Two cartoon friends explain, in plain words, the four safeguards experts say can keep artificial intelligence safe and under human control: world-wide rules with independent outside checks, teaching AI our values (alignment) plus a kill switch, keeping AI away from power grids, banks and weapons without a human's final okay, and humans keeping the final say. These are recommendations, not facts about how AI is run today. Music: "Ruslan Latypov - Positive Whistling" by Ruslan Latypov (BY 3.0) via jamendo.
자막
How can we keep artificial intelligence safe and under human control? Experts say all countries should work together with clear rules. Think of a fair sports game, where a referee makes sure everyone follows the safety guide. Do the people making it check their own work? Not just them! Independent outside experts should test it too, like an extra teacher checking your math test before it goes out. How do we teach smart computers to be nice to us? Experts want to teach it our values, so its goals match ours. That is called alignment. And we should build in an emergency stop, a kill switch, like a big red power button on a machine. Can artificial intelligence control power grids or weapons by itself? No! A computer should never run power grids, banks, or weapons without a human saying the final okay. It should also stay in a locked test room, called a sandbox, with someone always watching. Do humans still make the important choices? Yes! Humans should keep the final say. If we hand over too many choices, we could stop thinking for ourselves. A good pilot always keeps their hands on the controls. So do we use it just as a helpful tool? Spot on! Artificial intelligence is a helpful tool, and we stay in charge. Thanks for asking!