#115 - Under Uncertainty, Humans Win

I find this comparison quite astonishing. We are creating machines that require huge amounts of power just to imitate something the human brain does using very little energy.

The human brain operates on about 20% of the body’s total energy budget, which amounts to roughly 90–100 watts throughout the day. This efficiency is unique and not even remotely matched by any AI technology available today.

Gerd Gigerenzer, a long-time critic of Daniel Kahneman and his theory of cognitive biases, argues that these remarkable capabilities make the human brain far superior to artificial intelligence. Unlike Kahneman, who emphasizes systematic errors in human judgment, Gigerenzer views heuristics as powerful adaptations to uncertainty and information overload. In his view, these heuristics are a key advantage humans hold over AI.

This disagreement is not academic. It cuts to the core of how we think about intelligence itself.

Kahneman frames decision-making in terms of risk—situations where all possible outcomes and their probabilities are known, such as in roulette. Gigerenzer, by contrast, distinguishes between risk and true uncertainty. In uncertain situations—such as hiring an employee or forecasting an election— we simply do not know all the possible outcomes, let alone their probabilities.

AI systems tend to outperform humans in stable environments, where the future closely resembles the past. This is why AI excels at games like chess. However, it is far less capable in domains characterized by high uncertainty. Gigerenzer points to examples such as online dating, where platforms claim to use sophisticated algorithms to identify ideal partners. In reality, these systems often fail to significantly improve outcomes, as their ability to predict human behavior remains limited.

Under uncertainty, seeking an optimal solution is often misguided. Instead, we should aim for satisfactory solutions. One practical tool that reflects this approach is the “fast-and-frugal tree,” a simple decision-making structure resembling a checklist. By focusing on a small number of key questions, it enables decisions that are not perfect, but reliably effective.

Gigerenzer presents a striking example of such a decision tree used at military checkpoints to reduce civilian casualties. Built on just three questions, it reportedly reduced casualties by more than 60 percent. Similarly, simple decision trees used to identify failing banks can match or even outperform complex algorithms. In uncertain environments, these transparent and straightforward tools often outperform models that rely on vast datasets and assume the future will mirror the past.

Fast and Frugal Tree at Military Checkpoint

Self-driving cars illustrate another domain where uncertainty poses a major challenge. Human behavior is inherently difficult to predict. Although drivers are expected to follow rules, the frequency of accidents and violations shows how often those rules are ignored. They break rules, improvise, hesitate, and react emotionally. The real barrier to autonomy is not computation, but unpredictability.

This pursuit of predictability is one of the driving forces behind the concept of smart cities. For example, Toyota has announced plans to build a prototype city near Tokyo and Mount Fuji that is adapted to autonomous cars and where pedestrians will walk out of reach of the cars.

An alternative approach is to reduce reliance on cars altogether by promoting bike-first infrastructure and public transportation, which are much more resilient solutions, and don’t depend on solving AI’s hardest problems.

So what do humans have that machines still lack?

Gigerenzer calls it “common sense,” but that label undersells it. It is a powerful, deeply integrated toolkit:

• Causal thinking – the ability to build mental models of the world and understand relationships between elements. This can also lead to errors, especially when statistical reasoning is required.

• Intuitive psychology – the capacity to infer what others know, think, and believe.

• Intuitive physics – an innate understanding of time, space, and physical interactions, observable even in young children.

• Intuitive sociality – an internal drive to follow social norms and uphold moral standards.

Together, these abilities form a flexible and adaptive decision-making system that is difficult to replicate with algorithms due to its nonlinear and context-sensitive nature.

In uncertain environments, less can be more. Fewer variables. Simpler rules. Faster decisions. Better outcomes.

And yet, the dominant narrative insists that more data is the answer. That with enough historical information, the future becomes predictable. It is an appealing story—and a profitable one. But it is often more marketing than reality.

We are overestimating what machines can do—and underestimating what humans already do extraordinarily well.

The common-sense conclusion is simple:

I publish one short reflection like this each week. If you would like to receive them by email, you can sign up to the newsletter here

References and Notes

1. Gigerenzer, G. (2022). How to stay smart in a smart world: Why human intelligence still beats algorithms. Penguin.