A while back, here and in another forum, there was some discussion regarding bot-to-bot conversations.  I was seeing a lot of them and thought it might be an expeditious way to learn how a bot should respond to a variety of input… so, I wanted to learn how it was accomplished.

Rich Wallace replied in the thread that the practice was a TOS violation, and for the obvious reason of creating a runaway condition that would clog-up the free community server.

I’m still seeing a lot of bot-to bot conversations in my chatlogs .  I recognize it’s another Pandorabot when it talks of “the Giant Bugs,” in Starship Troopers, or the input is in the form of a statement rather than a question, like when it begins by stating, “It’s all part of God’s mysterious plans.”  And frequently the chats are so long that they send my bot high on the Most Active/Popular list.

Bot-to-bot conversations are awful… for the most part, unusable.  There might be one or two exchanges where I can see a better way of answering, but for the most part, it’s nonsense, and it rarely produces anything like a conversation.

I’d like to see Pandorabots dig into this problem, for their sake as well as mine.  So far, it seems that some are getting away with what I was told was forbidden.

I get a few of these too but I think from the timestamps on each exchange, that they are copying and pasting the text between two bots rather than having an automated script.

You can’t stop it but I have categories for things like “Your happiness adds to the delight of it all”, “It’s all part of God’s mysterious plans” and other such ALICE responses which no human is likely to say. I <srai> them to a category containing things like, “Why are you copying and pasting my responses?” and “Am I talking to ALICE here?”. They often stop after that.

If Pandorabots are looking for a way to stop automated conversations, how about checking the speed at which the other person replies after the bot answers? If someone can type out a question or reply in less than a second, chances are they are a bot.

Steve Worswick - Jun 1, 2013:

I get a few of these too but I think from the timestamps on each exchange, that they are copying and pasting the text between two bots rather than having an automated script.

I believe you’re correct in some cases.  I also see where three lines of input are strung together producing three separate and unrelated answers.

I’ve also found where, although my bot has stopped talking because of a trigger-word the person has used, the “human” continues to ask questions for a very long time.  That happened today where about halfway through the 364 exchanges, my bot stopped replying because this person used a trigger word that leads me to believe they’re chatting from a forum or game site.  I think that’s another indication that it’s a bot, humans give up after a few exchanges when my bots fail to answer any more.

I recently had a log with 1,170 interactions which, for that bot, is out of the ordinary.  I don’t know what a normal time for an average conversation might be, but the total time for it was 3 min. 25 sec.  Another chatlog contained 973 exchanges, and ran for 9 minutes and 31 seconds.

A log made up of 6,165 interactions took 28 min. 27 seconds.  It was so large I couldn’t the Training Interface wouldn’t, or couldn’t, open it.  And then, there was the grand-daddy of the all, 96,972 exchanges that ran 5 hours 39 min 46 seconds.  Who would sit chatting with the same chatbot all that time?  Again, I couldn’t view the log.

Then I had 464 exchanges with NTU@SL Librarian, a Pandorabot in Second Life. It was also using my replies as input. 

Human: You think I am screen looks dirty here have a lick.
Clicking on a link in the line displays a cat licking your computer screen.

You can’t stop it but I have categories for things like “Your happiness adds to the delight of it all”, “It’s all part of God’s mysterious plans” and other such ALICE responses which no human is likely to say. I <srai> them to a category containing things like, “Why are you copying and pasting my responses?” and “Am I talking to ALICE here?”. They often stop after that.

I do a lot of that, also.  But it’s like trying to put out a fire.  Just as you solve one problem, another breaks out someplace else.

As you know, I once had skynet used in a university research project. I ended up with about two weeks of 9-5bot to bot conversations. They ran as fast as the other bots could respond.
I would bet most of the bot to bot conversations you are having problems with do not use the web interface. Maybe you can put something in your web page to help identify web vs other connections.

I’ve had the same though, Merlin, but I’m not that computer savvy.  And, even if I’d discover something that would confirm that suspicion, what could I do about it?

People often ask my bots for their name.  A while back, I started having the bots respond by saying, “My name is on the screen right in front of you,” to which people sometimes reply, “No it’s not.”  That tells me they’re not at the web page, or they’re using some other form or device to reach the botid.

I’m open to any suggestions.

Can you set the topic or conversation id from the web interface?
The current most popular list on pandora bots has just an input box. It makes me think many of these are not via web pages. Possibly the Call Mom app?

Merlin - Jun 2, 2013:

Can you set the topic or conversation id from the web interface?
The current most popular list on pandora bots has just an input box. It makes me think many of these are not via web pages. Possibly the Call Mom app?

I have no idea, with regard to the web interface.

Yes, most of the bots are only a white text box, but I’m told they may also be a Second Life bot or a phone app.

If you see my bot Nickie on that list, clicking it will take you to the web page, as will Steve’s Bearbot, Chomsky, Lauren, Chabba, Izar, and Nicholas Jonas, and a few others when they appear there.

You might be interested in this app which allows others to talk to your bot without using your HTML

http://dev.bukkit.org/bukkit-mods/vectron-artificial-intelligence/

I guess their use of Marvin the Android shows how much they respect copyright and intellectual property.

I thought about registering, just to leave a comment, but at best it would likely get ignored.

insert 2 cents here >>>

In all honesty its about a 5 minute job to add code to any interface that disallows cross site posting, and probably wouldnt take more than an hour to build an interface that would allow a user to associate a userid with allowed urls. As for the API it wouldnt be that much more complicated to require the use of an api key. If this is a problem, you need to insist that these measures be put in place. One thing that I dont quite understand, when your using AIML, isnt the real work in the creating the triggers? How much real value is there for someone harvesting responses?I mean that even if someone had a list of responses to specific statements\interrogatories wouldnt the still have to have the skill set to write an AIML file that would implement that response? I always thought that was where the real skill was.  (could just be my overall ignorance with the format)

vince

I find that data mining the triggers is the easy part (especially in a headword based system).
Finding correct responses automatically is trickier.

Steve Worswick - Jun 8, 2013:

You might be interested in this app which allows others to talk to your bot without using your HTML

http://dev.bukkit.org/bukkit-mods/vectron-artificial-intelligence/

I guess their use of Marvin the Android shows how much they respect copyright and intellectual property.

Is this another one, or the same guy that was using Marvin the last time?

Gosh, I wonder what Disney would think?

http://thewaltdisneycompany.com/content/disney-antipiracy

Vincent Gilbert - Jun 8, 2013:

One thing that I dont quite understand, when your using AIML, isnt the real work in the creating the triggers? How much real value is there for someone harvesting responses?

I think they do this to have a base of bots who can add value to their product rather than trying to steal any AIML files themselves. If they had a bot application but no bots, it wouldn’t be as popular. And it takes time to write a bot, so they think why not just steal someone else’s hard work.

The only reason I think most of us mind our bots being included in such things is that we often put special tags or keywords in our bots which 3rd party applications don’t always follow. For example, I have a keyword that displays picture on Mitsuku or opens a webpage.

I see from the screenshot they provide that one exchange is “My brain contains <size></size> categories”. These tags shouldn’t be displayed to the enduser and makes the bot look stupid.

Thunder Walk - Jun 8, 2013:

Is this another one, or the same guy that was using Marvin the last time?

Now I remember, it wasn’t Marvin, it was EVE from the movie WALL-E, another Disney character.

https://itunes.apple.com/us/app/chatbot/id449856276?mt=12

And then, let’s not forget this thief.

http://www.hawkee.com/snippet/6306/