#bias #ai # [[Epistemic status]] #shower-thought # Cognitive bias learned by AI #to-digest First, let's not mistake statistical biases for [[Cognitive biases|cognitive biases]]. Most of the time when you hear talking about biases in ML, they talk about statistical biases (i.e., facial recognition trained on a dataset with more white than black humans), but cognitive biases are more fundamental to intelligence, I don't understand why it is so ignored? [[Artificial intelligence|AI]] learn a [[The Map is not the Territory|map of the territory]] largely based on the map humans have drawn, for example if you train a language model with a dataset that contains sentences evoking cats running after mouses, the model will associate cat & mouse (and maybe as predator and prey). All these datasets are obviously based on human language, and therefore you could find [[Cognitive biases|cognitively biased]] decision-making samples, such as [[Survivorship Bias]]: "Bob has earned $1000 short-trading Bitcoin, so he is confident now about investing $10.000". The learned model will have this bias imprinted in its intelligence. # External links More about statistical biases: <iframe width="560" height="315" src="https://www.youtube.com/embed/WjAkzPhqsxo" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>