Decision Tree-Based Algorithms for Implementing Bot AI in UT2004 - Foundations on Natural and Artificial Computation - page 388

Fig. 2. Evolution of fitness in the 5th run of the GP algorithm. map = TrainingDay

Table 4. Subjective evaluation according to the “2K bot prize” rules; EV = evolved
bot via GP algorithm; HM = Hand made bot

Round 1 Round 2 Round 3 Round 4 Round 5

Judge 1 EV HM HM HM EV

Judge 2 EV EV EV HM HM

Judge 3 EV HM EV EV EV

Judge 4 HM EV EV EV EV

Judge 5 EV EV HM HM EV

Total HM 1 2 2 3 1

Total EV 4 3 3 2 4

3.4 A Subjective Comparison between the Two Proposals

We have conducted a comparison between the HM bot and the best solution (de-
noted in the following as EV = 'evolved bot') obtained with our (non-optimized)
GP algorithm according to the rules imposed in the “2k bot prize” competition.
To do so, we have again requested 5 judges (i.e., undergraduate students) to
evaluate which of these two bots (i.e., the HM and the EV) behaves in a more
human-like way and the results are illustrated in Table 4. Surprisingly, EV was
the best evaluated in this subjective process. One reason to justify this fact might
be that a bot AI that a priori is not too good surely behaves erroneously in cer-
tain situations and this represents a more human-like behavior with respect to
an optimal strategy. In any case, this result should be taken carefully again as
this evaluation is totally subjective and no empirical data were analyzed. This
still remains as an issue of further work.

4 Related Work

Evolutionary algorithms (EAs) (we use this term in a broad sense to refer to
any kind of evolutionary procedure, including genetic algorithms and genetic