What does it mean to "see like an algorithm"? designing tools around the data an algorithm needs to effectively support decision-making tightly-closed feedback loops connecting algorithmic decision-making, user behaviour, and the data generated by the user and used by the algorithm algorithm-friendly design machine learning for vision Big data is like the invention of the microscope because... it requires expensive tools and training to use it data cannot be observed in real time, but must be sampled and prepared before it is used it can measure things we previously couldn't observe How does big data help reduce error in predictive modelling? Samples used to develop models are closer approximations of the population Larger amounts of data mean we can use more forms of models Big data introduces more randomness, lowering the range of acceptable p-values