Learning to Assist Humans without Inferring Rewards Vivek Myers 1 Evan Ellis 1 Sergey Levine 1 Benjamin Eysenbach