Is scripted chat possible?

Hi all!

I am building a machine that pays the user for chatting with it. So far it work nicely - it detects and keeps a list of all agents within 5 m in front of the machine and can therefore detect if the message was sent by an agent or not.

Is there a means to make an AGENT say something by using an LSL script? I don't know of such possibilities, but I fear if it exists, people will start using such scripts to suck my account dry...

Please comment on this

Best regards
Gregor

Not in LSL, but it's perfectly possible that a bot could be coded to speak gibberish for hours on end.

you mean by modifying the source code of the client?

I'm not sure you really need to detect and keep a list of agents either.

In the listen event...

listen (integer channel, string name, key id, string message)
{
     if (llGetOwnerKey(id) == id)
}

This makes sure that it will only listen to id's that own themselves, which only holds true for agents.

So long as the listening object and the agent are in the same sim this will work without having to use a sensor.

Yes. When a user clicks on a button in a menu generated by llDialog(), their agent will say whatever text was assigned to that button on the channel specified in the function call.


The method described above can't run unattended. It requires the user to click on a button for each message, so someone couldn't just "camp" overnight in front of your gadget and babble continuously.

SL clients that chat already exist.

By this I mean programs written using libsecondlife to connect with the LL servers, that produce an avatar which can, if written properly, appear pretty much or exactly like a regular human operated avatar. These "libsl chatbots" would typically make use of computer conversation libraries in addition to the libsecondlife library. Examples of conversation libraries would be AIML (Artificial Intelligence Markup Language, if my acronym memory is working right) or libraries that do the classic Eliza chat, or they might use RACTER or HAL libraries.

There's an article at http://nwn.blogs.com/nwn/2007/01/social_animal.html on this topic.

These avatar chatbots can't be distinguished with the" if (llGetOwnerKey(id) == id)" trick. They aren't LSL scripted objects, they are avatars, same as you and me, just operated by a program instead of a waterbag stuck to some calcium rods.

There's a good chance that if someone is paying enough money for some behavior and the behavior can be done by a program that's not too hard to write, and the libsecondlife programmers learn of it, there will be a libsecondlife based SL client written to take advantage of the income stream.

It sounds like the sensor, in addition to verifying chat messages come from Agents, is also used to filter messages for those intended for it (those from avs within 5m in front of it), rather than just any chat it receives, which could be from as far away as 100m, in any direction. Though there are other less passive methods, like requiring the chatting user to first click on the object, or start their message with some common identifier, e.g. "Computer, blah blah blah..."

Oh, I'm stealing that one.

You are completely right. The common identifier is used additionally, but as it makes no sense activating the machine from 100 m distance, the sensor checks for the distance, too, an also deactivates chatting if the user goes further than 5 m away.



That's bad..

I could use common words displayed on a screen every two minutes or so, and only pay out every 5 minutes. No payout if the captchas are not identified correctly. I have no idea yet how to make the user validate the captcha, since chatting would defeat the purpose (if multiple agents use the machine, the chatbot could just listen on the channel to get it).

Thanks to everybody so far!

Hmm, make it so they have to touch the machine every few minutes when prompted, preferably touching a colour coded button ?