Re: Meta-discussion vs. details

[Jonathan H. <vh...@ru...>]

On Wed 2004-08-04 (19:09), Paul L. Allen wrote:
> Or to put it another way, can anyone conceive of a legitimate type of
> inter-bot message that would not also be legitimate (in at least some
> situations) when given to the bot by a human (privileged to do so) in
> public or in private?  So we don't need a special inter-bot message
> format. 

It's rather the other way around.  There is no type of inter-bot
messages that aren't legitimate when coming from a (sufficiently
authorised) human, but there *are* legitimate human messages that
shouldn't be accepted from other bots.  The most obvious example is
avoiding bot-loops, where each bot tries to reply to the reply it just
got from the other one.

Because bots are trainable to behave in certain ways, having one bot
trust another bot is slightly dangerous.  If it is possible for me to
train a bot to say something to another person, then I can train him to
say something to another bot.  If a bot trusts another bot more than he
trusts me, I can use this method to get around the security.
Thus, I think, bots should trust other bots as much as J. Random User at
maximum, and possibly a lot less.