Researchers warn AI in policing may deepen discrimination against minorities. AI tools learn from biased past records and risk creating feedback loops. Crime predictions can reflect where police have looked rather than actual crime. Dirty data from unlawful or biased policing periods can train these systems. Independent checks of training data are urged to remove bias. Technical fixes alone are insufficient without oversight of data origins. AI can amplify existing human biases in policing decisions. Experts call for greater human oversight in AI policing use.