-LzR- said:
The Falcon Punch test doesn't really prove anything. They just notice you using it and then adapt to it like they are programmed. Why don't you do so that you play against a lv9 Falcon for an hour without doing anything, count the amount of punches. Then spam Falcon punch a lot and then run the test again. If there was a significant difference, it should tell something, which isn't learning, but adapting to stuff.
Uhh, ok...
What part of "I'm not saying they learn" didn't you understand?
And what part of "literally at least half of what the Falcon CPU did in regular games was falcon punch" didn't you understand (particularly the "literally" part)?
I'm not going to waste 2 hours re-doing the test just so you can get some "legit numbers". Before the test, the falcon cpu did maybe 2-3 falcon punches per game at most. After the test, literally half its' moves or more were falcon punches, including in ridiculous instances where if you were offstage it would jump off after you and try to falcon punch you.
Even without recording actual falcon punch counts, it was quite obvious that there was a significant difference in its move choices. If this isn't proof of at least some type of mimicry, then I don't know what is.
LzR, you seem to think we're saying the AI learns on a complex level. Nobody said this. All we are saying is they try to mimic humans in a very limited fashion. My falcon punch test proves that, if a human does falcon punch a lot, then the CPU will also have a tendency to falcon punch a lot. The test was never to determine that the CPU will falcon punch in exactly the same situations and with exactly the same spacings as a human would. Simply having the AI recall an input sequence is trivial to program. Determining common input sequences based on play records is also trivial in programming.
Sorry for the heated tone, but it kind of annoys me when people just ignore what you say, or add in factors to your argument that you simply never said. >.>