There are some practical reasons, and some pragmatic ones.
A "thinking" device is a device that is consuming power to think (to run the processor), and hardware and software designers have to keep a core user need in mind: battery life. The more the device "thinks" the lower the battery life. So devices are designed to be instant awake "on demand" computers that respond to user input. Siri, for example, probably uses a significant amount of processing power, and I'd wager if it was on all the time it would cut the "uptime" considerably.
Additionally you look at how people work with computers and computing devices, and the key idea of user control and freedom comes to play. Do you want to be "told" what to do by your phone, or to "tell" your phone what to do? Most users want to be/stay in control… And the idea of a "thing" telling them what to do just won't be appealing. This type of interaction has to be very carefully and well designed to work.
An area where this does "work" is with GPS, where the device "tells" the user what to do… But even then, there are limits. The user is still "in control" and the GPS is supporting the user's travel, not commanding. If a GPS comes off as "nagging" when informing the user of a traffic problem then the user will be… Well, I've tested of a similar solution, and it didn't work very well. Basically, the "tone of voice" is important in this and many other contexts. Get it wrong and you'll quickly get an angry user.
Some pragmatic reasons why devices are dumb… Well, we are still limited by old models and paradigms. We have desktop computers that are capable of millions of calculations per minute and we aren't doing anything innovative or new with that capacity. No one has come forward and invented an intelligent "butler" that used the innate information we all host on our own computing devices to, well, make suggestions. To help.
"I know you like Mythbusters, so I taped the new episode for you."
"There's a new Peter Gabriel album coming out. I downloaded the details and some samples and sent it to your iPad."
"You're electricity bill was due last week, would you like me to pay it for you?"
A smart assistant, who suggests but never takes control away from the user. All of these things are possible NOW, but no one has taken the next step to leverage the different systems and computing power together.
It's coming, though. Someday.
And if anyone wants to hire me to help design such a system you know where to find me…