Designing Autonomous AI

Chapter 5. Teaching Your AI Brain What to Do

In Chapter 3 I discussed how every decision is an action taken to get closer to a goal. This is true for all kinds of decision-making environments. Table 5-1 shows a few examples, most of which I’ve introduced in earlier chapters.

Table 5-1. Examples of decisions, goals, and goal states

Activity	Decision	Goal	Goal state
Chess	Which move to make (which piece to move and to which destination square)	There are a variety of goals related to specific strategies, but the overall goal is to pin the enemy king so that it cannot move.	Any board position where the enemy king is pinned in checkmate
Airport baggage handling	Choose a method to move each bag from one plane to the next, plus follow-on decisions (if a cart, which cart trip will take a given bag?)	Deliver as many bags as possible to a connecting flight before it departs	Maximum possible number of correct bags at correct gates before flight departs
Naval game fleet planning	How many ships to build and how much armor and armament to build into each ship	Sink all enemy ships	Any naval battle arrangement where all enemy ships have been sunk
Drone flight	How fast to spin each rotor and which direction to tilt each rotor	Travel to destination using minimal energy	Safe landing at destination location, maximum battery charge level
Rock crusher	How fast to run conveyor, how to adjust crusher arm and casing	Crush enough rock to produce enough final product to satisfy demand	Throughput ...