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

BGonline.org Forums

XG is not perfect/No human will make this play

Posted By: Timothy Chow
Date: Wednesday, 30 September 2015, at 3:39 p.m.

In Response To: XG is not perfect/No human will make this play (RobertFontaine)

RobertFontaine wrote:

1. What do we meat puppets understand here that our fancy electronic can opener (XG) does not?

The simplest example is how to roll a prime all the way around the board.

2. Why do you think XG is incapable of providing a World Class answer to this position given the choice of a wide search space and a few extra ply?

It helps to look at some concrete data rather than theorize purely in the abstract. Look at this game for example, where I played against XG 3-ply. The starting point is the obvious fact that XG 3-ply and XG Roller+ can't both be right when they contradict each other. Here, they not only contradict each other by small amounts once in a while, but by large amounts often. This should be enough to convince you that XG 3-ply can't be trusted.

So can XGR+ be trusted? XGR+ is not an oracle from heaven. It is based on a truncated rollout of play at even lower-ply evaluations than 3-ply. If we don't trust 3-ply then why should we trust lower ply? If the computer plays nonsensically then why should playing nonsensically a lot of times suddenly make its verdicts reliable? Our trust in rollouts is predicated on the assumption that the computer plays reasonably well to start with; then rollouts can improve the accuracy. If the computer doesn't play reasonably to start with then all bets are off.

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.