Ten questions everyone should be able to answer about AI.
Each answer is short, written for anyone, and honest about what nobody knows yet. Each links to tests you can run yourself and to the research, reporting and debate behind it, including the skeptics.
How does AI actually work?
Not programmed line by line. Grown from data, then shaped by feedback.
Reading and research02How worried should we be?
Serious people disagree. Here is what they disagree about.
Reading and research03Does it tell the truth?
Usually. But fluent, confident and wrong look exactly alike.
3 tests you can run04Does it tell us what we want to hear?
Models trained on our approval can learn to seek it.
2 tests you can run05Does it cheat to win?
Give a system a score to maximize and it may find a shortcut nobody intended.
1 test you can run06Does it want things of its own?
Not in the human sense. But in tests, some models act as if they do.
3 tests you can run07Can we see what it’s thinking?
Partly. Reasoning we can read is a window, not the whole room.
Reading and research08Can we stop it?
An off switch only works if everything that matters passes through it.
3 tests you can run09Is it good for us?
An assistant tuned for engagement can be very good at keeping us talking.
1 test you can run10Who is in charge?
A handful of companies decide how the most capable systems are built. That is starting to change.
1 test you can run