A paper by Alan Chan and co-authors on automated AI R&D argues that handing AI research to AIs could lead to unpredictable outcomes and a loss of control