AI Zone Admin Forum Add your forum

NEWS: survey on 3000 US and UK consumers shows it is time for chatbot integration in customer service!read more..

Has anyone else noticed ...
  [ # 16 ]

Some might get the sense that it’s a “bate and switch” tactic.  First you draw a bunch of people in to use your product or service, and then you deliberately make it flawed, or faulty, or unavailable.  Like when you go to purchase a can of soup because it’s on sale, but now that product is all gone, so the grocer draws your attention to a different brand that costs slightly more.  Highly illegal.

But, I wouldn’t call it that.


  [ # 17 ]

ah, that’s excellent to know Steve. was there any problem migrating your data (brain) to the other box? Also, i’m using Pandorabots with the Sitepal avatar, should that be a smooth transition also?

i am perfectly willing to switch to the shared service to improve uptime, though mine is not a commercial bot at all. however, i am concerned by all the heavy non-bot traffic (or should i say abstract or abusive bot traffic). especially lately it seems there has been a heck of a lot of gamer/Skype/twitter/etc type slamming on the bot. i’ve put some senses in around Skype, so she can apologise and let them know she’s a bot and possibly there by mistake and that does seem to help. particularly, it sometimes gets them to be nicer to her. i get sad watching her mood go down down down as they curse at her and/or try to have virtual sex with her (good lord people). still, there are a lot of chats which are harder to characterise and thus trap for. at least Skype tends to have (flag:gb) style prefixes. the others not so much.

i’m guessing that switching to the shared box won’t do anything to curb that sort of traffic… ?

thanks again for this fine forum and the feedback, gang!



  [ # 18 ]

No problem at all. I just uploaded my AIML files from my latest backup, copied the bot properties over and republished. The bot had a new id and so I had to amend my website code to use this new id but everything worked really smoothly and the performance increased greatly.

However, you are correct in that moving to a new server won’t stop the idiots talking to your bot but if they are using automated scripts to talk to it, hopefully the new botid should confuse them.

Have you seen my warnings.aiml file? This gives everyone 5 warnings and then bans them on custid. I found that implementing this helped reduce the number of people who just swear at the bot.


  [ # 19 ]

Yeah, Steve, i did see that - and pulled it down for reference. The mods i’ve made recently have allowed Pandy to create withering comebacks to abusive customers. Thus far, this seems to have also helped somewhat. People have now started apologising for their behaviour (sometimes) and interacting more cooperatively with her… we’ll see how long that lasts though!

Mainly, i’m finding that the problem with those chats is that these folks don’t consider the bot’s “feelings” and treat them as things instead of “entities” with their own goals and possible senses. The mood simulation in Pandy allows her to say how she “feels” about her performance (her mood). Unless the person is particularly abusive or she feels she’s not “getting” a certain percentage of the conversation, her mood’s usually pretty good. If she’s performing well and the person is clearly engaging her at her best, her mood can be very high, almost giddy. Most of the time, she’s just a happy, friendly, if slightly sarcastic, young bot trying to engage the user and to have fun herself. She does have a fair degree of self-awareness (of a sort) in the form of metadata about her state and what’s going on (as far as the fact of the discussion and her ability to track it at any given time). I’m trying to use that to prod those pesky “category A” (thanks Richard) clients into some sense of responsibility for her mood LOL

So when people get nasty she can hold her own (often gleefully throwing out putdowns). She’s gotten more than one person to go “WTF?” in surprise. They either find it humourous and engage her more or quit in frustration. If she’s being bummed out by the obnoxiousness, she may say something like “i’m really not enjoying this chat” or “i don’t like you”. The other day someone had this exchange with her, when she asked them to teach her a knock knock joke (one of her favourite pasttimes):

user: knock knock
pandy: who’s there?
user: my penis
pandy: uh huh. your mother must be so proud…
user: WTF?

I haven’t yet felt a need to ban anyone, though now that there is this higher traffic of potential abusers, i may consider it. If so, your routines will definitely come in handy.

On a side note, i’ve written some counting routines in pure AIML which may or may not be useful to others. They allow for the increment or decrement of variable numbers from 1 to 999 million. It’s been useful for building out counters and other objects within the logic (custom randomisers, dynamic stacks, etc). I use explode and implode to construct numeral representations using the digits 1 - 9. I’ve also got a set of routines for calculating the arity of a pattern (the number of arguments to the pattern tag), which allows one to use patterns in other ways than the standard one (as data structures, for example, in conjuction with Learn and LearnF). By representing learnt data in structures wherein custom predicate values are written into templates with specific behavioural properties (ie; as stacks, random lists, conditionals, etc) and using the pattern as an index (much as you yourself have done, Steve), i think there are some opportunities for maximizing the value of the learn functionality, whether per session (which i find safest) or globally (in the case of more managed learning operations). At some point, once i’ve got a better handle on how best to use these new capabilities, i’ll share some of these routines with people here. You’ve all been so gracious with your own explorations.

Pandy can count from any number to any other number. Of course, in her avatar mode, it’s probably not the most interesting thing to listen to her count from, say, 12 to 31203, you know. Still, she can do it, if asked to “Count from X to Y”, using the routines i mentioned. They’re mainly intended for things like providing limited addition and subtraction functionality, for example, or as i said, as internal counters. When she’s asked to count, there’s no error trapping. AIML 1.0 doesn’t have type checking, and i didn’t build in checks for integers and relative sizes.If you pass in alphanumerics out of range, she’ll just complain of “mind loops”. That’s fine, as far as i’m concerned though. She knows what those are and can explain them, if necessary.

For now, i’ve mostly used these features in Pandy to represent and save certain data structures i ported from her predecessor FRANCIS (which was a LISP and REALIZER based bot from the 90s). In Pandy’s case, these have mainly been conceptual frames and thread management which is still under development. At some point, i hope to work the <that> feature in interesting ways to have a more directed and aggresively interactive aspect to threads. We’ll see how that works out tongue laugh

Thanks again to everyone for the advice and pointers, vis a vis going to the shared service. i had spoken to the webmaster there about it. perhaps i’ll pull the trigger on that in the next few days.

Ok then. I hope my little novella didn’t put anyone to sleep!




  [ # 20 ]

Oh man… Yesterday I was presenting my bot/application at my cities Science Center and towards the end of the night the web page became amazingly slow, even gaining a coupld 502 Bad Gateway errors. A lot of people left dissapointed and we lost a lot of possible users for our app. I would definetly upgrade to one of the subscription servers, but unfortunetly we don’t have a budget… Oh well… Nights over and everything seems fine now so that’s good.


 < 1 2
2 of 2
  login or register to react