[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

Is bot analysis of manually altered positions reliable?

Posted By: Timothy Chow
Date: Saturday, 31 October 2015, at 10:55 p.m.

In Response To: Is bot analysis of manually altered positions reliable? (Karol Szczerek)

Everything is possible, of course. But I don't really see a compelling argument that the way the neural net generalizes is going to be so simply analyzed.

Let me put it this way. Let's say that the probability that your first position arises in a real game is—oh, I don't know—let's say 10–1000. Suppose the probability that the second position arises is instead 10–1001. Does that mean that the bot is ten times more likely to get confused about the second position than the first?

When I put it this way, it's clear that this is much too naïve a way to think about the two positions. It's not just the raw probability that the position arises that matters. Analyzing the net must take into account the features that it uses. But there is no reason to think that the neural net's features line up with the features that we humans use to assess the position. In particular, I don't think that any kind of analysis along the lines of "Position A looks slightly weirder to me than Position B; therefore the bot is more likely to misunderstand Position A than Position B" is at all trustworthy, unless you can back it up with hard data. Surely you have encountered many positions that seemed strange to you where the bot seems to have no trouble at all.

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.