Question 1

a) Letting R(Ag Env)denote the set of all possible runs of agent Ag in environment Env, letting u(r) denote the utility of a run r, and letting P(r|Ag,Env) denote the probability that run r occurs given agent Ag is placed in environment Env, write an equation that defines the properties of the optimal agent given u and Env.

b) Consider the environment Env₁ = 0, Τ> defined as follows:

E = {e₀, e₁, e₂, e₃, e₄, e₅}

Τ(e₀ →^α₀) = {e₁, e₂}

Τ(e₀ →^α₁) = {e₃, e₄, e₅}

There are just two agents possible with respect to this environment, which we shall refer to as Ag and Ag :

Ag₁(e₀) = α₀

Ag(e₀) = α₁

Assume the probabilities of the various runs are as follows:

P(e₀ →^α₀ e₁|Ag₁, Env₁) = 0.4

P(e₀ →^α₀ e₂|Ag₁, Env₁) = 0.6

P(e₀ →^α₁ e₃|Ag₂, Env₁) = 0.1

P(e₀ →^α1 e₄|Ag₂, Env₁) = 0.2

P(e₀ →^α1 e₅|Ag₂, Env₁) = 0.7

Finally, assume the utility function u₁ is defined as follows:

u₁(e₀ →^α₀ e₁) = 8

u₁(e₀ →^α₀ e₂) = 11

u₁(e₀ →^α₁ e₃) = 70

u₁(e₀ →^α₁ e₄) = 9

u₁(e₀ →^α₁ e₅) = 10

Given these definitions:

i) Determine the expected utility of the agents Ag₁ and Ag₂ with respect to Env₁ and u₁.

ii) State with explanation which agent is optimal with respect to Env₁ and u₁.

c) Briefly identify the main advantages and disadvantages of specifying tasks via utility func- tions over runs, as in part (a), above, and suggest the main alternative.

Question 2:

a) With particular reference to the way in which they communicate with one another, contrast the concept of an object (in the sense of object-oriented programming) with that of an agent (in the sense of multiagent systems).

b) "In order to successfully communicate, agents require a common ontology of terms". With the aid of examples, and with reference to the Knowledge Interchange Format (KIF), ex- plain what you understand by this statement.

c) With the aid of examples where appropriate, explain the relative advantages and disadvan- tages of deliberative, reactive, and hybrid agent architectures.

Question 3:

a) Explain, with the aid of examples where appropriate, what you understand by coordination and coherence in the context of cooperative distributed problem solving systems.

b) "The CONTRACT NET protocol takes inspiration from the way in which contracts are put out to tender in human organisations." With reference to the way in which the CONTRACT NET works, and the key issues that must be addressed in implementing it, explain what you understand by this statement.

c) Suggest how the CONTRACT NET protocol may be implemented in terms of the FIPA agent communication language.

Question 4:

a) In the context of logic-based argument systems:

i) Define the notion of an argument.

ii) Define the notion of one argument undercutting another.

iii) Define the notion of one argument rebutting another.

b) The following figure shows an abstract argument system.

Explain the status of the following arguments, justifying your answer in each case:

a b c d e f g h i j

Question 5:

a) Define the prisoner's dilemma, and in particular explain why it is a "dilemma".

b) Define the notion of Nash equilibrium, and identify and explain the single Nash equilib- rium in the prisoner's dilemma.

c) In negotiation scenarios, the negotiation set is the set of proposals, or deals, over which the agents are negotiating. It is usually defined as the set of possible deals that are individual rational and Pareto optimal. Define what is meant by these terms.

d) Briefly define the monotonic concession protocol for negotiation.

e) The Zeuthen strategy for negotiation answers two questions that must be answered on any given round of negotiation: who should concede? and how much should they concede? Explain the answers that the Zeuthen strategy provides to these questions.