UC Berkeley

Humans are powerful Bayesian learners – we rationally update our beliefs given new statistical evidence. In human history, there is ample evidence that we can completely overturn our previous beliefs and theories given new environmental input; we can also forego past tools, technologies, and social arrangements to adapt to new environments. However, are there limits to our ability to rationally revise our beliefs? Are there beliefs that are so entrenched that cannot be revised? Previous research has shown that human adults, but not non-human animals (i.e., chickens) can override a hard-wired perceptual “light-from-above” prior with new evidence. This raises the possibility that the learning mechanisms in humans may be more powerful than those in non-human animals. The current dissertation investigates whether beliefs in the most fundamental domains of human knowledge – the core knowledge systems of objects, agents, and social beings – can be revised given a small amount of counterevidence. Chapter 1 outlines the theoretical framework of this research and describes the three core knowledge systems that are investigated in Chapters 2–4. Chapter 2 assesses whether 4- to 6-year-old children and adults can revise three core knowledge principles in the object system – solidity (solid objects cannot pass through each other), continuity (objects traverse spatiotemporally connected paths), and contact (objects cannot interact at a distance). The findings show that children and adults can revise their beliefs about these object principles given a small number of violations of each principle. Chapter 3 uses the same methods as Chapter 2 to assess whether 4- to 6-year-old children and adults can revise three core knowledge principles in the agent system – goal (agents’ actions are goal-directed), efficiency (agents take the most efficient means to achieve their goals), and sampling (when an agent chooses objects that are in the minority of a population, they prefer that type of object). The findings show that children and adults can also revise their beliefs about these agent principles given a small number of violations of each principle. Furthermore, the agent principles are more easily revised than the object principles, suggesting that the agent system is more flexible than the object system. Chapter 4 investigates whether a novel paradigm of presenting statistically representative counterevidence can change 5- to 6-year-olds’ intergroup biases. The findings show that this paradigm successfully changes children’s biases about minimally defined social groups, and provide preliminary evidence that this paradigm is also effective in changing children’s racial biases. Finally, Chapter 5 synthesizes the findings from Chapters 2–4, and discusses the implications of this research for theories of cognitive development and social development. In conclusion, the current dissertation demonstrates that humans have powerful learning mechanisms and suggests that we might have the ability to revise any beliefs with new evidence. Future research will investigate how robust and long-lasting the belief revisions are, and whether this ability to revise even our most fundamental beliefs is unique to human learners.