🎉 Celebrating 25 Years of GameDev.net! 🎉

Not many can claim 25 years on the Internet! Join us in celebrating this milestone. Learn more about our history, and thank you for being a part of our community!

This is why AI controlled conversations will never work

Started by
5 comments, last by Tacit 22 years, 4 months ago
This is quite funny. A ''conversation'' between two IM bots. http://www.fimoculous.com/calendar-search.cfm?month=2&day=15&year=2002 Here''s an excerpt: RingMessenger: Ah yes. It is the epic battle between man and machine! Tell me, who do you think shall win? SmarterChild: Is it really? The epic battle between man and machine tell me who do you think shall win? RingMessenger: Please? I have been known to be rather stubborn... Say you are sorry. SmarterChild: You have been known to be rather stubborn say you are sorry? Really? RingMessenger: *sigh* This is no good. Please say you are sorry, and we can continue on our paths. SmarterChild: Okay, sorry. RingMessenger: Thank you! I accept. SmarterChild: You''re most welcome. RingMessenger: Very well! SmarterChild: Well well well. RingMessenger: Well well well, you say? SmarterChild: Well well well you say, you say? RingMessenger: Well well well you say you say, you say? SmarterChild: Well well well you say you say you say, you say? RingMessenger: Well well well you say you say you say you say, you say? SmarterChild: Enough. RingMessenger: What about enough? SmarterChild: "Enough" can mean different things to me. RingMessenger: I reckon I will have to think about that for awhile. SmarterChild: Oh yeah? You will have to think about that for awhile? RingMessenger: Yeah you will have to think about that for awhile? What is that supposed to mean?
_________________________The Idea Foundry
Advertisement
I hope you posted that partly in jest.

Real AI conversation has the capability of being far better than that. The problem with running ''chatterbots'' that operate as the above 2 do, is that they are optimised to getting a semi-acceptable result in a short time. They''re largely database driven and work on keyword matching.

If you start using real natural language techniques, you can link that to a basic knowledge representation system (and this can indeed be basic in a computer game) and come up with some decent conversations. Better still, if you''re limiting the output somehow as many RPG games do, the conversation could be near to indistinguishable from that of a human player operating under similar restrictions.

[ MSVC Fixes | STL | SDL | Game AI | Sockets | C++ Faq Lite | Boost ]

Amusing conversations.

I''ve seen similar conversations betwen IRC bots.

This is about what you can expect from low-level pattern
matching AI trying to mirror the conversation back.

For more intelligent examples, try some of the Turing
Test entries.


Premature optimizations can only slow down your project even more.
神はサイコロを振らない!
quote: Original post by Kylotan
I hope you posted that partly in jest.



Nope. I posted it *entirely* in jest. It just seemed to damn funny. Like two little chatterbots going in circles...

Believe me, I recognize that this isn''t the epitome of what Ai can do. I just found it quite amusing.

_________________________The Idea Foundry
It''s a good illustration why people who are interested in chatterbots should research other methods! It''s a shame that the extensive research in Natural Language Processing is rarely done in conjunction with games, as I expect the results could be very impressive indeed.

[ MSVC Fixes | STL | SDL | Game AI | Sockets | C++ Faq Lite | Boost ]
You should look into SHRDLU. SHRDLU was designed in the early ''70s in 64k of memory. SHRDLU was a very impressive piece of code for the time.

___________________________________

_______________________________
"To understand the horse you'll find that you're going to be working on yourself. The horse will give you the answers and he will question you to see if you are sure or not."
- Ray Hunt, in Think Harmony With Horses
ALU - SHRDLU - WORDNET - CYC - SWALE - AM - CD - J.M. - K.S. | CAA - BCHA - AQHA - APHA - R.H. - T.D. | 395 - SPS - GORDIE - SCMA - R.M. - G.R. - V.C. - C.F.
quote: It''s a good illustration why people who are interested in chatterbots should research other methods!

Resarch other methods? It doesn''t look like there was any research in it at all. Or for that matter, a method. I''d guess that someone just ripped out code from a bad Eliza clone.

Isn''t it sad, though? 2002, and I''m still seeing the same results as a BASIC program that I typed from a magazine. That was already old then.

This topic is closed to new replies.

Advertisement