[SpookyZalost]

Anyone remember chatbots back in the day?

the early DIY attempts at creating programs which could communicate with people, in some cases an effort to beat the turring test.

ALICE, and it's variants.

and of course the more... questionable one's designed for the more... adult of us.

anyone still using them or see them used more recently?

[Guardian]

Some sites still TRY to use them... terribly. I just find them annoying now. Leave me alone!

[SpookyZalost]

any positive memories with them though?

[Guardian]

any positive memories with them though?

Not really. :

For me anyway... my hatred for them started with Clippy and I never recovered.

[SpookyZalost]

ugh, clippy was an abomination, microsoft's effort to proffit of that crapshoot that was microsoft bob.

[Darth-Apple]

Well, if anyone from SM remembers Romulus...

I set them up in IRC rooms all over the place a long time ago. They were not necessarily true machine learning algorithms by any stretch of the imagination, and had virtually no AI component whatsoever. However, they did involve heuristics to identify "compatible" responses based on the input that it sees in the chatroom. They'd basically sit there and watch the chat all day and create a gigantic database of sentences that they would use in response to other sentences that people used. And once they decided to wake (usually invoked by certain keywords you'd set), they were so good that they'd leave people fooled for hours if you let them. They seemed like actual users in the chat that you could have a conversation with. That is, until you asked them a question. Once they started generating gibberish, they didn't stop. The conversations got interesting fast.

The true test was to set two separate bots, on two separate servers, and to train them with two separate sets of IRC chat rooms. While we always used a lot of diverse training data from multiple chat rooms, the primary training we gave them for this test was primarily a lot of political material. Man oh man, I wish I still had these logs sitting somewhere, but after hours (or days, honestly) of hardcore training from nothing but a ton of political propoganda, most of the responses that these bots generated were naturally political in nature. We put these two bots in the same room and made them talk to each other, and let's just say that while a measurable amount of what generated did come out to be relative gibberish, a lot of it seemingly was a little more concise than we expected. It was probably one of the most hilarious chat moments I had ever seen. A fair amount of nonstop (sometimes intelligent) political ramble ensued, unplagued by the fatigue that would have encapsulated any human mind.

I routinely ran these bots in various IRC chat rooms and they'd occasionally get invoked by keywords and respond to users. Every once in a while they generated some responses that were a little out of line, but for the most part, they blended in pretty well, albeit with the common gibberish they generated here and there. I never got in trouble for doing it, and ran these bots for months without much of an incident. We usually let people know out of courtesy, just in case.

Edit: we do have a couple logs of the bot running in short bursts...

[SpookyZalost]

@ darth: heh, you think romulus would win in a political debate with trump?