Question posted in 
I am new to AI and just started to reading Artificial Intelligence: A Modern Approach by Peter Novig and Stuart Russel. The second chapter talks about agents and says the following:
an agent’s choice of action at any given instant can depend on the
entire percept sequence observed to date, but not on anything it
hasn’t perceived.
According to this, you cannot call something an agent if at any given instant its action doesn’t depend on what it perceived some time in the past. Is this true? Let’s say a quad rotor starts its motors before any input is obtained, according to this definition it wouldn’t be an agent at that moment.
2
Answers
It’s still an agent.
All this means is that an agent doesn’t know anything that it hasn’t perceived, but an empty chain of perceptions is valid sequence. You might well program some reasonable defaults to handle this case, or the first step would be to read from your environment what you can.
Yes, it’s still an agent.
The key word here is can:
Agent’s action can depend on percept sequence but doesn’t have to.
A good example is a random agent which ignores percept sequence completely.