A cornerstone of scientific inquiry for centuries, the principle of Occam's razor, which advocates for the simplest explanation that fits the facts, is being re-evaluated by modern cognitive science. New research from the Santa Fe Institute suggests that this ingrained preference for simplicity might, in fact, be hindering our ability to uncover a more complete picture of reality.
Cognitive scientist and philosopher Marina Dubova is challenging the assumption that beginning with the simplest possible theory is always the optimal approach. Through a series of computer simulations and 'micro-world' experiments involving human researchers, Dubova's work delves into the psychological and cognitive biases that influence scientific discovery. Her findings indicate that some of the most cherished assumptions about how to best pursue truth may be on shaky ground, suggesting that a different approach, one that starts from complexity, could be more fruitful.
Occam's razor, attributed to the 14th-century friar William of Ockham, has historically guided scientists towards elegant solutions. Examples include Nicolaus Copernicus's heliocentric model, which replaced Ptolemy's Earth-centred universe, and the theory of continental drift, which more simply explained shared fossils across continents than the previous idea of sunken land bridges. However, Dubova argues that while this principle has served science well in many instances, it represents just one of several 'rules of thumb' that can inadvertently obscure deeper truths.
The tendency to favour simpler explanations is not limited to scientists. Psychologist Tania Lombrozo at Princeton University has conducted research showing that individuals generally prefer explanations that involve fewer causes or mechanisms and can account for the most data. For instance, in one study, participants asked to diagnose an alien with two symptoms overwhelmingly favoured a single disease explanation over two separate diseases, even when the latter was presented as more probable.
Dubova's own computational models further explore this bias. By creating AI agents tasked with developing theories based on limited data, she compared agents programmed to seek the fewest variables against those designed to embrace more complex explanations. The implications of this research are particularly pertinent as discussions turn to the automation of scientific processes and the development of 'AI scientists'. Embedding old ways of thinking into future AI systems, Dubova warns, could inadvertently limit their capacity for true discovery.