Leveraging Technology and Managing Risks in Decision Making

Key Insight

Effective decision making can be enhanced by treating every choice as an expected value calculation, weighing the probability and reward of being right against the probability and penalty of being wrong. A decision has a positive expected value if the reward multiplied by its probability of occurrence exceeds the penalty multiplied by its probability. This framework reveals that betting on what is most probable is not always optimal; sometimes a low-probability, high-reward bet with negligible downside (e.g., asking about a house not on the market) can be smart. Critically, increasing the probability of being right, even slightly, yields significant value, for example, raising odds from 51 percent to 85 percent is seventeen times more impactful than moving from 49 percent to 51 percent. Knowing when not to bet and favoring choices with more pros than cons, rather than those with no cons at all, further improves outcomes.

Simplifying complex situations and utilizing principles are vital shortcuts for becoming a great decision maker, reducing the number of decisions by a factor of 100000 by categorizing 'cases at hand' as 'another one of those' and applying pre-defined principles. Believability-weighted decision making involves triangulating with highly credible individuals and resolving disagreements based on principles and individual believability. Furthermore, converting these principles into algorithms allows for testing, refinement, compounding understanding, and removing emotional bias from the decision process, essentially enabling computers to make decisions alongside humans, which has been practiced for over thirty years at Bridgewater.

The partnership between human and artificial intelligence offers immense power. While computers excel in processing vast information faster, more reliably, and objectively, and are immune to biases and panic, human intelligence remains superior in imagination, common sense, and defining personal values. Trusting AI without deep understanding carries significant risks; it is essential to distinguish between 'expert systems' (based on specified cause-effect logic), 'mimicking' (pattern observation without logical understanding), and 'data mining' (finding patterns in massive data). Mimicking and data mining can fail when the future diverges from the past or when correlations are mistaken for causality, as computers lack common sense. The most powerful decision makers combine human intuition and values with the analytical strengths of technology, using systems as supplements, not substitutes, for their own judgment.