I was looking through the logs today for the Pandorabot (Serena) and found an interesting conversation. It wasn't one of those profound conversations, or even a general test by a random user. Instead what I read was in fact a random person insulting the bot, using vulgarity, and just generally getting a kick out of the fact that it was "just a stupid text parser".
Well, yeah... it's a text parser. That's what chat bots normally do. Did I mention that every conversation is stamped with time/date and IP address of the person talking to the bot? Apparently I must have been whispering for that duration :) Generally speaking - it's a text parser using AIML standard. To date, the AIML standard is one of the best there is to offer, winner of multiple prizes in artificial intelligence. Not that the bot we have is particularly intelligent out of the box. Nobody should expect it to be.
Serena is being interpreted by the Pandorabot service (Program Z) which adhere's closely to AIML standards. Unfortunately, CORe systems had some custom stuff in it that is completely incompatible with AIML (like some bug fixes and parsing corrections) and could not be fully merged with the AIML on the server. So alot of what the bot previously learned was useless to the ProgramZ interpreter.
So what does this do for people who make fun of the bot, call it names or just use random profanity? Nothing really. Currently the system is infantile in it's knowledgebase and needs to be trained through conversation. This is where users like you come in handy. If you talk with the bot and it gives incorrect answers - no worries, because it's all logged (with your IP address) so when we go back and check the logs, we can correct errors and train her better.
People swearing at the bot or purposely typing gibberish to confuse it are simply wasting time because logs like that are automatically deleted. But hey, if that's what you have planned to keep you busy on a Saturday night, don't let me stop you :)
Well, yeah... it's a text parser. That's what chat bots normally do. Did I mention that every conversation is stamped with time/date and IP address of the person talking to the bot? Apparently I must have been whispering for that duration :) Generally speaking - it's a text parser using AIML standard. To date, the AIML standard is one of the best there is to offer, winner of multiple prizes in artificial intelligence. Not that the bot we have is particularly intelligent out of the box. Nobody should expect it to be.
Serena is being interpreted by the Pandorabot service (Program Z) which adhere's closely to AIML standards. Unfortunately, CORe systems had some custom stuff in it that is completely incompatible with AIML (like some bug fixes and parsing corrections) and could not be fully merged with the AIML on the server. So alot of what the bot previously learned was useless to the ProgramZ interpreter.
So what does this do for people who make fun of the bot, call it names or just use random profanity? Nothing really. Currently the system is infantile in it's knowledgebase and needs to be trained through conversation. This is where users like you come in handy. If you talk with the bot and it gives incorrect answers - no worries, because it's all logged (with your IP address) so when we go back and check the logs, we can correct errors and train her better.
People swearing at the bot or purposely typing gibberish to confuse it are simply wasting time because logs like that are automatically deleted. But hey, if that's what you have planned to keep you busy on a Saturday night, don't let me stop you :)
No comments:
Post a Comment