The answers are in boldface.

1. How do we define an agent in this class?

   • An agent is a computer system that is situated in some environment, and that is capable of autonomous action in this environment in order to meet its design objectives.
   • An agent is a computer system that performs actions on the user's behalf and has adjustable autonomy.
   • An agent is a computer system that has intelligence and can perform complex tasks.
   • An agent is a computer system that communicate with other agents using an agent communication language and that is implement using beliefs, desires, and intentions.
   • An agent is an object that can say "no".

2. An accessible environment is one where

   • the agent can observe everything.
   • the agent's actions have guaranteed effects.
   • the environment does not change unless the agent takes an action.
   • there is a finite set of possible actions.
   • the agent can go anywhere.

3. Which one of the following is not a difference between agents and objects?

   • Objects encapsulate state, agents do not.
   • Objects do it for free; agents do it for money.
   • Agents have pro-active behavior, objects do not.
   • Agents have their own thread of control, objects do not.
   • Objects support inheritance, agents do not.

4. Let E be the set of all environmental states, A be the set of all actions, and R be the set of all possible runs of the system. We can then define a reactive agent as

   • A function from E to A.
   • A function from R to A.
   • A function from RxE to A.
   • A function from the last n > 1 steps of R to A.
   • A function from the agent's state S to A.

5. In the formal model of an agent with state, what is the state?

   • The internal state of the agent.
   • The current state of the world.
   • The last perception.
   • The state of the agent's interaction with another agent.
   • The function that maps environmental states to actions.

6. Deductive reasoning agents suffer from the transduction problem. What is the transduction problem?

   • Translating from the real world to a symbolic representation.
   • Manipulating symbols so as to arrive at reasonable conclusions.
   • Proving theorems fast enough so that the results are still relevant.
   • Translating what the other agent's say into the agent's own language.
   • Locating a position on the environment where the agent can maximize the amount of usable input it receives.

7. An agent that picks its actions based on theorem-proving does so by:

   • First trying to prove if it can derive Do(a) for some a from its database of knowledge.
   • For each possible action, trying to prove that taking this action will not lead to any contradiction in its database of knowledge.
   • Deriving all the possible statements from its database and then choosing the action that is mentioned more often.
   • Proving that a particular action can be taken, given its database of knowledge.
   • Finding a statement in its database of knowledge that states that the action should be taken now.

8. Why do we like to build agents using mentalistic notions?

   • Because we hope it will simplify the programming of agents.
   • Because Dennet's intentional stance says that this is a good way to build agents.
   • Because philosophical agents are more adaptive than programmatic agents.
   • We do not. Mentalistic notions are reserved for thinking about agents and not for building them.
   • Because traditional software engineering constructs are not useful for building software agents.

9. An Agent0 program consists of

   • a series of conditional commitment statements.
   • a set of of Lisp functions.
   • a set of intention plans and desire primitives.
   • knowledge areas and a blackboard for communication.
   • a knowledge database and a set of theorems.

10. Intentions are useful constructs for building agents because

    • they focus the agent's deliberation.
    • they are fleeting.
    • they reflect the tradeoffs between thinking and acting.
    • they can contradict each other so the agent can pursue multiple lines of thought at the same time.
    • they represent an agent's most preferred desires.

11. A practical reasoning agent uses a belief-revision-function, this function

    • eliminates contradictions from the agent's beliefs and incorporates new perceptions.
    • synchronizes the agent's beliefs to its desires.
    • synchronizes the agent beliefs to its intentions.
    • determines which intention should be chosen given the agent's beliefs.
    • revises the agent's beliefs so that they do not contradict any of the current perceptions.

12. The PRS-CL system we studied in class is

    • a BDI system.
    • a reactive system.
    • a means-end reasoning system.
    • a theorem-proving system.
    • a perfect reasoning system.

13. Which one of the following is not one of Rodney Brook's theses about intelligent behavior?

    • Simple interacting behaviors give rise to constructs which are akin to symbolic representations.
    • Intelligent behavior does not require explicit representations.
    • Intelligent behavior does not require abstract reasoning.
    • Intelligence is an emergent property of certain complex systems.
    • Intelligent agents need to be situated in an environment.

14. Say you build a subsumption architecture with five behaviors (using the (condition,action) notation we used in class):

    • A: (x, 1)
    • B: (x. 2)
    • C: (x, 3)
    • D: (x, 4)
    • E: (y, 5)
    and A inhibits B and C, B inhibits D, and E inhibits A. You then see the world state x. Which action does the system take?
    • A
    • B
    • C
    • D
    • E

15. Which one of the following is not a limitation of the subsumption architecture?

    • Its run-time is always n² in the number of behaviors.
    • Agents must have enough information in the local environment, at all times, in order to determine what to do.
    • It does not provide a way for the agent to incorporate information that it cannot directly see.
    • It is unclear how learning could take place in this architecture.
    • It is very hard to build agents that have many behaviors.

16. In a horizontal layered architecture

    • the perceptions are given to all the layers each of which chooses a, possibly different, action.
    • the perceptions are fed to the top layer, it filters down over the layers, and the bottom layer choses the action.
    • the perceptions are fed to the bottom layer, it filters up over the layers, and the top layer choses the action.
    • the perceptions are given to all the layers which then negotiate with each other in order to choose a layer which chooses the action.
    • there can be at most 3 levels of behaviors.
    • the perception processing is done by one of the layers, the rest of the layers then handle this processed data

17. Now that some years have passed since logic-based architectures were the norm and the subsumption architecture was introduced, what is the most popular approach to building agent architectures?

    • Use reactive architecture for low-level details and a logical approach for longer-term behaviors.
    • The subsumption architecture and its descendants..
    • BDI architectures.
    • Expert system shells such as Jess.
    • Use the subsumption architecture but replace some reactive behaviors with a more complex state-based behavior, if needed.

18. What is a performative verb?

    • One whose mere invocation causes the action it describes to be done.
    • One that asks another agent to do something.
    • One that serves to inform another agent about some fact.
    • One whose actor cannot be determined by a simple inspection of the phrase that contains it.
    • One that has several possible meanings depending on whether we interpret it literally or indirectly.

19. Which of the the following is a KIF representation of the sentence "An object is a Nimbus2000 if it is a broom and it flies"?

    • (defrelation nimbus2000 (?x) := 
         (and (flies ?x) (broom ?x)))

    • (and (nimbus2000 ?x)
         (and (flies ?x) (broom ?x)))

    • (nimbus2000 ?x)
      (flies ?x) 
      (broom ?x)

    • nimbus2000 <= flies = broom

    • [Commit (nimbus2000 ?x)
         [Cond
           (and (flies ?x) (broom ?x))]]

20. In KQML and FIPA-ACL the :reply-with is used for

    • to specify a string that the receiver should use when replying.
    • telling the receiving agent the language that should be used when replying to the message.
    • telling the receiving agent the time by which the sender expects the reply to arrive.
    • informing the receiving agent the preferred encoding mechanism.
    • telling the receiving agent the query that the sender wants answered.

21. In FIPA-ACL when an agent receives a message that it is unable to process it should respond with

    • not-understood
    • sorry
    • error
    • unable-to-handle
    • reply

22. If you want to send a FIPA-ACL message to a number of agents asking them to send you their bids for fixing your broken monitor, what performative would you use?

    • cfp
    • reply-with
    • query-ref
    • inform
    • confirm

23. How are the semantics of FIPA-ACL expressed?

    • Using a logical language based around the agents' beliefs and intentions.
    • Using detailed English descriptions.
    • Using first-order predicate logic and statements about the messages sent and received.
    • Using interaction protocols.
    • Using a derivative of the Agent0 language that was expanded to include uncertain beliefs.

24. Which one of the following statements about FIPA Interaction Protocols (IP) is not true?

    • An agent that claims to be FIPA-compliant must understand a small set of well-known IPs.
    • IPs are typical patterns of message exchange.
    • IPs are given well-known and fixed names.
    • An agent that claims to understand an IP must be able to do so.
    • Any agent can create new IPs.

25. AUML was a needed extension to UML because

    • agents run on their own thread.
    • UML has no way to represent message-passing.
    • agents can be reactive.
    • FIPA-ACL has a lot of performatives.
    • agents can have lifespans that are much longer than those of objects.

26. The Iterated Contract net Interaction Protocol differs from the Contract Net IP in that, in the Iterated Contract Net IP

    • The initiator can reject all proposals and send out a new call for bids.
    • The receiver can counter-propose.
    • There can be many receivers receiving the call for bids.
    • If the proposal is reject the agent that sent it can keep trying new proposals until, eventually, the initiator chooses somebody.
    • The initiator forwards its call for bids to an intermediary agent who negotiates directly with the other agents.

27. The FIPA Agent Message Transport does not specify

    • the semantics of the agent communication language.
    • the guidelines for the transport protocol.
    • the representation of the message envelope.
    • the representation of the agent communication language.
    • the encoding of a message as an envelope plus a body.

28. What is the job of the FIPA agent management system?

    • It tells what agents are on the system.
    • It serves as a "yellow pages" for the system, telling others what services are offered by which agents.
    • It manages the communications between agents.
    • It manages the interface between the agents' code and the FIPA code.
    • It provides agent mobility services.

29. FIPA provides

    • documents.
    • implementations.
    • money.
    • lobbying efforts.
    • testing services.

30. When one JADE agent sends a message to another JADE agent that resides on the same machine

    • the message might simply be copied into the other agent's queue.
    • the message will be sent using JAVA RMI.
    • the message will be sent using IIOP.
    • the message will first go to the AMS and then to the receiving agent.
    • the message will go directly to the agent via a socket.

31. On JADE, if the Behavior.done() function of a behavior returns true, then

    • the behavior will be deleted.
    • the next step of the behavior will be executed.
    • the behavior will be placed at the end of the queue.
    • the behavior's action method will be called.
    • the behavior's action method will not be called.

32. In JADE

    • each agent runs on a different and unique thread.
    • each behavior runs on a different and unique thread.
    • there is only one thread per agent container
    • each agent has two threads, one for communications and one for the behaviors.
    • each agent has three threads, one for the GUI, one for the behaviors, and one for communications.

33. The JADE function that an agent must call in order to send its information to the DF is called

    • register
    • addServices
    • setName
    • send
    • registerServices

34. In JADE, responder roles are

    • cyclic.
    • pre-defined to start after an initiator behavior.
    • used by slave agents.
    • complements to the relayer roles.
    • not used.

35. Which one of the following is not a benefit derived from using an ontology in agent communications?

    • It makes it unnecessary to use an encoding method.
    • Receiving agent can check the messages it receives for semantic consistency.
    • Messages can be understood by all agents that can understand the ontology used.
    • The actual information sent in the message is easier to identify, as opposed to using a system-specific encoding method.
    • It fulfills an aspect of the semantic web vision.

36. The Semantic Web vision calls for ontologies

    • to be created incrementally.
    • to reside in a central location.
    • to go through a standardization process.
    • to be defined with any encoding language available to the user.
    • to obey a top-level ontology.

37. Which one of the following statements about Semantic Web documents is not true?

    • They can reference terms without an ontology.
    • They are written in XML.
    • They are written in RDF.
    • They can reference many ontologies.
    • They can define new ontologies.

38. In the following piece of XML

    <alpha beta="gamma">
       text
    </alpha>
    

    which one is the element?

    • <alpha beta="gamma">
         text
      </alpha>
      

    • alpha
    • beta
    • gamma
    • text

39. Namespaces were a necessary addition to XML because

    • of the need to use multiple schemas in one document.
    • of the need to build really large XML documents.
    • of the proliferation of XML dialects.
    • of the excessive use of attributes.
    • of the move from DTDs to XML Schemas.

40. Using XML Schema one can define new data types by

    • restriction or extension
    • inheritance
    • grouping
    • aggregation
    • alliteration

41. What is the difference between the RDF ID and ABOUT attributes?

    • ID is used to define a concept, ABOUT is used when referring to an object.
    • They are both equivalent.
    • ID is used for resources, ABOUT is used for predicates.
    • ID can have any value, ABOUT's value must be a URN.
    • ID does not support the use of namespaces but ABOUT does.

42. Out of the following RDF snippets, which one is the equivalent of the English phrase "Anything that is a student is a person"?

    •   <rdf:Description ID="student">
          <rdf:type resource="http://www.w3.org/2000/01/rdf-schema#Class"/>
          <rdfs:subClassOf rdf:resource="#person"/>
        </rdf:Description>
      

    •   <rdf:student>
          <rdfs:subClassOf rdf:resource="#person"/>
        </rdf:Description>
      

    •   <rdf:Description ABOUT="student">
          <rdfs:isA rdf:resource="#person"/>
        </rdf:Description>
      

    •   <rdf:Class ID="student">
          <rdfs:isA rdf:resource="#person"/>
        </rdf:Description>
      

    •   <rdf:Description ID="student">
          <rdfs:isA rdf:resource="#person"/>
        </rdf:Description>
      

43. How many rdfs:classs does DAML define?

    • 12
    • 2
    • 0
    • 112
    • over 1000

44. Which one of the following is an example of a statement that can be defined in DAML but not in RDF/RDFS?

    • Things that have the isTall property cannot have the isShort property.
    • Anything that has the isTall property is a human.
    • All things that are men are human.
    • The property hasIncome maps a Worker to a Real.
    • There is a class called Firefly.

45. Which one of the following describes how a DAML-S service works?

    • serviceModel
    • serviceProfile
    • serviceGrounding
    • serviceInternals
    • serviceDetails

46. What is the significance of the fact that DAML-S uses IOPE's?

    • We can use AI planners to put together services.
    • There is not conflict with WSDL.
    • They allow for the automatic generation of code to execute the service.
    • They determine the time it will take for the service to execute.
    • We can have more fine-grained access control to services.

47. Which one of the following is not one of DAML-S's control structures?

    • Some-of
    • Sequence
    • Split
    • Choice
    • If-Then-Else

48. What is the Nash equilibrium(s) strategy for this game?

    	A	B
    A	4,4	5,0
    B	0,5	6,6

    • (A,A) and (B,B)
    • (A,B) and (B,A)
    • (B,B)
    • (A,A)
    • (A,A) and (A,B)

49. What is the social welfare strategy(s) for this game matrix?

    	A	B
    A	4,4	5,0
    B	0,5	6,6

    • (B,B)
    • (A,A) and (B,B)
    • (A,A)
    • none
    • (A,A) and (A,B) and (B,A) and (B,B)

50. What is the pareto optimal strategy(s) for this game?

    	A	B
    A	4,4	5,0
    B	0,5	6,6

    • (B,B)
    • (A,A)
    • (A,A) and (B,B)
    • (A,B) and (B,A)
    • none