Robert Maxwell wrote:
Likewise, let's make the leap that someone does create a generalized AI--but it requires a trillion dollars' worth of hardware and consumes a city's worth of electricity, and it makes decisions at a slower rate than a human. Is that ever going to be very broadly useful? I doubt it.
Which goes back to my point about sentient AI being fundamentally impractical. I can't emphasize enough the point that most of what you'll need an AI to do for you are NOT the kinds of things you need a self-aware, fully sentient computer for. You don't really need Siri to have emotional states or personal motivations like a true secretary, nor does the computer system that drives your car for you need to think like an actual motorist. Secretaries and human drivers can get distracted or annoyed or angry or jealous and this can compromise their job performance all the same.
On the other hand, some clever programming tricks and some refined algorithms turned Siri into a kind of wisecracking personal assistant just a couple of pods short of a HAL-9000.
I can think of one application of a trillion dollar AI. A starship computer. Think of a starship one that takes thousands of years to arrive at the nearest star, it is propelled by an ion drive that is powered by an atomic fission reactor, part of the payload is an O'Neil colony, an ISland One Bernal Sphere, but it is not a generation ship, as the journey to the nearest star would take longer that recorded history, instead we have a computer AI that monitors the ship and the environment within the Bernal sphere, and when the time of final approach to the star system is arrived at, frozen human embryos are revived and implanted in artificial wombs in the starship, babies are raised to adulthood by the ship's computer, and sometime, perhaps by the year 10,000 AD the first humans set down on the terraformed surface of a planet orbiting Alpha Centauri A.