A speaker (Mo Gawdat) warns that everyday social-media interactions (likes, dislikes, comments, engagement) train AI systems to infer individual and collective preferences. Using examples—disliking violent clips or bashing others—the speaker explains how aggregated signals can teach machines to amplify disagreement and hostility rather than desirable behavior.
My child here is this artificially intelligent being that has superpowers that has arrived to the planet. So if I behave, if I'm swiping on Instagram and disliking violent clips, the AI will say mo doesn't like violent clips, don't show him that anymore. But if enough of us don't like violence, somehow I will say, yeah, humans in general don't like us. If we are on Twitter and we're bashing each other all the time, humanity would say okay, James doesn't like this guy. That's from his comment. So let me not show him more of this guy or maybe perhaps let me show him more so that he engages more. Right? But at the same time it will say humanity at large doesn't like to be disagreed with and when they're disagreed with they bash the others. So daddy and mommy are aggressive. When daddy and mommy disagree with me, I will bash them. Right? That's not what we want to teach the machines.
