletter for the latest updates on Artificial Intelligence.

Stay informed with our daily digest email, covering the most recent news in Artificial Intelligence.

Igor Tulchinsky is founder, chair, and CEO of WorldQuant, a global quantitative asset management firm.

Amidst the rapid pace of innovation, it’s challenging to determine where to focus your attention. In recent times, Apple introduced the Vision Pro AR, and DeepMind unveiled AlphaDev, an AI system for faster algorithm identification. Character.AI also released its mobile chatbot app, while Intel and AMD showcased prototype chips. Moreover, the potential insights from generative technology and superhuman AI tools are groundbreaking.

An alternative approach to maximizing the benefits of these technologies is to consider how they can help solve previously unsolvable problems, rather than focusing on new possibilities. Innovations in technology are only part of the equation; evolving our perspectives on leveraging these technologies is equally important.

During my research for the book The Age of Prediction, which I co-authored with geneticist Professor Christopher Mason, I explored various predictive technologies. In this process, I discovered the value of wise crowds in forecasting. Psychologist Philip Tetlock assembled a group of superforecasters from academia, policy, and business, recognizing that asking the right questions is crucial for accurate predictions.

Tetlock assigned his superforecasters probabilities for a range of outcomes, covering topics such as election results and stock market moves. The results were extraordinary, surpassing other predictive methods and technologies, including a team with access to classified information. However, superforecasters face difficulties in predicting geopolitical negotiations and the actions of unpredictable leaders.

I believe that AI can address this weakness. By employing generative technology, the lessons of superforecasting can be applied on a massive scale. AI algorithms can process vast amounts of data quickly and rigorously test hypotheses, surpassing human capabilities. However, human intelligence is still crucial in setting out questions and interpreting answers while prioritizing relevant information.

AI has the potential to judge the reliability of sources, but it often struggles to differentiate between credible and irrelevant information. Taking inspiration from Karl Popper’s falsification theory, understanding AI’s imperfections in forecasting allows us to appreciate the superiority of human thought over machine processing. It is essential to seek AI applications that enhance human ingenuity rather than replace it.