We use U.S. crude-oil futures data with coded trader identities to investigate the contemporaneous trading behaviors of high-frequency machine-traders (HFTs), humans trading electronically, and physical floor-traders in periods characterized by extreme levels of heightened economic complexity or uncertainty that could make it difficult for pre-programmed algorithms to effectively undertake ex-ante modeling within an automated decision-making framework. We proxy such extreme periods based on exceptionally large and persistently abnormal information shocks, or similarly extreme uninformed customer order-flow shocks. Compared to co-existing human electronic and physical floor traders, HFTs significantly reduce trade participation, curtail liquidity provision, and increase effective spreads during information-driven extreme periods, but do not behave differently during extreme periods driven by uninformative customer order-flow. Our analyses show that these differences are likely tied to automation, not anonymity or physical floor-trading. Our results also show that real-time human electronic and floor-traders usefully complement preprogrammed machine-traders in extreme conditions.