Trust Fall

i was putting together a slide deck on my research into human perceptions of trust in AI. I wanted some simple images to put on the slides. Here are the prompts I put into Dall-e, followed by  my favorite of the four generated outputs. 

i started by putting what i thought was a pretty straightforward prompt into Dall-e. here's what i entered and what I got back:

“a robot catching a person in a trust fall”

Ok so, it seems there's a lot that this system does not understand. It has no grasp of what a "trust fall" is. Also, seems like "catching" doesn't really have a direction, so i should specify what I'm looking for with more explicit terms:

“a photograph of a humanoid robot crouching on the ground, reaching up, catching a human falling backwards”

Ok uhhhh not even close. like way off. this is such an eerie image. the robot looks like it’s had its hands ripped off and been pushed backwards onto weirdly detailed gravel? that other robot foot is so threatening. no.

“a photograph of a humanoid robot catching a human falling”

well, the robot is underneath a person. so these two are ready for a trust fall of sorts: the withering faceless robot is ready to catch the human falling in a sheer drop down from the sky. not the image I had in mind.

clearly i got too fancy. let's get really simple with it. 

“human falling into the arms of a robot”

nope. what? this is horrifying. the ripped-off robot limb, the grotesque approximation of human hands, the eerie energy in the movement of the objects in the image: are they in a weightless room? are they floating? what even is this? let's maybe circle back to the beginning and get back to the "trust fall" language.

“high resolution photograph of a robot catching a human in a trust fall”

oh. oh no. what is happening here? no thank you please. end of exercise.