The Microsoft cofounder warned the next five years are critical for setting up guardrails to rein in the potential harms of AI.