## 9.2 One-Off Decisions

Basic decision theory applied to intelligent agents relies on the following assumptions:

- Agents know what actions they can carry out.
- The effect of each action can be described as a probability distribution over outcomes.
- An agent's preferences are expressed by utilities of outcomes.

It is a consequence of Proposition 9.1 that, if agents only act for one step, a rational agent should choose an action with the highest expected utility.

**Example 9.4:**Consider the problem of the delivery robot in which there is uncertainty in the outcome of its actions. In particular, consider the problem of going from position

*o109*in Figure 3.1 to the

Thus, the robot has to decide whether to wear the pads and which way to go (the long way or the short way). What is not under its direct control is whether there is an accident, although this probability can be reduced by going the long way around. For each combination of the agent's choices and whether there is an accident, there is an outcome ranging from severe damage to arriving quickly without the extra weight of the pads.

To model one-off decision making, a **decision
variable** can be used to model an agent's
choice. A
decision variable is like a random variable, with a domain, but it
does not have an associated probability distribution. Instead, an agent
gets to
choose a value for a decision variable. A **possible world** specifies
values for both random and decision variables, and for each
combination of values to decision variables, there is a probability
distribution over the random variables. That is, for each assignment
of a value to each decision variable, the measures of the worlds
that satisfy that assignment sum to *1*. Conditional probabilities
are only defined when a value for every decision variable is part
of what is conditioned on.

Figure 9.4 shows a **decision
tree** that depicts the different choices
available to the agent and their outcomes. [These are different from
the decision trees used for
classification]. To read the decision tree, start
at the root (on the left in this figure). From each node one of the
branches can be followed. For the decision nodes, shown as squares,
the agent gets to choose which branch to take. For each random node, shown
as a circle, the agent does not get to choose which branch will be
taken; rather there is a probability distribution over the branches
from that node. Each path to a leaf corresponds to a world, shown as *w _{i}*,
which is the

**outcome**that will be true if that path is followed.

**Example 9.5:**In Example 9.4 there are two decision variables, one corresponding to the decision of whether the robot wears pads and one to the decision of which way to go. There is one random variable, whether there is an accident or not. Eight possible worlds exist, corresponding to the eight paths in the decision tree of Figure 9.4.

What the agent should do depends on how important it is to arrive quickly, how much the pads' weight matters, how much it is worth to reduce the damage from severe to moderate, and the likelihood of an accident.

The proof of Proposition 9.1 specifies how to measure
the desirability of the outcomes. Suppose we decide to have utilities
in the range [0,100]. First, choose the
best outcome, which would be *w _{5}*, and give it a utility of

*100*. The worst outcome is

*w*, so assign it a utility of

_{6}*0*. For each of the other worlds, consider the lottery between

*w*and

_{6}*w*. For example,

_{5}*w*may have a utility of 35, meaning the agent is indifferent between

_{0}*w*and

_{0}*[0.35 : w*, which is slightly better than

_{5}, 0.65:w_{6}]*w*, which may have a utility of 30.

_{2}*w*may have a utility of 95, because it is only slightly worse than

_{1}*w*.

_{5}**Example 9.6:**In

**diagnosis**, decision variables correspond to various treatments and tests. The utility may depend on the costs of tests and treatment and whether the patient gets better, stays sick, or dies, and whether they have short-term or chronic pain. The outcomes for the patient depend on the treatment the patient receives, the patient's physiology, and the details of the disease, which may not be known with certainty. Although we have used the vocabulary of medical diagnosis, the same approach holds for diagnosis of artifacts such as airplanes.

In a one-off decision, the agent chooses a value for each
decision variable. This can be modeled by treating all the decision variables as a single
composite decision variable. The
domain of this decision variable is the cross product of the
domains of the individual decision variables.
Call the resulting composite decision variable *D*.

Each world *ω* specifies an
assignment of a value to the
decision variable *D* and an assignment of a value to each random variable.

A **single decision** is an assignment of a value to the decision
variable. The **expected utility** of single decision *D=d _{i}* is

E(U|D=d_{i}) = ∑_{ω (D=di)}U(ω)×P(ω),

where *P(ω)* is the probability of world *ω*, and
*U(ω) * is the value of the utility *U* in world
*ω*; *ω (D=d _{i})* means that the decision variable

*D*has value

*d*in world

_{i}*ω*. Thus, the expected-utility computation involves summing over the worlds that select the appropriate decision.

An **optimal single decision** is the decision whose expected
utility is maximal. That is, *D=d _{max}* is an optimal decision if

E(U|D=d_{max})=max_{di∈dom(D)}E(U|D=d_{i}),

where *dom(D)* is the domain of decision variable *D*. Thus,

d_{max}=argmax_{di∈dom(D)}E(U|D=d_{i}).

**Example 9.7:**The delivery robot problem of Example 9.4 is a single decision problem where the robot has to decide on the values for the variables

*Wear_Pads*and

*Which_Way*. The single decision is the complex decision variable

*⟨Wear_Pads,Which_Way⟩*. Each assignment of a value to each decision variable has an expected value. For example, the expected utility of

*Wear_Pads=true∧Which_Way=short*is given by

**E**(U|wear_pads∧Which_Way=short)*= P(accident |wear_pads∧Which_way=short) ×utility(w*

_{0})*+(1-P(accident |wear_pads∧Which_way=short)) ×utility(w*

_{1}),where the worlds *w _{0}* and

*w*are as in Figure 9.4, and

_{1}*wear_pads*means

*Wear_Pads=true*.