Third edition of Artificial Intelligence: foundations of computational agents, Cambridge University Press, 2023 is now available (including the full text).

### 10.4.2 Learning to Coordinate

Due to the existence of multiple equilibria, in many cases it is not clear what an agent should actually do, even if it knows all of the outcomes for the game and the utilities of the agents. However, most real strategic encounters are much more difficult, because the agents do not know the outcomes or the utilities of the other agents.

An alternative to the deep strategic reasoning implied by computing the Nash equilibria is to try to learn what actions to perform. This is quite different from the standard setup for learning covered in Chapter 7, where an agent was learning about some unknown but fixed concept; here, an agent is learning to interact with other agents who are also learning.

This section presents a simple algorithm that can be used to iteratively improve an agent's policy. We assume that the agents are repeatedly playing the same game and are learning what to do based on how well they do. We assume that each agent is always playing a mixed strategy; the agent updates the probabilities of the actions based on the payoffs received. For simplicity, we assume a single state; the only thing that changes between times is the randomized policies of the other agents.

**Procedure**PolicyImprovement(

*A,α,δ*)

2:

**Inputs**

3:

*A*a set of actions

4:

*α*step size for action estimate

5:

*δ*step size for probability change

6:

**Local**

7:

*n*the number of elements of

*A*

8:

*P[A]*a probability distribution over

*A*

9:

*Q[A]*an estimate of the value of doing

*A*

10:

*a_best*the current best action

11:

12:

*n ←|A|*

13:

*P[A]*assigned randomly such that

*P[a]>0*and

*∑*

_{a∈A}P[a]=114:

*Q[a] ←0*for each

*a∈A*

15:

**repeat**

16:

**select**action

*a*based on distribution

*P*

17:

**do**

*a*

18:

**observe**

*payoff*

19:

*Q[a]←Q[a]+ α(payoff-Q[a])*

20:

*a_best ←argmax(Q)*

21:

*P[a_best]←P[a_best]+n×δ*

22:

**for each**

*a'∈A*

**do**

23:

*P[a']←P[a']-δ*

24:

**if**(

*P[a']<0*)

**then**

25:

*P[a_best]←P[a_best]+P[a']*

26:

*P[a']←0*

27:

28:

29:

**until**termination

The algorithm *PolicyImprovement* of
Figure 10.9 gives a controller for a
learning agent. It maintains its current stochastic policy in the *P* array
and an estimate of the payoff for
each action in the *Q* array. The agent carries out an action based on
its current policy and observes the action's
payoff. It then updates its estimate of the value of that action and
modifies its current strategy by increasing the
probability of its best action.

In this algorithm, *n* is the number of actions (the number of
elements of *A*). First, it initializes *P* randomly so that it is a
probability distribution; *Q* is initialized arbitrarily to zero.

At each stage, the agent
chooses an action *a* based on the current
distribution *P*. It carries out the action *a* and observes the
payoff it receives.
It then updates its estimate of the payoff from doing *a*. It is doing
gradient descent with learning rate
*α* to minimize the error in its estimate of the payoff of action
*a*.
If the
payoff is more than its previous estimate, it increases its estimate
in proportion to the error. If the payoff is less than its estimate,
it decreases its estimate.

It then computes *a_best*, which is the current best action according to its estimated *Q*-values. (Assume that, if there is more than one best action, one is
chosen at random to be *a_best*.) It increases the probability of the best action by *(n-1)δ* and reduces
the probability of the other actions by *δ*. The *if* condition
on
line 24 is
there to ensure that the probabilities are all non-negative and sum to 1.

Even when *P* has some action with
probability 0, it is useful to try that action occasionally to update its current
value. In the following examples, we assume that the agent chooses a
random action with probability 0.05 at each step and otherwise
chooses each action according to the probability of that action in *P*.

An open-source Java applet that implements this learning controller is available from the book's web site.

**Example 10.17:**Figure 10.10 shows a plot of the learning algorithm for Example 10.12. This figure plots the relative probabilities for agent 1 choosing football and agent 2 choosing shopping for 7 runs of the learning algorithm. Each line is one run. Each of the runs ends at the top-left corner or the bottom-right corner. In these runs, the policies are initialized randomly,

*α=0.1*, and

*δ=0.01*.

If the other agents are playing a fixed strategy (even if it is a
stochastic strategy), this algorithm converges to a best response to
that strategy (as long as *α* and *δ* are small enough, and
as long as the agent randomly tries all of the actions occasionally).

The following discussion assumes that all agents are using this learning controller.

If there is a unique Nash equilibrium in pure strategies, and all of the agents
use this
algorithm, they will converge to this equilibrium. Dominated strategies will
have their probability set to zero. In
Example 10.15, it will find the Nash equilibrium.
Similarly for the prisoner's dilemma in Example 10.13, it
will converge to the unique equilibrium where both agents take. Thus,
this algorithm does *not* learn to **cooperate**,
where cooperating agents will both *give* in the prisoner's dilemma to maximize their payoffs.

If there are multiple pure equilibria, this algorithm will converge
to one of them. The agents thus learn to **coordinate**. In the
football-shopping game of Example 10.12, it
will converge to one of the equilibria of both shopping or both going
to the football game. Which one it converges to depends on the initial strategies.

If there is only a randomized equilibrium, as in the penalty kick game of Example 10.9, this algorithm tends to cycle around the equilibrium.

**Example 10.18:**Figure 10.11 shows a plot of two players using the learning algorithm for Example 10.9. This figure plots the relative probabilities for the goalkeeper jumping right and the kicker kicking left for one run of the learning algorithm. In this run,

*α=0.1*and

*δ=0.001*. The learning algorithm cycles around the equilibrium, never actually reaching the equilibrium.

Consider the two-agent competitive game where there is
only a randomized Nash equilibrium.
If an agent *A* is playing another agent, *B*, that is
playing a Nash equilibrium, it does not matter which action in its
support set is performed by agent *A*; they all have the same value to *A*. Thus, agent *A* will tend to
wander off the equilibrium. Note that, when *A* deviates from the
equilibrium strategy, the best response for agent *B* is to play
deterministically. This algorithm, when used by
agent *B* eventually, notices that *A* has deviated from the
equilibrium and agent *B* changes its policy. Agent *B* will also deviate from the
equilibrium. Then agent *A* can try to exploit this deviation.
When they are both using
this controller, each agents's deviation can be exploited, and they
tend to cycle. There is nothing in this algorithm to keep the agents on a randomized equilibrium. One
way to try to make agents not wander too far from an equilibrium is to adopt a
**win or learn fast (WoLF)** strategy: when the agent is winning
it takes small steps (*δ* is small), and when the agent is losing
it takes larger steps (*δ* is increased). While it is winning, it
tends to stay with the same policy, and while it is losing, it tries
to move quickly to a better policy. To define winning, a simple
strategy is for an agent to see whether it is doing better than the average payoff
it has received so far.

Note that there is no perfect learning strategy. If an opposing agent
knew the exact strategy (whether learning or not) agent *A* was using, and could predict
what agent *A* would do, it could exploit that knowledge.