I think it might come down to whether the transistor is making a decision. If the transistor is being used as a switch, that would seem to qualify as a decision, even though it's an extremely rudimentary decision.
Intelligence, in reference to Artificial (or Algorithmic) Intelligence, is not restricted to high intelligence. A brute force Tic-Tac-Toe AI has extremely low, narrow intelligence but still constitute AI. An automatic switch, which makes the most simple decision possible, a binary choice, would seem to be the most basic form of intelligence.
Norvig's definition seems rooted in game theory, which is important in terms of utility of intelligence. But in a condition of intractability one is only assuming one's decision is more optimal than other choices. Outcomes can be evaluated, and a determination made as to whether the algorithm was "smart" or "dumb", but these terms refer to relative positions on a spectrum.
It's worth noting this fundamental definition of AI opens up a can of worms in that pre-transistor, automated mechanical switches such as the Strowger switch would also probably qualify.
And automata do not have to be electrical. The History of Automation wiki suggests the first feedback control system was for a water clock invented by Ctesibius. This device dates to the 3rd Century BCE, and water clocks were said to be the most accurate time pieces until Huygens.
This type of intelligence I tend to think of as autonomic, in the sense of involuntary, and distinct from higher functions, which are more commonly associated with "intelligence".
Note: The characteristics of an autonomic system in a computing context are quite interesting and include self-optimization, self-learning, and self-awareness.
ADDENDUM: After much thought, in reference to @JT's answer, I can't see how decisions can be separated from goals—there needs to be an intent, or the decision is merely random. This might prompt the question "can a simple switch be said to have a goal?"