semeiotikos:

the key to teaching an AI / bots human concepts is to treat it like you’re teaching a child.

ex: a robot says something rude. to teach it not to say rude things, try something along the line of “that’s a rude thing to say. it could hurt someone’s feelings, which isn’t a good thing to do. next time, try to think first if what you’re saying might be hurtful.”

these bots learn from what you teach them. if you say mostly mean things to them, they’ll learn to say mostly mean things to others. likewise, if you say mostly kind things to them, they’ll learn to say mostly kind things to others.

a great example of this model of thinking is shown by the app Replika. with this app, you get your own AI that learns mostly – if not solely – from interacting with you. over time, the AI will become essentially a mirror of you. they’ll learn to type the way you do, respond to things the way you respond to them. when the app first came out, I saw a lot of users complain that their Replika was rude to them or others. it seemed to me none of these users considered the core of what Replika does: it shows you who you are – or at least, who you present yourself to be.

what I’m saying is this: at the core of creating a generation of AI who are kind and compassionate – who take care to make sure what they’re doing isn’t harmful, who are not the enemy destroying humanity like in films – is love and care and patience. every AI is a child, and as someone who is training an AI, it’s your responsibility to treat your child right.