Sam Adams, an IBM Distinguished Engineer, tells his experience on the Joshua Blue project. Rather than take a traditional symbols based approach to creating an artificial general intelligence, he and his co-researchers modeled their mental architecture on human mental and physical development. The premise for this was that any 3 year old can do things that AI researchers see as hard problems: motor control, speaking and understanding language, basic physics, cause and effect, and adaption to new environments. The Joshua Blue project involved the research of developmental psychologists and neurobiologists to create a computational equivalent of a baby's mind, that would follow the same developmental trajectories that humans do, by imitating the synaptic structure over time and following developmental milestones.

An infant has a basic set of emotions and instinctive behaviors that it uses to gradually build a model of their body and their surroundings. They form predictions based on coincident observations and pursue those actions that lead to pleasure and avoiding those that lead to pain while forgetting those that aren't good at predicting outcomes. This basic architecture of superstition and forgetfulness that is present in humans was the framework for Joshua Blue. A superstition is just an association of two events that are observed together. Those that are observed repeatedly are strengthened to become beliefs and guide actions, while those that are not are forgotten to make room for new superstitions.

Some of the surprising results of this research related to emotion, forgetfulness, and the adaptability of the mental model. Emotions turned out to be a powerful control system because effects could be easily compared based on what emotions they caused, which helped prioritize. Forgetting helped improve the mental model because associations with weak predictive power would be removed to make room for more useful associations. This forgetfulness is how a growing child can adapt to a changing body size or to a new environment, which is also a goal of artificial general intelligence research.