An AI will be extremely effective at whatever it learns. If it learns from bad people—by mimicking unethical people—an AI will be unconscionably evil.