Spot -- Level 4
1I'm not convinced that this set of videos is particularly "healthy" for me in hindsight, because this game, under any of its particular names, is quite addictive and suddenly you could devour a whole bunch of games in rapid succession.
I actually had an unusual number of technical failures while trying to capture these...and it's surprising, because it was a whole bunch of issues I thought I'd already figured out and got squared away long ago. Sometimes the video capture wouldn't be recognized and would fail our mid-stream. Other times I had whole captures turn out to have no sound, which is in particular is a pretty hard to pin down problem, to the point that I always make a test capture first just to be sure...this time I didn't. I've lost entire (short) projects this way, actually.
So anyway, the only reason why I mention this is because it gave me a lot of time to get back settled in, but it also gave me a lot of time to observe the AI opponent in action and under different sets of circumstances. For one thing, it's surprisingly inconsistent at times, even on this supposedly advanced setting. I mean, don't get me wrong, when it's presented a good move, it'll often take it...and when it's presented a "good" move with a serious opening afterwards, it won't take it. It's even good enough to not take a move that I didn't even realize would've made for a really bad follow-up afterward.
The "problem" it has seems to be in the beginning, where the most variation in play occurs. It's not surprising, to say the least, because the more options it has and the more of them being apparently "equal" under the circumstances, the harder it would be to expect it to choose between them. So I must assume that in addition to a one-move lookahead, the AI is programmed to select randomly among otherwise numerically equivalent options.
Of course, this goes completely out the window now and then, particularly at the lower difficulties. It would be funny to find out that the lower-grade AI players are simply based off the highest level and then composed of a random decision engine to make a "good" move only some of the time and a more or less random one otherwise. It's hard to tell, much less reverse engineer or pass judgment upon its routines...but it does get more predictably exacting and borderline cutthroat when the chips are down.
Thus, the AI can be safely said here to be much better in the mid- to endgame than in the early stages.
---
One thing I found amusing that I didn't recall, probably from an extreme lack of messing with it, is that the AI has some kinda funky "thinking" times. I'm pretty bad at estimating the kind of computing power necessary to run even middling-complexity algorithms, particularly these days with Moore's Law (not Murphy's Law, mind you...but you might think that Murphy has you in mind when it comes to AI opponents in games like this) running amok and computers getting incredibly powerful and even more incredibly compact...I'm not actually sure offhand what the Game Boy was capable of, but the results are interesting.
The interesting thing is that it appears as if the AI uses up all its turn time on five seconds sometimes, but it never gets clocked out and forced to skip a turn. In fact, contrary to its appearances of being "rushed," most times I tried it on shorter times, it seemed to make even better moves than usual. Of course, given my earlier hypotheses as to the nature of the AI's thinking patterns, this could've been entirely incidental.
Anyway, the ultimate point to all this seems to be that I couldn't discern any serious deviation in terms of playing ability regardless of how long it's given to make a move, so if it's capable of feeling pressured, it doesn't show it particularly strongly.
---
In fact, the greatest deviation of all was actually the beginning of the first match shown here. I was honestly quite surprised how poorly it set up, falling directly into my usual starting "gambit" of trying to offer only groups of two for it to assimilate, and trying (and usually failing) to prevent it from cloning into capturing such a group. If you play this way, a group of four spaces can relatively safely be made your own.
To put it (relatively) simply, "triangle" formations of three are weak on one side, and easily made into "squares" that offer weaknesses on no sides. Furthermore, to attack a "square" means to create a "triangle" in the process, with one of the pieces from the "square" able to retaliate in kind. Of course, all this flies right out the window and becomes many times more complex after the first iteration, because then there are additional adjacent pieces involved.
But, don't let that make you feel as though the whole notion of defense is entirely out of your hands, because a strong beginning makes the rest of the game dramatically more controllable. Of course, it's also exceedingly difficult to stamp out another player entirely, as you might imagine.