For those in a hurry, this:

To align an AI's goals with ours, we must build-in alignment from the start,
To keep an AI honest, we must build-in honesty from the start,
To get an AI to understand anything, we must invest it with something of what it's like to be conscious.
In this book, a theory of consciousness is cast into an AI architecture that allows interventions in concept formation by design.
For the rest of you, who enjoy reading and mulling things over, this:
Can a computing device appreciate the smell of coffee on a Sunday morning, or contemplate the Earth as seen from the Moon, or worry about inflation and the price of fuel?
Not without being conscious and understanding the world. And one can't be done without the other,
surely?
In this book, Carter Blakelaw uses a theory of what makes us conscious to present a machine that will genuinely think for itself.
Not only that, but once he has his machine, he looks at how to ensure its interests align with our own, and how to keep it honest and true (alignment and hallucinations being two of the biggest issues in AI).
Discover what he discovers about the machine, about our world, and about us.