Evaluation Process in Practice
The process of choosing is mostly based on evaluating the benefit of an object. The theory behind evaluating fitness values has already been considered. However, things are much different in practice. Providing a good estimate of fitness for any object is extremely difficult.
Fundamentally, the process is about mapping features onto possible choices according to criteria. To make this easier, humans use two different techniques: one based on logical reasoning, and the other based on experience.
It's certainly possible to perform critical analysis of any situation. This process breaks down the decision into smaller parts that can be understood easily.
The first step of this process involves extracting key features of the situation (for instance, low health, constrained environment, and being chased). From these facts, it is possible to induce what weapon properties are necessary. This requires a simple form of knowledge in the form of statements (for instance, if being chased, use a weapon with a fast firing rate)—very similar to the production rules covered in Chapter 11, "Rule-Based Systems." Finally, it's possible to select the most suitable weapon by finding one that matches the induced properties.
This approach assumes that it's possible to model this knowledge and decision process. It certainly is feasible to do so in many cases, although it's questionable whether humans do. When applying the rigorous method, we can only take into account features that we are aware of. It's quite easy to overlook secondary features and criteria when evaluating a choice! Often, a simpler empirical approach is favored.
Humans have the gift of memory, which can save them from countless deliberations. With memory, the history of fights can be drawn upon to estimate the benefits of weapons; "I was shot from behind last time here, I have a bad feeling ... I'll prepare my weapon for ambush!"
Certainly, when using an approach based on experience, it is essential to have history in combat—at least if the decisions are expected to be any good! This is where games (and simulations generally) have the advantage; it's possible for players to run through different scenarios quickly, without running as many risks. So, with simulation, it's eventually feasible to try every scenario. For each specific situation, it's possible to know how well the weapon performs.
Humans can get close to this kind of knowledge by sharing their experience with each other and training extensively. Therefore, this approach is highly applicable to people as well.
Even if the outcome of individual situations is unpredictable, with experience, it's possible to extract a trend from the outcome (like gathering statistics). This is sufficient to estimate the effectiveness of a weapon.
Both alternatives—whether based on deduction or experience—have their advantages and disadvantages. The first uses heavy reasoning based on simple knowledge, whereas the second uses no reasoning but assumes extensive experience. Both approaches prove useful, but an option somewhere down the middle may be more applicable.
Humans use a combination of reasoning and experience. Using past experience improves the estimates, whereas deduction enables us to apply the knowledge to new situations; "I won the duel in a similar situation with this weapon, so I'll try it again."
There is a clever balance in gathering knowledge by experience, and using logical reasoning to manipulate this knowledge further. Such a compromise is beneficial for simulations, too, because there is a trade-off between the time needed to acquire data and the accuracy of the estimates.